DEVELOPMENT OF FUNCTIONAL STUDIES AND METHODS TO BETTER UNDERSTAND VISUAL FUNCTION
DISSERTATION
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the
Graduate School of The Ohio State University
By
Nasser Hussam Kashou, M.S.
*****
The Ohio State University
2008
Dissertation Committee: Approved by
Dr. Cynthia J. Roberts, Co-Adviser Dr. Ronald X. Xu, Co-Adviser Co-Adviser Dr. Lawrence E. Leguire Co-Adviser Graduate Program in Biomedical Engineering c Copyright by
Nasser Hussam Kashou
2008 ABSTRACT
In the study of visual function an understanding of the visual pathways is essential.
Once this is achieved then quantitative measurements can be made in order to assess the quality of vision. However, this development can at times be problematic and may lead to visual disorders. Some of these visual disorders are directly related to the development but others may not. We are concerned with mainly one of these visual disorders, infantile nystagmus syndrome (INS). Common ways INS is assessed is through visual evoked potentials (VEP), or electroretinigrams (ERG). The current work is a comprehensive multidisciplinary attempt to develop new tools and methods for assessing these visual functions in order to both complement as well as introduce new clinical tools that will help in finding efficient treatments by identifying the activation patterns in the brain. This is divided into three stages: functional magnetic resonance imaging (FMRI) of oculomotor movements, development of a near infrared spectroscopy system (NIRS) for visual cortex monitoring, and finally an MRI post processing scheme to enhance the cortical imaging. These three stages are an attempt to develop tools in order to aid in visual function studies.
ii Dedicated to my mother
iii ACKNOWLEDGMENTS
I would like to praise Allah, may He be glorified and exalted, for blessing me with an education and guiding me throughout my life and I would like to thank my mother, Ibtisam El-Khatib, for her continuous support, motivation and guidance. I would also like to thank my advisors and mentors Dr. Lawrence E. Leguire, Dr.
Cynthia J. Roberts, and Dr. Ronald X. Xu for allowing me this opportunity and helping me get through it successfully.
iv VITA
November 16, 1978 ...... Born - Beiruit, Lebanon
March 16, 2001 ...... B.S. Electrical Computer Engineering, The Ohio State University June 15, 2004 ...... M.S. Electrical Engineering, The Ohio State University August 26, 2007 ...... M.S. Biomedical Engineering, The Ohio State University January 19, 2007 - Present ...... PhD Candidate, The Ohio State University.
FIELDS OF STUDY
Major Field: Biomedical Engineering, studies in Vision Science, Functional MRI and Near Infrared Spectroscopy. Electrical Engineering, studies in Computer Engineering, VLSI Design and Image Processing.
v TABLE OF CONTENTS
Page
Abstract ...... ii
Dedication ...... iii
Acknowledgments ...... iv
Vita...... v
List of Tables ...... xi
List of Figures ...... xv
Chapters:
1. INTRODUCTION ...... 1
1.1 Statement of Problem ...... 1 1.2 Significance of Research ...... 1 1.3 Organization of Dissertation ...... 1
2. BACKGROUND OF THE ASCENDING VISUAL PATHWAY ...... 3
2.1 Introduction ...... 3 2.2 Visual Pathway ...... 5 2.2.1 Ganglion Cells ...... 5 2.2.2 Lateral Geniculate Nucleus (LGN) ...... 9 2.2.3 Primary Visual Cortex (V1) ...... 10 2.3 Development of the Ascending Pathway ...... 14 2.3.1 Disorders of the Ascending Pathway ...... 17 2.4 Conclusion ...... 21
vi 3. OVERVIEW OF BRAIN FUNCTIONAL STUDIES ...... 23
3.1 Introduction ...... 23 3.2 Functional Magnetic Resonance Imaging(FMRI) ...... 24 3.2.1 FMRI Visual Cortex Studies ...... 24 3.3 Near Infrared Spectroscopy (NIR) ...... 31 3.3.1 FNIR Visual Cortex Studies ...... 32 3.4 Discussion ...... 35
4. FMRI on Look vs Stare Optokinetic Nystagmus (OKN) ...... 37
4.1 Abstract ...... 37 4.2 Introduction ...... 37 4.3 Materials and Methods ...... 39 4.3.1 Subjects ...... 39 4.3.2 Scanning sequences ...... 39 4.3.3 Optokinetic Nystagmus Stimulation Protocol ...... 40 4.3.4 Data Analysis ...... 42 4.4 Results ...... 44 4.4.1 Activation effects in individual subjects ...... 44 4.4.2 Group activation effects ...... 44 4.5 Discussion ...... 49 4.6 Conclusion ...... 54
5. PARADIGM EFFECTS ON LOOK VS STARE OPTOKINETIC NYS- TAGMUS (OKN): AN FMRI STUDY ...... 55
5.1 Abstract ...... 55 5.2 Introduction ...... 56 5.3 Materials and methods ...... 57 5.3.1 Subjects ...... 57 5.3.2 Scanning sequences ...... 58 5.3.3 Optokinetic Stimulation Protocol ...... 58 5.3.4 Data Analysis ...... 60 5.4 Results ...... 61 5.5 Discussion ...... 62 5.6 Conclusion ...... 65
6. FUNCTIONAL MAGNETIC RESONANCE IMAGING (FMRI) OF VI- SUALLY GUIDED SACCADES AND SMOOTH PURSUIT AT 3T . . . 78
6.1 Abstract ...... 78
vii 6.2 Introduction ...... 79 6.3 Materials and Methods ...... 79 6.3.1 Subjects ...... 79 6.3.2 Scanning sequences ...... 80 6.3.3 Saccade and Pursuit Stimulation Protocol ...... 81 6.3.4 Data Analysis ...... 81 6.4 Results ...... 83 6.5 Discussion ...... 84 6.6 Conclusion ...... 89
7. COMPARISON OF AXIAL, SAGITAL, AND CORONAL IMAGING FOR SIMPLE FINGER TAPPING EXPERIMENT: AN FMRI CASE STUDY 90
7.1 Abstract ...... 90 7.2 Introduction ...... 90 7.3 Materials and Methods ...... 91 7.3.1 Subjects ...... 91 7.3.2 Scanning sequences ...... 91 7.3.3 Stimulation Protocol ...... 92 7.3.4 Data Analysis ...... 92 7.4 Results ...... 93 7.5 Discussion ...... 94 7.6 Conclusion ...... 95
8. TWO AND THREE DIMENSIONAL IMAGE COMBINATION FOR MRI ACQUIRED USING DIFFERENT SCANNING PLANES ...... 99
8.1 Abstract ...... 99 8.2 Introduction ...... 99 8.3 Background ...... 100 8.3.1 Significance ...... 100 8.3.2 Methods ...... 101 8.3.3 2 dimensional phantom (2-D) ...... 102 8.3.4 3 dimensional(3-D) phantom ...... 103 8.4 Results ...... 110 8.5 Discussion ...... 117 8.6 Conclusion ...... 118
9. USING FMRI AND FNIRS FOR LOCALIZATION AND MONITORING OF VISUAL CORTEX ACTIVITIES ...... 119
9.1 Abstract ...... 119
viii 9.2 Introduction ...... 119 9.3 Technology Review ...... 121 9.3.1 Near Infrared Light (NIR) ...... 121 9.3.2 NIR Systems ...... 122 9.4 Experimental Design ...... 123 9.4.1 FMRI Scanning Parameters ...... 123 9.4.2 FMRI Data Analysis ...... 124 9.4.3 FNIR Instrument and Sensor ...... 125 9.4.4 Task ...... 125 9.4.5 FNIR Data Analysis ...... 126 9.5 Results ...... 126 9.6 Discussion ...... 127 9.7 Conclusion ...... 130
10. DESIGN AND DEVELOPMENT OF A FUNCTIONAL TOOL USING NIRS FOR NON-INVASIVE DETECTION OF VISUAL CORTEX AC- TIVITIES ...... 132
10.1 Abstract ...... 132 10.2 Background ...... 133 10.3 Methods and Development ...... 133 10.3.1 System ...... 135 10.3.2 Sensor Head ...... 137 10.3.3 Software ...... 140 10.4 Pre Trial test ...... 142 10.5 Conclusion ...... 146
11. ALGORITHM SIMULATION ...... 147
11.1 Introduction ...... 147 11.2 The Heterogeneous Diffusion Equation ...... 147 11.3 Born Approximation ...... 148 11.4 Rytov Approximation ...... 150 11.5 Analytical Inverse Algorithm ...... 152 11.6 Model and Simulation ...... 154 11.7 Discussion ...... 169
12. CONCLUSION AND FUTURE WORK ...... 170
Appendices:
ix A. Current Studies Using FMRI and FNIRS ...... 171
B. StO2 Derivation ...... 174
B.1 Pulse Oximtery (PO) ...... 174 B.2 Near Infrared Spectroscopy (NIRS) ...... 177
Bibliography ...... 180
x LIST OF TABLES
Table Page
4.1 Visual acuity and demographics of subjects that participated in the study...... 39
4.2 Talairach coordinates of mean (meanstdev where more than one coor- dinate is available per correlate) location of the look- and stare- OKN related activity at the Culmen. Z-scores were thresholded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01...... 45
4.3 Talairach coordinates of mean (meanstdev where more than one coor- dinate is available per correlate) location of the look- and stare- OKN related activity at the Middle Occiptal Gyrus. Z-scores were thresh- olded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01...... 45
4.4 Talairach coordinates of mean (meanstdev where more than one coor- dinate is available per correlate) location of the look- and stare- OKN related activity at the Precentral Gyrus. Z-scores were thresholded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01...... 46
4.5 Talairach coordinates of mean (µ ± σ where more than one coordi- nate is available per correlate) location of the look- and stare- OKN related activity at the Cingulate Gyrus. Z-scores were thresholded us- ing clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01...... 46
xi 4.6 Talairach coordinates (meanstdev where more than one coordinate is available per correlate. Anatomical correlates across entire brain for mean look voluntary OKN. Z-scores were thresholded using clusters determined by Z − score > 3.0 and a (corrected) cluster significance threshold of P − value = 0.01. L/R (left or right hemispheres) . . . . 50
4.7 Talairach coordinates (meanstdev where more than one coordinate is available per correlate. Anatomical correlates across entire brain for look voluntary - stare involuntary OKN using 2 sample paired t-test. Z-scores were thresholded using clusters determined by Z −score > 3.0 and a (corrected) cluster significance threshold of P − value = 0.01. L/R (left or right hemispheres)...... 51
5.1 The visual, stereo acuity and demographics of subjects that partici- pated in the study...... 70
5.2 The contrast sensitivity functions of subjects that participated in the study...... 71
5.3 The group mean anatomical sites exclusively seen for either look or stare as well as common to both with P − value < 0.05...... 72
5.4 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the occipital lobe for look voluntary - stare invol- untary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation...... 73
5.5 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the frontal lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activa- tion...... 74
xii 5.6 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the parietal lobe for look voluntary - stare involun- tary OKN using 2 sample paired t-test with P −value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activa- tion...... 75
5.7 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the temporal lobe for look voluntary - stare invol- untary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation...... 76
5.8 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the limbic lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activa- tion...... 76
5.9 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the posterior lobe for look voluntary - stare invol- untary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation...... 77
5.10 The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the anterior lobe for look voluntary - stare involun- tary OKN using 2 sample paired t-test with P −value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activa- tion...... 77
6.1 The activation sites from the group averages of saccade and pursuit eye movements for P < 0.05. Overall there were seventeen areas associated solely with saccades, one solely for pursuit and twelve areas of overlap 87
7.1 The anatomical correlates from the three plane scans...... 96
xiii 8.1 The iteration scheme for 2-D interpolation of combined images for sev- eral different pixel dimensions. The number of iterations and weighting factors are dependent on the difference between the small and large pixel dimensions, i.e. spacing. Note this algorithm is also valid for floating point resolution...... 103
8.2 MSE ...... 111
8.3 MSE for 3-D phantoms of sizes 64, 96 and 128. The MSE between the ideal phantom and the combined subphantoms without interpolation (top row) as well as after interpolation (bottom row). We expect a decrease from row 1 to 2 as well as a decrease from 64 to 128 in the second row...... 111
10.1 The minimum source detector distances allowed for sensor design 1. . 140
10.2 The source detector distances for sensor design 2...... 142
A.1 Pathologies investigated in FMRI and FNIR...... 171
A.2 NIRS system setup for relevant NIRS studies...... 172
A.3 Specifications of FMRI studies performed on normal eye movements. All the groups used a θ = 90o for the flip angle except Miller et al 2005 which used θ = 20o...... 173
xiv LIST OF FIGURES
Figure Page
2.1 The visual pathway [1]...... 6
2.2 The layers of the retina and component cells. Note that light has to pass from the front of the eye through the vitreous body and all the retinal layers to finally reach the receptor cells, rods, and cones [1]. . 7
2.3 Retinal ganglion cell projections to the lateral geniculate nucleus (LGN) of the thalamus. Note that layers 1,4, and 6 of the LGN receive visual information from the contralateral retina, whereas layers 2,3, and 5 receive visual information from the ipsilateral retina [1]...... 11
2.4 The visual field representation in the retina and primary cortex [1]. . 12
2.5 The block diagram of ganglion cell mapping from retina through LGN, V1, and other cortical areas...... 13
4.1 Experimental task design. The contrast for all gratings was 35%. There was a total of two scans (one for look OKN and one for stare OKN) for each subject and each scan consisted of three ON-OFF cycles. . . 41
4.2 Activation of look (blue) and stare (red) overlayed for each subject with cingulate gyrus marked by green ellipse. In general more brain activation was found with look OKN at the cingulate gyrus with all subjects being activated; however, subjects 1,2,6 did show activation for stare at these slices. Although not evident by slice selection, subject 5, did show activation for stare OKN. Z-scores (Gaussianised T/F) statistic images were thresholded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01...... 47
xv 4.3 Precuneus, cingulate and middle frontal gyrus activation for look (red) OKN after a group mean of significant areas activated across all six subjects with look using clusters determined by Z − score > 3.1 and a (corrected) cluster significance threshold of P − value = 0.01. . . . . 48
4.4 Culmen and parahippocampal gyrus activation for look (red) OKN after a 2 sample t-test comparison of significant areas activated with look using clusters determined by Z − score > 3.1 and a (corrected) cluster significance threshold of P − value = 0.01...... 48
5.1 The Experimental task design. The counter phase gratings was intro- duced. The contrast for all gratings was 35%. The temporal frequency was 7Hz. The counter phase frequency and moving grating frequency were the same...... 66
5.2 The visual paradigm and response from one voxel in the brain of one subject showing high correlation between block design and physiolog- ical response. All the subjects showed a similar correlation pattern. P − value < 0.05, z cluster threshold = 5.1...... 67
5.3 The activation of look voluntary (red) and stare involuntary (blue) overlayed from one subject. This trend was seen across the other sub- jects as well. We see more cerebellar and higher brain activation with look OKN. P − value < 0.05, z cluster threshold = 5.1...... 68
5.4 The average for look voluntary OKN (blue) and stare involuntary OKN (red). Overall more brain activation is observed for look versus stare OKN with many sites of overlap. Occipital lobe activation from look include: lingual gyrus, cuneus, middle occipital gyrus, inferior tempo- ral gyrus, fusiform gyrus. Occipital lobe activation from stare include: lingual gyrus, cuneus, middle occipital gyrus, inferior occipital gyrus, fusiform gyrus. p − value < 0.05, z cluster threshold = 4.0...... 69
5.5 The 2 sample paired t-test for look voluntary - stare involuntary OKN (blue) and stare involuntary - look voluntary OKN (red). More brain sites are observed solely for look OKN but not for stare OKN. P − value < 0.05, z cluster threshold = 4.0...... 69
xvi 6.1 The experimental task design for 1 cycle. The OFF condition consisted of a white fixation dot for 30sec. The ON condition consisted of the white dot moving back and forth at 0.5Hz for 30sec. For pursuit the motion was smooth and for saccade the motion was ballistic. The cycle repeated three times for pursuit and saccade separately...... 82
6.2 The chart illustrates the number of subjects that showed activation for saccades and pursuits and is split by regions of interest for P < 0.05. On the y-axis are the number of subjects that showed activation for that region. The red indicating saccades and the blue indicating pursuit. 85
6.3 The activation differences between saccade and pursuit for one subject. This trend was typical for all seven subjects with P < 0.05...... 86
6.4 The center slice from a three plane view of the group mean activation across the seven subjects for both saccade and pursuit with P < 0.05 and a minimum Z threshold of 2 and maximum of 11.5...... 86
6.5 The axial view of several lower brain slices of the group mean activation across the seven subjects for both saccade and pursuit with P < 0.05 and a minimum Z threshold of 4 and maximum of 11.5...... 88
7.1 The functional activation maps from the axial, coronal and sagital acquired EPI volumes...... 97
7.2 The brainstem and cerebellar activation for all three planes from slice number 29 from the anatomical image of the brain...... 98
7.3 The brainstem and cerebellar activation for all three planes from slice number 22 from the anatomical image of the brain...... 98
8.1 Project pipeline...... 100
8.2 The Shepp-Logan phantom and subsampled test phantoms in x and y directions...... 101
8.3 The 3-D subsampling in three planes and the combination of original voxels spacing from each plane into a new volume...... 102
xvii 8.4 The step by step iterative reconstruction process for 1x4 and 4x1 pixel spacings for subsampled test phantoms resulting in a 1x1 pixel dimen- sion and 64x64 matrix...... 104
8.5 The three volumes orientated in different planes are the phantoms sim- ulating isotropic in plane resolution for coronal, axial and sagital di- rections...... 105
8.6 This is an illustration of the first steps of the original information method on a 128x128x128 phantom. Each subphantom is upsampled to a common grid (top). Then each is subsampled based on original pixel spacing to extract original information from each volume (middle).106
8.7 The extension of the original information algorithm to 3-D on a 64x64x64 phantom. The last steps remap the original data in x, y, and z planes (bottom row)...... 107
8.8 The ideal phantom, combined subphantoms after remapping and final reconstructed volume (left to right) for 64, 96, and 128 volume sizes (top to bottom)...... 108
8.9 The iterative process of preserving original information in the last two steps of algorithm for 64, 96 and 128 (top to bottom) for axial, coronal and sagital (left to right) center slices...... 109
8.10 The SE for the reconstructed volume. Slice position of 30x31x33, 31x31x31 and 32x32x32 (left to right)...... 112
8.11 The SE of 64, 96 and 128 size volumes between ideal and final recon- structed phantoms...... 112
8.12 The SE of axial, coronal, and sagital (left to right) for 64, 96 and 128 (top to bottom). Slice coordinates chosen to highlight worst case for 64 (27,27,27), 96 (50,50,50) and 128 (63,63,63)...... 113
8.13 The original data (2-D anatomical images with dimensions of 128x512 and 512x128) remapped to a common grid of 512x512 (top) and then combined (middle). The result of different interpolation points used (bottom)...... 114
xviii 8.14 3-D anatomical volumes in the axial (128x128x32), coronal (128x32x128) and sagital (32x128x128) planes (top). Each was mapped to 128x128x128 grid (second from top) then subsampled (second from bottom) preserv- ing original data and combined (bottom)...... 115
8.15 The top row is the first step of the algorithm and the bottom are the last two. Common interpolation methods stop at the middle row whilst this algorithm combines both interpolation and original information to reconstruct the volume and thus stopping at the bottom row. Note top row is not the same slice positions as the bottom two...... 116
9.1 The actual sensor head design. This straps onto the head at the region determined from the FMRI trial. We used a symmetric sensor head design with one detector and eight sources. Sources 1-4 were 692nm and 5-8 were 834nm...... 126
9.2 The relative change in concentration of oxyhemoglobin (HbO) and deoxyhemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 4 and 8. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases. . . 127
9.3 The relative change in concentration of oxyhemoglobin (HbO) and deoxyhemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 1 and 5. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases. . . 128
9.4 The relative change in concentration of oxyhemoglobin (HbO) and deoxyhemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 2 and 6. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases. . . 128
9.5 The relative change in concentration of oxyhemoglobin (HbO) and deoxyhemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 3 and 7. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases. . . 129
10.1 An overview for the motivation of the development and validation of NIRS alongside FMRI...... 134
10.2 The ISS frequency domain system originally used...... 135
xix 10.3 The RS232C QualSys-2044 continuous wave prototype system with a baud rate of 19200, sampling frequency of 14Hz, and power of 30mW. 136
10.4 The first steps taken to modify the RS-232 CW system at the detector interface in order to increase coupling efficiency. The internal compo- nents and the final system are illustrated...... 136
10.5 The modification of the source coupling interface in order to reduce power loss...... 137
10.6 The schematic for the first sensor head prototype included adjustable source and detector slots. Note: not to scale...... 138
10.7 Sensor head design 1 fabricated and tested...... 138
10.8 The schematic for the second sensor head prototype no longer included adjustable source and detector slots. Note: not to scale...... 139
10.9 Sensor head design 2 fabricated and tested...... 139
10.10The schematic for the third sensor head prototype no longer included adjustable source and detector slots and the source diameters were increased to 1mm. Note: not to scale...... 141
10.11The different configurations for sensor head design number 3...... 141
10.12The Labview code hierarchy...... 143
10.13The Labview graphical user interface...... 144
10.14The test setup for functionality evaluation of system, sensor head and software...... 144
10.15The resulting signal from complete setup from one source detector combination. The other combinations showed similar trends...... 145
10.16The experimental setup consists of the modified CW system, sensor head, labview GUI and the visual stimulus...... 145
11.1 The basic pulse oximetry optical path with one source and one detector.152
xx 11.2 A simple near infrared oximetry optical path with one source and one detector...... 153
11.3 The phantom model (10x10x10 cm) used for simulation with an em-
bedded slab (7x7x0.3 cm). The absorption coefficient (µa) of the slab varied as SpO2 was varied...... 155
11.4 The simulated raw data from the Rytov approximation as a result of introducing a pulse inside the slab by varying Hb and HbO for period of 60 seconds for λ = 690 and 830 nm...... 156
11.5 The physiological AC and DC was extracted from the output signal by averaging over each second interval in order to capture the minimum and maximum of the pulse across the 60 seconds and averaging for 690 and 830 nm...... 157
11.6 The actual oxygen saturation plotted against the calculated oxygen saturation as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 1 cm
for SNR of 100% and the inverse calculation for SpO2 based on pulse oximetry...... 159
11.7 The actual oxygen saturation plotted against the calculated oxygen saturation as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 2 cm
for SNR of 100% and the inverse calculation for SpO2 based on pulse oximetry...... 160
11.8 The actual oxygen saturation plotted against the calculated oxygen saturation as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 3 cm
for SNR of 100% and the inverse calculation for SpO2 based on pulse oximetry...... 161
11.9 The actual oxygen saturation plotted against the calculated oxygen saturation as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 4 cm
for SNR of 100% and the inverse calculation for SpO2 based on pulse oximetry...... 162
xxi 11.10The actual and calculated oxygen saturation plotted against the ra- (AC/DC)690 tio of ratios, R = (AC/DC)830 ; as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 1 cm for SNR of 100% and the inverse calculation for
SpO2 based on pulse oximetry...... 163
11.11The actual and calculated oxygen saturation plotted against the ra- (AC/DC)690 tio of ratios, R = (AC/DC)830 ; as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 2 cm for SNR of 100% and the inverse calculation for
SpO2 based on pulse oximetry...... 164
11.12The actual and calculated oxygen saturation plotted against the ra- (AC/DC)690 tio of ratios, R = (AC/DC)830 ; as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 3 cm for SNR of 100% and the inverse calculation for
SpO2 based on pulse oximetry...... 165
11.13The actual and calculated oxygen saturation plotted against the ra- (AC/DC)690 tio of ratios, R = (AC/DC)830 ; as a result of using the forward Rytov approximation for a slab at a depth of 0.5 cm and a source detector separation of 4 cm for SNR of 100% and the inverse calculation for
SpO2 based on pulse oximetry...... 166
11.14The calculated oxygen saturation plotted against SNR values of 20 − 100; for a slab at a depth of 0.5 cm and a source detector separation of 1 cm and oxygen saturation levels of 50 − 95% using the inverse
calculation for SpO2 based on pulse oximetry...... 167
11.15The calculated oxygen saturation plotted against SNR values of 0−10; for a slab at a depth of 0.5 cm and a source detector separation of 1 cm and an oxygen saturation levels of 60% using the inverse calculation
for SpO2 based on pulse oximetry...... 168
xxii CHAPTER 1
INTRODUCTION
1.1 Statement of Problem
Visual disorders such as amblyopia and infantile nystagmus syndrome (INS) cause a reduction in the ability to see (i.e. visual acuity) and thus patients with these disorders can not fully feel the comfort of 20/20 vision. And if not treated or diagnosed early in life then patients have to live with them throughout their lives without opportunities for treatment. The reason being that a cure does not exist.
1.2 Significance of Research
From the functional magnetic resonance (FMRI) perspective, by mapping the anatomical areas associated with some of these visual disorders we may gain insight and a better understanding. From the near infrared spectroscopy perspective if a device can be introduced into the clinic as an early diagnosis device, then proper treatment can be prescribed before it is too late for the patients, specifically for visual development disorders.
1.3 Organization of Dissertation
This dissertation is divided into several chapters.
1 1. Chapter 2 is a background on the visual pathway with clinical relevance.
2. Chapter 3 is an overview of brain functional studies as it relates to vision.
3. Chapter 4 is an FMRI study on optokinetic nystagmus (OKN).
4. Chapter 5 is a follow up FMRI study on OKN.
5. Chapter 6 is an FMRI study on saccade and pursuit eye movements.
6. Chapter 7 is an FMRI protocol comparison using three planes.
7. Chapter 8 is an algorithm for neuroimage enhancement incorporating three
planes.
8. Chapter 9 is a study to utilize NIRS and FMRI for visual cortex monitoring.
9. Chapter 10 is the development process of a NIRS device for use in the clinical
environment.
10. Chapter 11 is the mathematical simulation and results for the forward and
inverse NIRS algorithm development.
11. Chapter 12 is the concluding remarks.
2 CHAPTER 2
BACKGROUND OF THE ASCENDING VISUAL PATHWAY
2.1 Introduction
In this section will be a description of specific stages of the visual pathway, be- ginning from the distal stimulus and ending in the visual cortex. More importantly, the development of ascending visual pathway will be discussed in order help in un- derstanding various disorders associated with it such as monochromacy, albinism, amblyopia (refractive, strabismic) and infantile nystagmus syndrome (INS).
To motivate the discussion we begin by asking, what is the problem in visual perception? This will be answered briefly. In visual perception, we have both a distal and a proximal stimulus. The distal stimulus is what the subject is looking at, usually at a distance. In the case of vision, it determines the pattern of light arriving at the cornea. The proximal stimulus hits the sense organs directly. In the case of vision, it is the pattern of light arriving at the retina, for instance as a result of looking at the distal stimulus. There are several features that distinguish the distal and proximal stimuli. The distal stimulus is 3-dimensional, independent of point of view, upright, and has no lens blur or filter. An example, of the latter
3 two is that when we look at a person their head is on top and their feet are on the bottom and the physical person does not get blurred. The proximal on the other hand is 2-dimensional, depends on point of view, inverted, blurred and filtered by the lens. So the main problem in visual perception becomes clearer; that is to retrieve information about the distal stimulus with only the proximal stimulus to work with. This is important because it affects the perceptual representation which is the endpoint of the perceptual process. Perceptual representation is the state of the visually-guided motor behavior (keeps us from bumping into things), visual pattern recognition, visual understanding, and memory. Basically, as the subject sees an object (distal stimulus), the input falls on the retina (proximal stimulus) and an output of the distal stimulus is perceived via perceptual representations. Note, that this is not the same as the distal stimulus, because there are two kinds of perception, veridical and illusory. There are many examples of visual illusions, in which the perceptual representation suggests an incorrect distal stimulus. That is, the apparent distal stimulus differs from the veridical distal stimulus. We will not go any further than this on the topic of illusions. With this concept, we can now refine the problem in visual perception, as trying to understand how the visual system creates a perceptual representation of the distal stimulus with only the proximal stimulus as an input.
Why is this a problem? Because the relationship of distal to proximal is not one to one, that is a distal stimulus can be seen as many proximal stimuli and proximal stimuli can be many distal stimuli. This leads to the inverse problem of trying to recover a visual representation from the input, even when many representations are consistent with the proximal stimulus. Thus, this is a motivation to begin discussing the visual pathway and understand the retinal (proximal) input to the brain.
4 2.2 Visual Pathway
The visual pathway consists of many stages as illustrated in Figure 2.1. A dis- cussion of each block in this diagram is beyond the scope of this report; however, we will focus on the ganglion cells, lateral geniculate nucleus (LGN), and the primary visual cortex (V1). The ascending visual pathway begins when light hits the back of the retina and stimulates the photoreceptors (rods and cones). These photoreceptors transform radiant energy into electrical activity, which is transmitted to retinal bipo- lar cells and then into retinal ganglion cells. Figure 2.2, presents a detailed layout of the retinal layer and labels all the sub-layers and their respective cells. Each of these cells play a role in the visual system and have their own receptive fields. Again, in this report we choose to focus and discuss the ganglion cells.
2.2.1 Ganglion Cells
There are two major classes of ganglion cells. The smaller midget, or parvo, cells comprise about 80 percent of these cells and the larger parasol, or magno, cells about
10 percent [2]. As with other cells in the retina, these ganglion cells have their own receptive fields. Namely, they have a center surround receptive field, with either on-center (off-surround) or off-center (on-surround). There are several differences between these two types of cells. Parvo cells are dominant in the fovea as opposed to the magno cells, which are dominant in the periphery. The parvo cells are also characterized as having a sustained response while the magno have a transient re- sponse [3, 4]. At any given eccentricity, parvo cells have a higher spatial resolution, lower contrast sensitivity, slower conduction velocity, and a more sustained response than do magno cells [5]. The parvo cells have low contrast sensitivity and detect
5 Figure 2.1: The visual pathway [1]. 6 Figure 2.2: The layers of the retina and component cells. Note that light has to pass from the front of the eye through the vitreous body and all the retinal layers to finally reach the receptor cells, rods, and cones [1].
7 color and form, while the magno have high contrast sensitivity and detect motion.
Parvo cells rarely respond well to luminance contrasts below 10%, whereas magno cells often respond to stimuli with contrasts as low as 2% [5–7]. In addition to these two, there are other types of ganglion axons that exist; the more common of these are the konio cells which are small bistratified cells [8]. They are common in the parafovea and have low contrast sensitivity and detect color. The major difference between the konio cells and the other two is that the konio have a uniform receptive
field and thus have no spatial opponency. To many investigators the term konio has become synonymous with the blue-yellow pathway, just as parvo is now equated, too simplistically, with the red-green pathway [9]. But this is not always the case because, konio cells constitute a heterogeneous population of cells, some lacking blue-yellow color opponency [10]. The axons of all these ganglion cells exit the eye, forming the optic nerve and synapse in the midbrain. Since the diameter of the optic nerve and the number of the ganglion cell axons it contains are limited by the structure of the skull, not all the information that falls upon the retina is transmitted to the brain proper [11]. Although there are more than 100 million photoreceptors within the retina, there are only 1 million ganglion cells, revealing an extensive degree of neural convergence [12, 13]. At the optic chiasm, ganglion cell fibers from the nasal retina of each eye cross over to join the temporal fibers of the fellow eye to form the optic tract [11]. The long axons of the retinal ganglion cells leave the eye, form the second cranial nerve (the optic nerve), and synapse in the dorsal lateral geniculate nucleus
(dLGN), a midbrain structure [11]. We will now discuss the LGN.
8 2.2.2 Lateral Geniculate Nucleus (LGN)
The primary target of the optic tract is the dorsal lateral geniculate nucleus
(dLGN), a thalamic nucleus. In higher vertebrates, such as carnivores and primates, axons from the two eyes converge onto their primary target, the dorsal lateral genicu- late nucleus (dLGN), but occupy distinct regions (the eye-specific layers) within this target [14–16]. In primates [17, 18], the axonal terminals of ganglion cells of the two eyes initially share common territories within the dLGN, but through a process that eliminates inappropriately placed branches, projections from the two eyes become restricted to their appropriate layer. Most, but not all, retinal ganglion cells synapse in the six-layered structure. Layers 2, 3, and 5 receive input from the ipsilateral eye, whereas layers 1, 4, and 6 receive input from the contralateral eye. The dorsal four layers, which are constituted of comparatively small neurons called parvo, or
P-cells, are the parvocellular layers (layers 3,4,5,6). Larger neurons, commonly called magno or M-cells, comprise the two ventral magnocellular layers (layers 1,2). Axons from midget ganglion cells synapse on P-cells in the dLGN to form the parvo path- way, while axons from the parasol cells synapse on dLGN M-cells to form the magno pathway. The layers between the parvocellular and magnocellular layers contain very small neurons (konio cells). Studies have shown that konio cells provide the only di- rect geniculate input to layers 1-3 [19]. The subcortical projection from the retina to cerebral cortex is strongly dominated by the two pathways (M and P pathways) the magnocellular and parvocellular subdivisions of the lateral geniculate nucleus [20].
The parvo layers receive input from color-opponent midget ganglion cells, whereas the magno layers are supplied by broadband parasol ganglion cells [21]. Parvo path- way neurons show color opponency of either the red/green or blue/yellow type, which
9 means that they respond to color change regardless of the relative luminance of the colors [22]. The blue-yellow ganglion cells project to the konio layers just ventral to the third and fourth parvocellular layers [23]. Layers 5 and 6 have on-center receptive
fields, and layers 3 and 4 have off-center receptive fields. Layers 1 and 2 have both on- and off- center receptive fields. Figure 2.3, shows the projections from the retina to the LGN.
2.2.3 Primary Visual Cortex (V1)
The cells of dLGN send most of their axons to the cerebral cortex, specifically, the primary visual cortex (V1), as seen in Figure 2.4 along with the visual field representation in the retina and primary cortex. As a side note, Figure 2.4 also illustrates the cortical magnification factor, meaning the fovea has more space allo- cated to it on V1 than the peripheral retina. Inputs to V1, which are stratified by magno, parvo, and konio, become thoroughly intermingled by passage through the elaborate circuitry of V1 [9]. There are about 8/9 layers in V1. Layer 4 consists of three sublayers, 4A, 4B, and 4C. Layer 4C also is subdivided into 4Cα, and 4Cβ.
The projections from the LGN go specifically to layer 4C and the information flows up and down from there [24]. The projections from parvocellular layers terminate primarily in layers 4A and 4Cβ, whereas those from magnocellular geniculate termi- nate in layer 4Cα [25]. Layer 4B receives direct input from 4Cα (M pathway), but not 4Cβ (P pathway) [26,27]. Layer 4Cβ projects to the blobs and interblobs [28,29].
The blobs also receive major inputs from the M pathway by way of layers 4B and
4Cα [25,30–32]. Figure 2.5 gives the details of these connections.
10 Figure 2.3: Retinal ganglion cell projections to the lateral geniculate nucleus (LGN) of the thalamus. Note that layers 1,4, and 6 of the LGN receive visual information from the contralateral retina, whereas layers 2,3, and 5 receive visual information from the ipsilateral retina [1].
11 Figure 2.4: The visual field representation in the retina and primary cortex [1].
12 Figure 2.5: The block diagram of ganglion cell mapping from retina through LGN, V1, and other cortical areas.
More recently, Yazar et al. 2004 [33] have found that some geniculate fibers terminate in both layers 4Cβ and 4A, implying either a direct parvo input to 4A or a konio input to 4Cβ. In layer 3B the cells in blobs and interblobs receive input from parvo (4Cβ), magno (4Cα), konio (4A), or mixed (4B) layers, in a range of relative synaptic strengths [34]. Cells in both 4Cα and 4Cβ project to layers 5 and 6 [26,35].
Feedback from layer 6 to the LGN is segregated only partially with respect to magno and parvo, thus mixing the geniculate channels [36]. There two main types of cells in
V1, stellate and pyramidal. The stellate cells are small interneurons found in layers
2-6 and the pyramidal cells are large relay neurons found in layers 2, 3, 5, and 6. The stellate cells are simple cells because of their receptive fields. The pyramidal cells are complex cells. The simple cells’ receptive fields are of a certain size, are oriented in a certain way, and are sensitive to phase. They increase their rate of firing when stimulated in some places, and reduce it when stimulated in other places. The simple cells respond to a single spot of light and are additive and linear. The complex cells do not respond to a single spot of light, rather they respond to edges and bars, and are not sensitive to spatial phase. Many of the complex cells respond only or best
13 to stimuli that move in one direction. So, if the stimulus is stationary, in opposite direction or a spot of light then the complex cells receptive field will have no response.
The complex cells are non-additive and are non-linear. Both the simple and complex cells respond to most proximal stimuli. All together, these cortical cells are tuned for spatial frequency, position, and orientation.
2.3 Development of the Ascending Pathway
We now describe how the visual pathway develops and the affects of abnormal development. During development anatomical projection patterns are restructured and functional reorganization takes place [37–40]. And there are at least two ways by which neurons can be wired up accurately: Connections may be specified from the outset or synapse formation may initially follow an approximate wiring diagram, with precision achieved by the elimination of inappropriate inputs and the stabilization and growth of appropriate connections [41, 42]. The ganglion cells, LGN, and V1 are all wired up in a retinotopic fashion; meaning that the order of points on the retina
(proximal stimulus) are preserved. In this mapping, the points that are further away from each other on the retina will be further away on the brain. It is easy to see that the proximal image is retinotopically related to the distal stimulus, simply because of the optics of the eye. However the retinotopic mapping from the retina to the
LGN and from the LGN to V1 is harder to appreciate. Studies of patients with localized cortical damage showed that the receptive fields of neurons within area V1 are retinotopically organized [43–45]. As a matter of fact, the development of the retinotopic map is a general process for the central nervous system. Cell bodies are born early in embryogenisis; axons and dendrites come later. The nerve growth is then
14 guided mechanically, probably by glial cells, to their overall destination. The patterns of activity of the neurons themselves determine the exact position of the synapses that are formed. Ganglion cells travel up the concentration gradient to the LGN. Target cells send guiding chemical messages, giving crude directions to the cells’ overall destination by their concentration gradient. The chemical signposts act like beacons that attract the cells to project to approximately the correct part of the target tissue.
At the same time the chemical signposts repel growth cones from the wrong axons.
These guidance molecules also govern the decussation at the optic chiasm by signaling the retinal ganglion cells to either cross or not to to cross. The activity of adjacent retinal ganglion cells is correlated [46], and waves of activity sweep across the retina during early life [47]. Although the waves could potentially underlie the refinement of many retinal projection patterns, activity may not be required for establishing the
M and P pathways of the primate retina that develop prenatally, and which show no apparent gross structural refinement with ensuing development [48]. The immature and light-insensitive retina spontaneously generates a pattern of rhythmic bursting activity during the period when the connectivity patterns of retinal ganglion cells are shaped [49]. After the cells find a region, the wave then enforces precise ordering at the target. Thus the retinotopic map is finalized via the wave. Prenatal refinement of the retinotopic projections is achieved by these spontaneous waves of activation that propagate across the retina. Here ganglion cells are linked together by means of electrical synapses in a rough network and charge fluctuates randomly. The random response of one cell starts a wave of activity and the cells that fire together will eventually wire together. These spontaneous waves cause neighboring retinal regions to fire at about the same time. In fact, the correlation between the responses of
15 cells is directly related to their separation on the retina [49]. So, the first principle of refinement is that cells that are neighbors tend to respond together. The second principle of refinement is that cells that fire together wire together. If there are two cells, 1 and 2, that are close to each other on the retina then when they fire together they will form neighboring synapses at the LGN. But cell 3, which is far from the first two on the retina will fire separately and thus synapse at the LGN separately. This is how the LGN is retinotopically wired up at birth along with V1 and other retinotopic cortical areas. Hence, the waves in the prenatal retina setup the relation between retina and brain. As for the postnatal retina, responses to stimuli set up the relation between the proximal stimulus and the brain. The postnatal wave may help guide the formation of synapses and determine which erroneous synapses are cut out for the normal mapping. When they arrive at their destinations, each process synapses over a relatively large area. These synapses are over a large region where they are likely to synapse eventually. Since target cells have lots of cells synapsing onto them, there are a lot more synapses present in V1 at 6 months and 1 year than in an adult.
The process of the synapse starts as each axon from different cell bodies tries to take over a large piece of visual cortex and inevitably overlap occurs. At these regions of overlap a competition occurs, and the cell with the most or strongest synapse claims that region and the other synapses pull back. This synaptic elimination is a key element in the refinement of connectivity in both the central and peripheral nervous systems [41, 42, 50–52]. This produces a retinotopic map that has less overlap than before, and has many fewer synapses. If there is a vacant area then other nearby cells synapse onto it without meeting any competition and in turn increase their synaptic field. This process of being able to change as a result of experience is called
16 plasticity, and is required for development. It determines how the visual system is wired up during normal or abnormal development. The synaptic development occurs at different time scales across the brain. For V1 the development ends from about 8 to 16 years and culling happens at about 1-2 years. If there is any difficulty or blur in one eye or an eye turn while these synapses are being formed and refined, the subject will develop a visual disorder. This leads us into the next section.
2.3.1 Disorders of the Ascending Pathway
We will now discuss several visual disorders associated with the ascending pathway.
The disorders are: albinism, refractive amblyopia, and strabismic amblyopia.
Albinism
Albinism is characterized by a systematic misrouting of the connections between the retina and the visual cortex. The ascending projection in an albino favors the contralateral hemisphere. Note the normal projection that is crossed is about 55%.
This miswiring can produce nystagmus and strabismus. The clinical features of al- binism include hypopigmentation of the fundus, and iris. There are variable degrees of pigmentation of the iris, hair, skin. Tyrosenase negative albino (oculocutaneous) individuals may be completely white with a visual acuity range from 20/60 - 20/400, but is usually worst than 20/200. Tyrosenase positive albino may look hypopigmented or even essentially normal with visual acuity range from 20/60 - 20/400, but is usu- ally better than 20/200. More clinical features related to the eye include a very light fundus because there is no melanin in the retinal pigment epithelium (RPE). There is little differentiation of the fovea from the surrounding retina. Albinos also have high myopia or high hyperopia. In the albino system there is more than 90% decussation at
17 the optic chiasm. This means that the guidance molecules during development failed to stop the neurons from going the opposite direction. For a better understanding of the ascending pathway abnormalities in albinos we will do a comparison with normals.
If a distal stimulus is presented on the right hand side of a normal subject then the expected pathway from the right eye nasal retina would cross the optic chiasm and end up in the contralateral visual cortex (left visual cortex). For the same stimulus on an albino subject, the resulting signal would be the same as the normal. If the distal stimulus is changed to the left hand side for the normal, and looking at the right eye temporal retina, then the signal would not cross the optic chiasm and would end up in ipsilateral visual cortex (right visual cortex). The same repeated for the albino reveals the opposite since the majority of the neurons cross the optic chiasm and end up in the contralateral visual cortex. The primary abnormality in albinism is a genetically determined lack of melanin or melanosomes and abnormal decussation of the early afferent visual pathways. As a side point, melanin is very important for many aspects of neurological development. For instance, the neural crests pigment and its location on the embryo is determined by melanin. Melanin is also involved in production of dopamine and serotonin and many other neurotransmitters related to neuroendocrine function.
Refractive Amblyopia
Refractive/deprivation amblyopia is a result of the receptive fields not being used early in life thus the culling at about 1 year postnatal removes their synaptic connec- tions because lack of function. Specifically, the proximal stimulus is blurred during the critical period, meaning the high spatial frequencies are reduced or eliminated from the visual image, causing high spatial frequency tuned channels to either never
18 develop, or be lost. In this case, the low spatial frequencies pass unattenuated, so the low spatial frequency tuned channels develop normally. The effect of this blur in refractive amblyopia is the direct loss of contrast sensitivity at high spatial frequen- cies, which is equivalent to a loss of visual resolution acuity. As for the remaining spatial frequency channels, they stay relatively normal because they are stimulated approximately normally during the critical period. This illustrates the principle that the receptive fields must be used if they are to be maintained. If the proximal stim- uli do not stimulate the receptive fields effectively, the cells tend to stop responding to the intended stimulus even if it is presented occasionally. The cell may begin to respond to other stimuli, and therefore develop a new receptive field. The input from the other eye is likely to grab the synapse area because of competition. As a result of this anisometropia, an unequal refractive error in the two eyes, the receptive fields will be different than normal fields. Thus, the eye with the larger refractive error continues to experience chronic blur. Dominance of the good eye becomes exagger- ated during development, because of competition between incoming signals. Most cells in the primary visual cortex come to have predominant input from the good eye. If one eye is handicapped during the competition, it tends to lose its synaptic connections. Thus, the development of ocular dominance columns in amblyopia is distorted, and depends on the age at which deprivation begins. The most dangerous periods of refractive amblyopia are in the first 6 months.
Strabismic Amblyopia
The cells in the ascending pathway are labeled lines. Labels relate to position on the retina and therefore position in the proximal stimulus. Labels also relate to spatial frequency and orientation. Labeled lines are important because the brain only
19 knows what the ascending pathway tells it. If the labels are abnormal, vision is also abnormal. In strabismic amblyopia, the lines are mislabeled, which leads to distorted vision. In normal retinotopic organization, labels relate position in the distal stimulus to position upon the retina. Strabismic amblyopia is thought to be due to disordered
(scrambled) retinotopic mapping between the LGN and V1 of the signals from one eye; therefore, leading to abnormal visual experience. The waves that happen after birth are not normal because the eye is not always pointing in the right direction. Re- call that cells fire together after birth because of the wave of activity produced by the usual retinal stimulus. This postnatal wave may help guide the formation of synapses and determines which erroneous synapses are cut out for the normal mapping. This eye turn in early childhood produces an abnormal wave. The connection between the retina and the LGN remains normal because it is wired up prenatally, but the connection between the LGN and V1 can be influenced by postnatal development.
When cortical cells fire together abnormally they wire together abnormally. Clinical consequences of this disorder at the primary visual cortex are impaired visual recog- nition, crowding (nearby stimulus information obscures attended item), poor vernier acuity, poor stereo acuity, poor grating orientation identification acuity, and often near normal grating resolution acuity. The high spatial frequency gratings do not look like uniform gray, so they can be detected, but they are badly distorted, so the amblyope cannot discriminate between vertical and horizontal.
Infantile Nystagmus Syndrome (INS)
What causes infantile nystagmus syndrome (INS)? INS may be considered a com- mon phenotypic manifestation due to an abnormality in one of many genes influencing the development of oculomotor control [53]. At least three genes are responsible for
20 INS, as X-linked, autosomal dominant and autosomal recessive pedigrees have been described. Chromosome 6p12 is the first reported genetic locus linked to INS [54].
The nystagmus itself is the direct result of an ocular motor control instability that may develop with or without an accompanying sensory deficit [55]. This ocular motor instability may affect the oculomotor nerve (cranial nerve III). This nerve originates in the midbrain below the cerebral aqueduct, and supplies all the extrinsic muscles of the eye except the lateral rectus and the superior oblique muscles (innervated by cranial nerves VI and IV respectively), and supplies the elevator muscle of the up- per eyelid, the ciliary muscle, and the sphincter muscle of the pupil. For a better understanding of these muscles we will now discuss the a brief anatomy of the eye.
INS is one of about 49 forms of nystagmus as has been described through eye movement recordings [55]. Some other forms of nystagmus are latent, pendular, and jerk. Latent nystagmus is where the nystagmus is worst when one eye is occluded.
Pendular nystagmus is a slow rhythmical and horizontal movement of the eyes. Jerk nystagmus is where the nystagmus varies depending upon which way the eyes are looking. There are 12-14 INS waveforms identified and are variations of two of the above types - pendular and jerk - with periods of extended target foveation present during each cycle [55,56].
2.4 Conclusion
There are three main principles in visual development: labeled lines, cells firing together wire together, and synaptic competition. In summary, sensory cells send the same kind of signal regardless of how or how strongly they are stimulated (labeled lines). The relations between the retina and the LGN, and between the LGN and
21 the cortex, are crudely wired up at birth, by prenatal visual experience of the wave.
That wire up is refined and related to the proximal stimulus by genuine postnatal visual experience and synaptic competition. This refinement includes creation of new synapses and culling of old ones. Abnormalities early in life can cause disorders in the visual pathway. Rod monochromats do not have the normal photoreceptor connections from the retina and thus the rods take over the synaptic fields where the fovea usually falls in V1. Albinos seem to have a disfunction in the chemical signposts that separate the nasal and temporal retina projections. In refractive amblyopia, there is a blur in the proximal stimuli of one eye and high frequency cells are not fully developed in V1 because they are cut out during the refinement process. Strabismic amblyopes suffer from an eye turn early on that causes an abnormal wave which leads to miswiring between the LGN and V1. As for INS, it is still not clear what the source is. For all these disfunctions, the use of technology can help in diagnosis or even isolate the cause of the problem in order to assess effectiveness of treatment.
This will be discussed in the next section, specifically the use of functional magnetic resonance imaging (FMRI) and functional near infrared spectrsocopy (FNIRS).
22 CHAPTER 3
OVERVIEW OF BRAIN FUNCTIONAL STUDIES
3.1 Introduction
Studying brain functional activities is an area that is experiencing rapid interest in the fields of biomedical imaging. Although when the term functional appears, the
first thing that most people would think of is magnetic resonance imaging (MRI), which indeed is a robust modality. However, just like with any other modality, it has some drawbacks. This is why functional MRI (FMRI) researchers are turning to new modalities either to simultaneously use or run standalone. Near infrared spectroscopy
(NIRS) is one of these modalities because of its high temporal resolution. Currently there have been and are efforts to combine NIR with MRI [57–62]. In this chapter, we will discuss both NIR and MRI studies with focus on visual cortex applications. The remaining sections of this chapter will be highlighting the work done on the visual cortex from both the FMRI and NIRS ends. The first half will be on FMRI studies and the second will be on FNIR studies. We will then conclude with a discussion.
23 3.2 Functional Magnetic Resonance Imaging(FMRI)
Over the last decade, functional magnetic resonance imaging (FMRI) has been developed as a technique for mapping brain activation and has found widespread interest in basic and clinical research aimed at better understanding brain function.
The FMRI technique is based on the detection of local perturbations of the deoxyhe- moglobin concentration in the vicinity of neuronal activity. Neuronal activity at the synaptic level results in both an increase in oxygen consumption by the active cortex and an even greater increase of blood flow to the site. Because oxygen delivery exceeds oxygen utilization, the net effect is a local decrease in deoxyhemoglobin concentration near the activation site. The decrease in deoxyhemoglobin concentration at the site of neuronal activity causes a local increase the magnetic resonance signal. This effect has been termed the Blood Oxygenation Level Dependent (BOLD) contrast mech- anism [63]. BOLD FMRI capitalizes on the difference between two conditions; an active condition in which stimulus related neural activity is generated and a passive condition in which stimulus related neural activity is absent or kept to a minimum.
3.2.1 FMRI Visual Cortex Studies
As a result of the increase in general FMRI studies, there has also been an increase of studies investigating many aspects of the visual cortex. These studies include nor- mal eye movements such as optokinetic nystagmus (OKN) [64–72], saccades [73–86], and smooth pursuit [68, 77, 79, 83, 87–89]. There have also been studies that look at
24 varying aspects of visual perception such as: effect of age [90, 91], retinotopic map- ping [92–97], magnocellular (M) and parvocellular (P) pathways [57,98], ocular dom- inance [99–101], binocular rivalry [102], illusory contours [103, 104], contrast detec- tion [105], visual attention [106,107], perceptual filling-in [108], lateral geniculate nu- cleus (LGN) [109–119], superior colliculus (SC) [120], motion perception [121], and il- lusory perception of real motion [122]. There have also been FMRI studies undertaken for abnormal visual functions such as: amblyopia [90,91,123–133], albinism [134], op- tic neuritis (ON) [135–141], Autism [142, 142, 143], and macular degeneration [144].
Other studies include looking at callosal agenesis and colpocephaly [145], vascular lesions and therapeutic intervention [146], migrane aura [147], and idiopathic Parkin- sons disease [148], as examples. We will now discuss some of the results of functional studies in normal visual development.
FMRI and Normal Vision
FMRI studies of the oculomotor function have been mostly limited to normal sub- jects and have concentrated on voluntary pursuit and saccadic eye movements and optokinetic nystagmus (OKN). Tanabe et al. 2002 [89] have noted that FMRI studies of oculomotor function have employed few subjects and the reliability of mapping-out brain sites involved in oculomotor control have not been established. This statement was made almost 5 years ago and a lot has been accomplished since then. Overall there appears to be two parallel cortical oculomotor systems for pursuit and saccadic eye movements. Both pursuit and saccadic eye movements appear to activate the same cortical areas including the frontal eye fields (FEF, precentral cortex), supplemen- tary eye fields (SEF, superior frontal cortex), parietal eye fields (PEF, intraparietal cortex), precuneus, and MT/V5. However, pursuit or saccadic eye movements may
25 selectively activate subregions of these cortical areas. Petit and Haxby [68] found that the pursuit related activation areas were usually smaller than and consistently inferior to or/and posterior to the saccadic related activation areas. Dieterich et al.
2000 [66] have shown that small field horizontal OKN as well as voluntary saccadic eye movements activate areas of both cerebellar hemispheres including the superior semilunar lobule, simple lobule, quadrangular lobule and inferior semilunar lobule. In addition, activation was found in the middle cerebellar peduncle, dentate nucleus, cul- men (medially), and uvula of the cerebellar nuclei. Fixation during OKN suppressed activation in the uvula and culmen. Dieterich et al. 1998 [65] also found OKN to activate subcortical areas including the caudate nucleus, putamen, globus pallidus and paramedium thalamus. Fixation increased activity in the FEF and anterior cin- gulate gyrus. Dieterich et al. 2000 [66] used a rotating drum that contained colored
figures to stimulate OKN amplitude that ranged from 2−13o visual angle, suggesting a mixture of voluntary and involuntary OKN or only voluntary OKN. Most recently it has been shown that voluntary OKN generates more higher level cortical activa- tion than does involuntary OKN [69, 72]. Bense et al. 2006 [70] found that there was no direction dependent activation in cortical eye fields, but there was asymmetry in the paramedian visual cortex areas. Also they found stronger activation in the hemisphere contralateral to slow OKN phase (pursuit). Bense et al. 2006 [71] found cerebellar activation was localized in the oculomotor vermis.
Saccades in humans have been found to activate the precentral sulcus in FEF and in the precuneus along the intraparietal sulcus (IPS), extending in both superior and inferior parietal lobules [78]. Saccades are traditionally divided into reflexive and voluntary saccade. Mort et al. 2003 [84], demonstrated that voluntary saccades
26 produced greater activation within FEF and the saccade related area of IPS. An oculomotor study on oscillatory, predictable and unpredictable saccade [85] showed that predictable saccades which have the shortest saccadic latency led to the most pronounced cerebral activity both in terms of cortical areas involved and signal inten- sity. Saccades are also distinguished as either pro or anti if they are made toward or away from a stimulus respectively. Cornelissen et al. 2002 [82] found similar BOLD activation in FEF during both pro- and antisaccades. It was also suggested that the presupplementary motor area (pre-SMA) coordinates with the FEF to maintain a controlled, preparatory set for task appropriate oculomotor execution in a study looking at functional interactions between pro- and antisaccades [86]. Saccade fre- quency and amplitude was varied [80] and high correlation between frequency and
BOLD signal was found along with higher BOLD activation in antisaccades over prosaccades. Merriam et al. 2001 [81] found that comparison of visually guided sac- cades with fixation revealed activation in all three cortical eye fields: supplementary eye field (SEF), FEF, and parietal eye field (PEF).
FEF activation during smooth pursuit performance was found to be smaller than during saccades [77]. The performance of pursuit eye movements induced activations in the cortical eye fields also activated during the execution of visually guided saccadic eye movements, namely in the precentral cortex [frontal eye field (FEF)], the medial superior frontal cortex (supplementary eye field), the intraparietal cortex (parietal eye field), and the precuneus, and at the junction of occipital and temporal cortex
(MT/MST) cortex [68]. Rosano et al. 2002 [83] localized the saccade-related area to the upper portion of the anterior wall of the precentral sulcus and the pursuit-related area to a deeper region along the anterior wall, extending in some subjects to the
27 fundus or deep posterior wall. It was suggested that the lateral occipitotemporal cor- tex has extraretinal signals during pursuit [87]. Significant activation in V1 and V2 in both hemispheres as well as additional bilateral activation in the lateral extent of
Brodmann’s area 19 and 37 (BA 19/37) was evident during smooth pursuit [88]. Pur- suit performance, relative to visual fixation, elicited activation in three areas known to contribute to eye movements in humans and in nonhuman primates: the frontal eye field, supplementary eye field, and intraparietal sulcus. It also activated three me- dial regions not previously identified in human neuroimaging studies of pursuit: the precuneus and the anterior and posterior cingulate cortices. All six areas were also activated during saccades [79]. Tanabe et al. 2002 [89] found activation consistently in dorsal cortical eye fields and cerebellum. Many studies are still being pursued on normal eye movements with hopes of mapping out or isolating specific anatomical areas responsible with the goal of future diagnostic and therapeutic interventions.
The general perception studies are too many to list/discuss thus we briefly mention a few related to visual perception. In general, the volume and degree of FMRI activation was found to decrease with increasing age, particularly over the age of
40 years [90, 91]. Differentiation between the magnocellular and parvocellular visual pathways has been recently demonstrated [98]. Goodyear and Menon [100] were the
first to demonstrate reproducible high resolution (0.55mmx0.55mm) capabilities of
FMRI in humans when using short duration (< 6sec) visual stimuli. Conner et al. 2004 [96] compared retinotopic maps of children with adults in hopes that the study would be useful reference for studies of children with visual disorder, such as amblyopia.
28 FMRI and visual dysfunctions
FMRI studies have been undertaken in normal subjects and in patients with am- blyopia, commonly known as lazy-eye [90, 91, 123, 126, 127, 129–131]. Goodyear et al. 2000 [123] showed that there were always fewer activated FMRI voxels during amblyopic stimulation than during normal eye stimulation. Algaze et al. 2002 [126] also showed that the volume and level of occipital visual cortical activation was less from the amblyopic eye compared to the dominant eye of amblyopes or to normal eyes. Rogers [129] and Algaze et al. 2005 [127] have shown that L-dopa, a drug used in the treatment of Parkinson’s disease, caused a reduction in volume of activation of occipital visual cortex while it improved visual acuity - a counterintuitive finding.
Yang et al. 2003 [128] showed that the volume ratio between the amblyopic and sound eye stimulation significantly increased after L-dopa treatment. More recently, the amblyopic eye showed marked reduction in activation in the fusiform gyrus, with normal activation in the collateral sulcus [132]. Responses to grating stimuli showed reduced responses in higher areas on the central visual pathway [133].
In albinism, there is an abnormal chiasmic projection system which favors the con- tralateral hemisphere [134]. For example, in oculocutaneous albinism and in ocular albinism, monocular stimulation yields a greater FMRI response in the contralateral hemisphere than the ipsilateral hemisphere because of misrouting of the eye’s affer- ents favoring the contralateral hemisphere. After using standard FMRI statistical analysis tools, the number of voxels activated in each hemisphere were counted for each subject. A crossing ratio was then computed by subtracting the voxels activated contralaterally from the ipsilateral ones and dividing by the total number activated.
The mean of these ratios for left and right eyes were then calculated for correlations.
29 Reduced signal and greater asymmetry in the visual cortex has been shown in optic neuritis (ON) patients, compared with controls [138]. They also showed that the volume of visual cortical activation was significantly correlated to the result of the contrast sensitivity test. They used an asymmetry index, Ia, to calculate the relative difference between size of activated area in the left and right hemisphere, in a similar fashion to the above study. This was done by simply counting the number of voxels in each hemisphere and taking the absolute value of the difference and dividing by the total number of voxels in both hemispheres. A value of Ia=1 meant 100% asymmetry while a value of Ia = 0 meant no asymmetry.
Toosy et al. 2002 [139] showed that visual cortex activation is reduced dur- ing photic stimulation, whilst extra-occipital areas are extensively activated with a peak blood oxygen level dependent response during the OFF phase of the stimulus paradigm. More recently they suggested a genuine adaptive role for cortical reor- ganization within extrastriate visual areas early after optic neuritis [140]. Reduced activation was seen in V1 during stimulation of the affected eye, compared to the normal eye [141].
Parents of children with autism or Asperger Syndrome (AS) showed atypical brain function during both visual search and emotion recognition [143]. Hadjikhani et al.
2004 [142], found that retinotopic maps of individuals with autism were similar to normal subjects, indicating that low level visual processing is normal. A case study by Sunness et al. 2004 [144], illustrated that retinotopic mapping can be performed successfully in patients with central scotomas from macular disease. The ability to look at anatomical reorganization of the visual cortex was demonstrated in a case of callosal agenesis and colpocephaly [145], and in alteration by vascular lesions [146].
30 Data from Hadjikhani et al. 2001 [147], suggested that an electrophysiological event such as cortical spreading depression (CSD) generates migraine aura in the visual cortex. This was determined using a standard t statistic computing the difference between activation amplitude during off period preceding aura. The time courses for independent voxels were then extracted from specific visual areas. A reference baseline (mean) and standard deviation was computed on the first 6 cycles and the pixels that exhibited a higher mean plus standard deviation and a standard deviation less than the reference standard deviation for at least 2 cycles were considered as activated. The visual cortex of patients with idiopathic Parkinsons disease with and without visual hallucinations were examined by Holroyd and Wooten [148]. They found that patients with visual hallucinations had increased activation in the visual association cortex and deficits in the primary visual cortex. Again these are samples of the FMRI studies published in literature. As we will see in the next section NIR is not nearly as advanced as FMRI.
3.3 Near Infrared Spectroscopy (NIR)
Functional near-infrared spectroscopy (FNIRS) allows the ability to monitor brain activation by measuring changes in the concentration of oxy- and deoxy-hemoglobin
(Hb) by their different spectra in the near-infrared range [61,149–157,157–193]. Func- tional imaging with NIR light is made possible in a spectrum window that exists within tissues in the 700−900nm NIR region, in which photon transport is dominated by scattering rather than absorption. For more than two decades, the single channel measurement technique of NIR spectroscopy has been successfully used to measure the hemodynamic response to brain activity in both adults and neonates [171, 180, 184].
31 NIR light interacts with biological tissue through scattering and absorption. Its res- olution is intrinsically limited by the diffusive nature of NIR light in tissue. More recent studies using the multi channel NIRS technique, NIR optical topography (OT), have improved spatial and temporal resolution in both adults [194] and infants [182].
3.3.1 FNIR Visual Cortex Studies
The first clinical research use of NIRS in 1985 was in newborns [149]. Wyatt et al. 1986 [150] were the first to perform quantitative spectroscopy. Although a few, there have been studies that have shown NIRS can detect responses by using visual stimuli [155, 160, 174, 178, 179]. Jasdzewski et al. 2003 [179] compared the hemody- namic response of motor and visual stimulation. Plichta et al. 2006 [193] analyzed reproducibility and reliability of visual cortex measurements using a 52 channel sensor head. They showed good reliability at the group level and excellent reproducibility whereas single subjects’ were mediocre. Schroeter et al. 2006 [187] showed that ig- noring circadian variability in FNIRS is valid in the primary visual cortex. Kusaka et al. 2004 [185] used photostimulation in sleep infants and showed they have dif- ferent results than in adults. This could be a consequence of the still developing brain of the infant. Wolf et al. 2003 [183] were able to detect small fast neuronal signals in the human visual cortex during visual stimulation. They also showed that the brain responds stronger to a double reversal over a single checkerboard stimulus.
Stimulating different quadrants of the visual field in a replication study showed good predictability of results [181].
32 NIRS may be useful in detecting visual dysfunction objectively and noninvasively in patients with visual disturbance, especially when used at the bedside [167]. Specif- ically, Miki et al. 2005 [189] demonstrated that a decreased activation of the visual cortex in patients with optic neuritis can be seen with NIRS. This is the only clinical study related to the visual cortex that has been performed. In the analysis they per- formed a t test (p < 0.05) on each subject between mean values during two conditions
(right eye stimulation versus resting). Also, Wilcox et al. 2005 [190] has studied the relation between object processing and brain function in human infants, and observed neural activation in the primary visual cortex and the inferior temporal cortex. Meek et al. 1995,1998 [160,161] also looked at visual cortex activation in awake infants.
Unfortunately, using a single channel NIRS technique has the major constraint of spatial resolution; hence multiple NIRS systems were used for optical topography
[158]. The first report of optical topography on premature babies was by Chance et al. 1998 [164] that used nine sources and four detectors using a continuous wave
(CW) system. In a more recent study, optical topography in awake infants [182] was demonstrated using a Hitachi system. This multi-channel system used a CW source and generated two wavelengths (780 and 830nm). Twenty laser diodes and eight avalanche photodiode detectors were used and separated into individual light sources by 48 lock-in amplifiers. Schroeter et al. 2004 [186] also used a Hitachi system where twenty two channels were simultaneously measured at a sampling frequency of 10Hz in reflection mode and with an emitter diode spacing of 3cm. Toronov’s group [173], used a 32 source-detector channels (16 sources, 2 detectors) and a TD system for optical topography of the motor system. Wavelengths of 758 and 830nm were sourced through 400µm core diameter optical fibers with the 16 sources operating
33 in a sequential multiplexing mode. Two glass fiber bundles (3.2mm internal diameter)
were connected to the photomultiplier tube (PMT) detectors. Another CW system
was used by Franceschini et al. 2003 [170], which employed 16 laser diodes (8 at
690nm, and 8 at 830nm) and 16 APDs. The laser diodes were frequency encoded by
steps of approximately 200Hz between 4kHz and 7.4kHz, so their signals could be acquired simultaneously by 16 parallel detectors. Each detector output was digitized at 40kHz. There have also been FD system studies as well by Franceschini et al.
2000 [169] with a time resolution of 160ms. Their system which was produced by
ISS Inc. used eight sources at each of two wavelengths (758 and 830nm) and two detectors. Everdell et al. 2005 [188] used an FD system in which they introduce a novel approach to optical topography by using software to demultiplex multiple source signals which are modulated at different frequencies in parallel. Currently, their system uses 16 laser diode sources (8 at 785nm, and 8 at 850nm) and 8 APDs
(3mm diameter) but will eventually increase this to 64 sources and 32 detectors. This would then require 100 lock-in amplifiers. Xu et al. 2005 [191] used a continuous wave system with a single-wavelength (808nm) laser diode. Their detector head included
9 sources and 4 detectors forming 16 source detector pairs with a distance of 2.5cm.
These studies are some of many brain functional studies done on infants and adults. The majority of these systems were CW even though it does not have the best resolution and some studies used APD while others use PMTs. Also, there were variations among the wavelengths of the light sources used and the number of both the source and detectors. As for optical tomography, the first tomographic images of the neonatal were recorded by Benaron et al. 2000 [168], by reconstructing measurements of mean photon flight time made across neonatal head. Thirty-four pairs of sources
34 and detectors were attached to the head using a headband [165] and were used to obtain 2D tomographic slice images of anatomical disorders [166,168] and functional activation [168]. From these optical tomography studies it is apparent that NIR 3D imaging is still in its early stage and it may some time before any good 3D resolution images can be achieved.
3.4 Discussion
FMRI and NIR are complimentary to each other in that they allow for the mon- itoring and measurement of oxygen. While FMRI is well developed, NIR is still lagging behind with respect to imaging the visual cortex and is not ready for the clin- ical environment. NIR measurement accuracy can vary depending on the design of the detector head. Constant optode distance is crucial; if head circumference changes even by a fraction of a millimeter, the trends are significantly biased [192]. Strang- man et al. 2003 [177] did a comprehensive study on factors affecting the accuracy of
NIR concentration calculations and found that the wavelength selection and optode placement to be important factors in reducing error. A limitation of NIR is its low spatial resolution [156]. Statistical parametric mapping, as in FMRI, cannot be di- rectly applied to NIRS studies because there are too few measurement portions [180].
On the other hand compared to other imaging methods, optical approaches have an excellent temporal resolution [162,195] that enables analysis in the frequency domain.
NIR is also low cost, non-invasive, and portable.
The use of functional MRI has proved to be a successful imaging modality but unfortunately, like many other modalities, it has its drawbacks. The main being cost and portability. Therefore, the development of novel, more selective, and noninvasive
35 diagnostic techniques is a priority. The inherent potential of this optical technique is that it is noninvasive and can be used during behavioral tasks in order to provide temporal and spatial information, making it ideal for infant research [190]. However, this technique is limited by its low spatial resolution [156]. With all these in mind, it is clear that there is a significant clinical need for low cost, non-invasive, portable imaging modalities that combines both high spatial resolution and high physiological sensitivity to improve the clinical sensitivity and specificity for visual cortex (and long term motor cortex) diagnosis.
FNIRS has several advantages in comparison with other imaging methods, such as high flexibility, portability, low cost and biochemical specificity. Moreover, patients and children who might not stand the confined environment of functional magnetic resonance imaging (FMRI) experiments can be repetitively examined. Therefore, it is useful to further establish FNIRS as a method for functional imaging [172]. NIRS may be useful in detecting visual dysfunction objectively and noninvasively in patients with visual disturbance, especially when used at the bedside [189]. Specifically, Miki et al.
2005 [189] demonstrated that a decreased activation of the visual cortex in patients with optic neuritis can be seen with NIRS. As for the field of recording eye function via the cortex, this again does not seem to have been recognized much yet. The use of FNIR and FMRI concurrently may lead to new findings in the area of functional imaging. Studies combining FMRI and FNIR are now emerging [57–60, 62, 187].
Appendix A lists groups that have and are currently trying to develop both FMRI and NIRS for visual and oculomotor studies.
We will now report the work that we have contributed in regards to these fields.
36 CHAPTER 4
FMRI ON LOOK VS STARE OPTOKINETIC NYSTAGMUS (OKN)
4.1 Abstract
The purpose of this study is to identify the anatomical correlates of voluntary look
OKN compared to involuntary stare OKN based on FMRI. FMRI was undertaken utilizing a 3.0T GE system and the BOLD technique. Data analysis was carried out using FEAT (FMRI Expert Analysis Tool) Version 5.4. Look OKN and stare OKN were generated under identical stimulus ON conditions (vertical sine wave grating of
1.14 c/deg drifting right to left with binocular viewing). Subjects included 6 normal adults ranging in age from 18-54 years with normal visual acuity (20/20 or better) and normal stereoacuity (40 sec of arc or better). Look OKN generated significantly more cortical FMRI activation than did stare OKN. These preliminary results suggest that look OKN involves more brain sites than stare OKN. The anatomical correlates of look vs stare OKN are discussed.
4.2 Introduction
There are two main types of Optokinetic Nystagmus (OKN); Look OKN charac- terized by a series of voluntary pursuit and saccadic large amplitude and low frequency
37 eye movements and Stare OKN characterized by a series of involuntary pursuit and
saccadic low amplitude and high frequency eye movements [196]. Look OKN is charac-
terized by large amplitude (e.g., > 5o va) and low frequency (e.g., < 1Hz) alternating
voluntary pursuit and saccadic eye movements. Stare OKN is characterized by low
amplitude (e.g., < 5o va) and high frequency (e.g., 1 − 5Hz) involuntary alternating
eye movements. The typical frequency of involuntary stare OKN is about 3Hz; higher
than normal subjects can voluntarily move their eyes in terms of saccadic or pursuit
movements (about 2.25Hz) [197].
In previous FMRI studies of OKN, a detailed review of OKN may not answer the question of which type of OKN was utilized, particularly if not all subject OKN waveforms or details of the OKN characteristics were published. An issue of concern in
FMRI studies of OKN is the lack of specification of whether voluntary (look OKN) or involuntary (stare OKN) OKN was assessed [65–67,70,71]. [67] used a rotating drum that contained colored figures, thus a stimulus more inclined to evoke voluntary, look
OKN. They also reported that the OKN amplitude ranged from 2 − 13o visual angle,
suggesting voluntary OKN or a mixture of the two types of OKN. [71] instructed
their subjects to look passively at the screen thus favoring involuntary stare OKN.
However, Konen et al. 2005 [69] suggested that looking passively at the stimulus may
not maintain purely involuntary or voluntary OKN. As a consequence, the literature is
ambiguous regarding the type of OKN evoked in FMRI studies. In the present study,
the brain responses based on FMRI to both voluntary look OKN and involuntary
stare OKN in normal subjects was investigated. Our working hypothesis was that
the two forms of OKN would yield different anatomical correlates and, further, that
voluntary OKN would activate more cortical sites because of its volutional nature.
38 4.3 Materials and Methods
4.3.1 Subjects
Subjects included 6 normal adults ranging in age from 18-54 years with normal visual acuity (20/20 or better) and normal stereoacuity (40 sec of arc or better).
The study was approved by the IRB at Children’s Hospital, Columbus, OH. All subjects were given a written explanation of the study and signed a consent form. For each subject LogMAR visual acuity (ETDRS), contrast sensitivity function (VCTS
6500) and stereoacuity (Randot) were measured and assessed as previously described utilizing alternative forced-choice procedures [198]. Details of these measurements along with subject demographics are provided (Table 4.1).
Group 1 LogMAR
Age Visual Acuity Stereoacuity
Subject Years Gender Race RE LE min of arc. CSF
S1 34.35 FB -0.12 -0.02 20 N1
S2 20.61 MW -0.16 -0.18 20 N1
S3 37.71 FW -0.18 0.06 40 N1
S4 17.95 FW -0.16 -0.02 40 N1
S5 24.20 MW -0.18 -0.16 25 N1
S6 54.41 MW -0.12 -0.12 20 N1
µ 31.54 -0.15 -0.07 27.50
σ 13.62 0.03 0.09 9.87
Table 4.1: Visual acuity and demographics of subjects that participated in the study.
4.3.2 Scanning sequences
FMRI was performed with a 3.0T GE Medical Systems Signa Excite and the
BOLD sensitive T2* weighted echo-planar (EPI) sequence [199] using an 8 channel
39 array head RF coil. Scanning protocol included a screening brain MR scan, including
sagittal T1-weighted and axial T2-weighted scans, to exclude any anatomic brain ab-
normality (e.g., Arnold - Chiari malformation) [200]. The locations of activation and
deactivation clusters were defined with respect to the Montreal Neurological Institute
(MNI) coordinates and anatomical landmarks [201–203]. These sites were identified
with the Talairach and Tournoux atlas [204]. The FMRI acquisition parameters were:
TE = 35ms; TR = 1.5s; flipangle = 90o single shot; full k-space; 64 64 acquisi-
tion matrix with a field of view (FOV) = 24cm, generating an in-plane resolution of
3.75mm2 with a max total of 23 axial slices. A total of 120 volumes (time points)
were acquired. For anatomic imaging, we used a three-dimensional volume spoiled
gradient-echo pulse sequence (0.469mm2) in the axial plane and obtained 1.3mm thick slices. Example values of the parameters were: TR = 5.852ms; TE = 2.1ms; matrix = 352x224; flipangle = 45o. An additional Axial Fast Relaxation Fast
Spin Echo sequence (FRFSE) T2 weighted anatomic MR image (0.234mm2) was ac-
quired. For the T2 weighted anatomic MR images, the image sets slice thickness
was the same as their respective functional EPI scans. Example values: TR = 3.4s;
TE = 99.268ms; matrix = 320x256; flipangle = 90o.
4.3.3 Optokinetic Nystagmus Stimulation Protocol
Visual stimuli were programmed in Matlab (MathWorks, Natick, MA) and pre- sented on a Toshiba laptop. Visual stimuli were delivered within the bore of the
MRI scanner using an LCD projector (Sony VPL-X1000) fitted with a custom lens
(Buhl Optical, Pittsburgh, PA). Stimuli were projected onto a rear-projection screen mounted on the head coil and visible to the subject through a tilted mirror placed
40 above the subject’s head [126]. The screen was placed 20 − 23cm from subjects’ eyes, depending on head size. The fields of view were 30o visual angle in the horizontal and 11o visual angle in the vertical direction. Head motion was restricted by firm cushions packed around the head and by use of a head strap. Visual stimuli followed a standard block design (Figure 4.1) [72]. During the ON stimulation condition, a vertical sine wave grating of 1.14 c/deg drifted right to left (10Hz) at 37% contrast and duration of 30s. During the OFF condition, the subject was presented a station- ary grating of 2.53 c/deg. The difference in grating spatial frequencies between OFF and ON conditions was designed to minimize grating adaptation effects. The grating in the OFF condition was stationary, to minimize pattern specific activation. There were two ON conditions, look and stare. The ON-OFF cycle was repeated three times for each subject and both ON conditions.
Figure 4.1: Experimental task design. The contrast for all gratings was 35%. There was a total of two scans (one for look OKN and one for stare OKN) for each subject and each scan consisted of three ON-OFF cycles.
41 Look OKN and stare OKN were generated under identical stimulus conditions.
The only difference between look and stare OKN conditions was the subject instruc-
tions. Subjects were either instructed to stare at the drifting grating stripes but not
to follow them (involuntary OKN) or to actively pursue the drifting grating stripes
from the center of the display to the edge of the display (about 15o) and back again in
repeat fashion (voluntary OKN). These instructions were given so that the subjects
did not simply look passively [71] at the screen which may elicit a mixture of look
and stare nystagmus [69]. An eye movement practice session (look versus stare OKN)
was performed prior to the scanning session so that subjects understood directions
prior to FMRI data collection. Also, subjects were verbally instructed in the MRI
room on what to do for the next set of stimuli (look or stare) between the scans.
4.3.4 Data Analysis
Multi-stage data analysis was carried out using FEAT (FMRI Expert Analysis
Tool) Version 5.4 (FMRIB Software Library, www.FMRIb.ox.ac.uk/fsl). This multi-
stage process involved analyzing individual level statistics followed by higher/group
level analysis. Motion correction using MCFLIRT [205], mean-based intensity normal-
ization of all volumes by the same factor, and highpass temporal filtering (Gaussian-
weighted LSF straight line fitting, with σ = 30s) were applied as pre-statistics pro- cessing. No spatial smoothing of the data was performed. Post processing motion correction included mean voxel displacements: absolute (each time point with re- spect to the reference image = 0.14mm) and relative (each time point with respect to the previous image = 0.04mm). Before further analysis, these relative displace- ments were compared across the subjects to assess image motion. For this study,
42 the subjects for higher level analysis had an average and maximum relative motion displacement of 0.03mm and 0.05mm, respectively. Independent Component Anal- ysis (ICA)-based exploratory data analysis was carried out using MELODIC [206], in order to investigate the possible presence of activation or unexpected artifacts.
Time-series statistical analysis was carried out using FILM with local autocorrelation correction [207]. Z (Gaussianized T/F) statistic images were thresholded using clus- ters determined by an average Z > 5.1 and a (corrected) cluster significance threshold of P < 0.01 [208–210]. Registration to high resolution and/or standard images was carried out using FLIRT [205,211].
The first level of statistical analysis, individual subject and session level, was car- ried out using a general linear modeling (GLM) approach [212]. The input stimulus function was convolved with a gaussian (σ2.8s and peak lag 5s) to yield the regres- sor for the general linear model. The purpose of this convolution was to form the hemodynamic response function (HRF) which will blur and delay the original wave- form to match the difference between the input function (i.e., stimulus waveform) and the output function (measured FMRI HRF). A single contrast on-state (e.g., OKN) versus off-state (e.g., no OKN) was formed.
Group statistics (the higher level analysis) was carried out using FLAME (FM-
RIB’s Local Analysis of Mixed Effects) [213]. A paired analysis was applied on a voxel-wise basis between the stimulus-evoked BOLD signal change in the on-state and the off-state to yield a statistical parametric map. We analyzed the BOLD [63]
MRI signal activation for the following subtraction pairs: OKN look - stationary grat- ings (off-state); OKN stare - stationary gratings (off-state); OKN look - OKN stare; and, OKN stare - OKN look. FMRI activation maps were obtained accordingly.
43 4.4 Results
4.4.1 Activation effects in individual subjects
Look OKN yielded more activation sites compared to stare OKN, although there were some overlapping sites. Tables 4.2 4.3 4.4 4.5 highlight these results at the individual subject level using a Z-score of 5.1 and a P-value of 0.01. All the subjects had activation in the Culmen during look OKN, however; only two subjects displayed activation with stare OKN. Activation in the middle occipital and precentral gyri was balanced between the two OKN conditions across the subjects with the exception of subject 3. The cingulate gyrus was activated in all subjects for look OKN. In the stare OKN, the cingulate gyrus was activated in subjects 1,2,5 and 6 but not present in subjects 3 and 4. The typical activation trend seen across the individuals was similar with consistent activation with look OKN for all subjects, however; little or no activation with stare OKN for the same subjects (Figure 4.2).
4.4.2 Group activation effects
In a group average activation across the subjects of look and stare OKN separately, look OKN generated significantly more FMRI activation than stare OKN (Table 4.2,
4.3; Figure 4.3 4.4). The mean results showed areas specific to look OKN such as the cingulate gyrus (BA 24,32), inferior frontal gyrus (BA 47), middle frontal gyrus, anterior cingulate, parahippocampal gyrus (BA 27,28,35,36), lingual gyrus (BA 18), precuneus (BA 7, 31), inferior parietal lobule, cerebellar tonsil, declive and superior temporal gyrus at a Z-score of 3.0 and a P-value of 0.01 (Table 4.6). The group mean activation of stare OKN across the six subjects yielded no significantly activated areas at a Z-score of both 3.0 and 2.3 and a P-value of 0.01.
44 Look vs. Stationary Stare vs Stationary
x y z x y Z
S1 L 0 -43 -5 -36 -48 -25
R 16 -55 -6 - - - - - S2 L -54±8 14±11 9±14 25±3 55±7 19±10 - - R 16±6 -55±8 11±1 -6±2 11±1 46±9 - - - S3 L 17±14 53±10 13±5 R - S4 L -25±6 -54±8 20±8 R 11 -49 -8
S5 L -7 -65 -5
R 5 -60 -1 - - S6 L -45±5 30±11 24±6 R 18 -31 -11
Table 4.2: Talairach coordinates of mean (meanstdev where more than one coordinate is available per correlate) location of the look- and stare- OKN related activity at the Culmen. Z-scores were thresholded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01.
Look vs. Stationary Stare vs Stationary
x y z x y z
S1 L -43 -73 1 -47 -61 -3
R 42 -67 7 - - S2 L -32 -81 -7 6±11 46±5 76±3 R 39 -71 3
S3 L -50 -82 2 - R 40±6 1±5 77±8 S4 L -38 -72 -9
R 41 -80 3 24 -88 21
S5 L -55 -65 -4 -55 -65 -4 - R 48±0 -2±8 52 -79 4 77±4 - - - - S6 L 14±0 5±8 32±3 83±4 43±9 79±9 R 46 -76 8
Table 4.3: Talairach coordinates of mean (meanstdev where more than one coordinate is available per correlate) location of the look- and stare- OKN related activity at the Middle Occiptal Gyrus. Z-scores were thresholded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01.
45 Look vs. Stationary Stare vs Stationary
x y z x y Z - S1 L -7±6 30±9 45±7 R 42±11 -3±2 38±9 42 -8 40 - - S2 L -6±5 31±6 -6±9 30±17 49±12 51±6 - - R 50±9 25±18 48±8 30±22 3±11 3±13 S3 L -47 -4 34
R 45±8 -7±8 32±16 - S4 L -53±7 32±18 -43 -4 54 3±11 R 44 -13 44
S5 L -45±8 -4±4 41±6 -47 -2 40
R 47±14 0±8 39±2 - - S6 L -4±4 34±8 -7±4 31±8 49±10 53±9 R 48±10 -2±6 29±14 14±6 2±11 22±23
Table 4.4: Talairach coordinates of mean (meanstdev where more than one coordinate is available per correlate) location of the look- and stare- OKN related activity at the Precentral Gyrus. Z-scores were thresholded using clusters determined by Z−score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01.
Look vs. Stationary Stare vs Stationary
x y z x y z - S1 L -19±8 34±1 -11 -17 32 12±5 R 4 6 33 7±6 -15±27 30±3 - S2 L -9±5 -10±17 37±8 -17±18 38±6 9±4 R 7±6 -9±23 31±4 8±4 -15±20 35±6
S3 L -9±4 12±9 33±6
R - S4 L 5±21 34±6 11±1 R 12 -23 32
S5 L
R 13±5 -11±18 36±3 9 4 32 - S6 L -9±7 -13±29 32±5 7±5 40±1 7±6 R 6±4 -18±33 32±8 5±6 20±15 33±8
Table 4.5: Talairach coordinates of mean (µ ± σ where more than one coordinate is available per correlate) location of the look- and stare- OKN related activity at the Cingulate Gyrus. Z-scores were thresholded using clusters determined by Z −score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01.
46 Figure 4.2: Activation of look (blue) and stare (red) overlayed for each subject with cingulate gyrus marked by green ellipse. In general more brain activation was found with look OKN at the cingulate gyrus with all subjects being activated; however, subjects 1,2,6 did show activation for stare at these slices. Although not evident by slice selection, subject 5, did show activation for stare OKN. Z-scores (Gaussianised T/F) statistic images were thresholded using clusters determined by Z − score > 5.1 and a (corrected) cluster significance threshold of P − value = 0.01.
47 Figure 4.3: Precuneus, cingulate and middle frontal gyrus activation for look (red) OKN after a group mean of significant areas activated across all six subjects with look using clusters determined by Z − score > 3.1 and a (corrected) cluster significance threshold of P − value = 0.01.
Figure 4.4: Culmen and parahippocampal gyrus activation for look (red) OKN after a 2 sample t-test comparison of significant areas activated with look using clusters determined by Z − score > 3.1 and a (corrected) cluster significance threshold of P − value = 0.01.
48 To further test our hypothesis that look OKN would activate more sites than stare
OKN, we performed a two sample paired t-test in order to directly compare look and stare OKN. In essence, the test takes the average activation of all subjects performing the look OKN task and subtracts the activation from the average activation of all the subjects perfoming the stare OKN task. Stare - look OKN yielded no significant brain sites activated using a Z-score of 3.0 and a P-value of 0.01. Thus, we could not find activation sites only with stare OKN at these statistical values. However, we found several areas significantly activated with look OKN and not stare OKN using a Z-score of 3.0 and P-value of 0.01 (Table 4.7). These areas included the culmen, parahippocampal, lingual, middle temporal gyri, inferior and superior parietal lobules and precuneus, all of which were unilaterally activated in the left hemisphere. The middle occipital gyrus was unilaterally activated in the right hemisphere while the cuneus was bilaterally activated.
4.5 Discussion
In the current study, we found that look OKN generated significantly more ac- tivation sites and an overall greater degree of activation than stare OKN in normal subjects. This finding supports our original hypothesis. The results of the current study indicate that it is very important to speciify the type of OKN studied in FMRI research; depending on the type of OKN different FMRI results would be expected.
These differences were seen at the individual level where several activation areas were not seen with stare OKN, as well as in the group level activation results where con- sistent activation for the group on average was seen for look OKN but not stare
OKN.
49 L/R z-score x y z
Frontal - Cingulate Gyrus (BA 32) L 6.5±0.5 13.3±1.5 35.0±1.0 12.7±0.6 R Inferior Frontal Gyrus (BA L 5.8 -38.0 23.0 -5.0 47) R 5.8 40.0 15.0 -6.0
Middle Frontal Gyrus L
R 6.8±0.2 36.3±3.4 35.8±3.1 26.0±3.7
Limbic
Anterior Cingulate L 4.6 -10.0 26.0 -8.0
R - Cingulate Gyrus (BA 24) L 5.9±0.8 12.3±2.9 33.3±3.2 14.3±1.5 R Parahippocampal Gyrus (BA - L 5.9±0.2 -29.5±1.7 -9.5±5.2 27, 28, 35, 36) 22.3±2.2 R 6.3±0.6 22.0±1.2 -30.0±1.9 -10.2±2.9
Occipital
Lingual Gyrus (BA 18) L
R 7.5 4.0 -96.0 -4.0
Precuneus L
R 9.4 17.0 -73.0 24.0
Parietal
Inferior Parietal Lobule L 5.7 -52.0 -32.0 30.0
R
Precuneus (BA 7, 31) L 6.8±0.1 -9.0±2.1 -50.0±1.3 39.2±1.3
R 7.4 18.0 -73.0 22.0
Posterior
Cerebellar Tonsil L
R 7.8 36.0 -54.0 -31.0
Declive L 7.6 -11.0 -75.0 -18.0
R Temporal
Superior Temporal Gyrus - L 6.1±1.0 -9.2±26.0 -1.0±14.1 (BA 38) 52.8±5.3 R
Table 4.6: Talairach coordinates (meanstdev where more than one coordinate is avail- able per correlate. Anatomical correlates across entire brain for mean look voluntary OKN. Z-scores were thresholded using clusters determined by Z − score > 3.0 and a (corrected) cluster significance threshold of P − value = 0.01. L/R (left or right hemispheres)
50 L/R z-score x y z
Anterior - Culmen L 4.8±0.2 -15.0±0.0 -50.7±3.5 8.3±0.6 R
Limbic
Parahippocampal Gyrus L 4.8 -26.0 -57.0 -6.0
R
Occipital
Cuneus (BA 17, 18) L 7.0±0.0 -3.0±1.2 -80.0±2.2 14.4±4.0
R 5.3±0.2 21.0±2.0 -80.8±2.2 23.3±1.3 - Lingual Gyrus (BA 19) L 5.1±0.1 -19.0±1.4 -65.0±0.0 2.5±0.7 R
Middle Occipital Gyrus L
R 4.3±0.6 30.0±1.4 -80.5±0.7 13.5±4.9
Middle Temporal Gyrus L 6.9 -39.0 -81.0 25.0
R
Parietal
Inferior Parietal Lobule L 6.1±0.2 -36.0±3.2 -57.2±1.6 44.2±0.8
R
Precuneus (BA 7) L 6.0±0.3 -22.8±2.8 -69.5±3.7 43.3±2.6
R Superior Parietal Lobule (BA L 5.9±0.4 -22.5±2.1 -69.0±0.0 45.5±0.7 7) R
Table 4.7: Talairach coordinates (meanstdev where more than one coordinate is avail- able per correlate. Anatomical correlates across entire brain for look voluntary - stare involuntary OKN using 2 sample paired t-test. Z-scores were thresholded using clus- ters determined by Z − score > 3.0 and a (corrected) cluster significance threshold of P − value = 0.01. L/R (left or right hemispheres).
51 We chose only one direction of movement (right to left), as [70] found that there was no direction dependent activation in cortical eye fields, but there was asymmetry in the paramedian visual cortex areas. Our findings did show asymmetry in the occip- ital cortex. Also, Bense et al. 2006 [70] found stronger activation in the hemisphere contralateral to slow OKN phase (pursuit). We were not able to replicate this finding.
However, because the amplitude of pursuit involved in look OKN is greater than in stare OKN, look OKN theoretically should have stronger activation than stare OKN in the contralateral hemisphere (right). This was the case in the visual cortex but not in the parietal lobe (Table 4.7). We also found more voxels activated in the contralateral hemisphere in the occipital cortex for look OKN.
Dieterich et al. 2000 [66] have shown that small field horizontal OKN as well as voluntary saccadic eye movements activate areas of both cerebellar hemispheres in- cluding the superior semilunar lobule, simple lobule, quadrangular lobule and inferior semilunar lobule. In addition, activation was found in the middle cerebellar peduncle, dentate nucleus, culmen (medially), and uvula of the cerebellar nuclei. Unfortunately, the type of OKN was not specified. [65] also found OKN to activate subcortical areas including the caudate nucleus, putamen, globus pallidus and paramedium thalamus.
We did not find FMRI activation in these latter areas, possibly due to the fact that our scanning protocol was not optimized to detect these subcortical areas.
However, with our current scanning protocol, for look OKN, we still found acti- vation in the cerebellum specifically in the cerebellar tonsil, declive and culmen - a
finding that suggests that [65] study may have included look OKN.
A direct comparison of look and stare OKN, based on the subtraction process as described in the Results section, revealed that areas of activation in the frontal,
52 posterior and posterior were not significantly different. Instead, areas specific to the limbic, occipital and parietal lobes showed significant activation only in look OKN
(Table 4.7). Again, the lack of significant differences in the cerebellum is largely due to the scanning protocol. The current protocol has an in plane resolution of 3.75 x 3.75 which is too coarse to differentiate between regions of activations when comparing the two types of OKN. This confound was confirmed in a follow up study by our group to investigate the differences of look and stare OKN. We ran three subjects at a higher in plane resolution of 1.875 x 1.875 and were able to find significant differences at the cerebellar level. We also found significant differences between look and stare OKN in the follow up study confirming our current hypothesis.
We found significant activation for look OKN, which is a series of voluntary pur- suit and voluntary saccadic eye movements, in the cingulate gyrus which has been previously been associated with voluntary smooth pursuit eye movements [89]. Also, activation was seen for look OKN, in the precuneus which is known to be activated in voluntary saccadic [73,78,214,215] and voluntary pursuit [68] eye movements.
A limitation to the present study as well as to most previous FMRI studies on oculomotor function is that an MRI compatible eye tracker was not employed to confirm the subject’s eye movements. As a consequence, we can not be sure that there was pure look or stare OKN throughout the “ON” condition in the current study. However, if there was a large amount of mixing between the two types of OKN within the “ON” conditions we would not expect to find differences between look and stare OKN conditions. On the contrary, statistically significant differences between the two OKN “ON” conditions were found, suggesting that indeed the subjects were following the proper instructions.
53 One aim of this study was to determine if there exist differences between the look and stare OKN. The results are in general agreement with Konen et al. 2005 [69].
We found more activation sites with higher intensity in the less reflexive look OKN as opposed to the more reflexive stare OKN as Konen et al. 2005 [69] reported.
4.6 Conclusion
These preliminary results suggest that look OKN involves more brain sites than stare OKN. Therefore, future FMRI studies on OKN should specify OKN type.
54 CHAPTER 5
PARADIGM EFFECTS ON LOOK VS STARE OPTOKINETIC NYSTAGMUS (OKN): AN FMRI STUDY
5.1 Abstract
FMRI studies are dependent on the type of paradigm used. In our study we changed the visual stimulation as well as the MRI acquisition protocol in order to see if the trends in activation are consistent with our previous study. This is possible by exploiting the higher magnetic strength of 3T as opposed to all previous optokinetic nystagmus (OKN) studies which were performed at 1.5T. The secondary purpose of this study is to identify the anatomical correlates of voluntary look optokinetic nystagmus (OKN) compared to involuntary stare OKN based on FMRI. FMRI was undertaken utilizing a 3.0T GE system and the BOLD technique. Data analysis was carried out using FEAT (FMRI Expert Analysis Tool) Version 5.4. Look OKN and stare OKN were generated under identical stimulus “ON” conditions (vertical sine wave grating of 1.21 c/deg drifting right to left with binocular viewing). The only difference between look and stare OKN conditions was the subject instructions. To minimize stimulus pattern and movement related FMRI activation, “OFF” condition included a stationary grating (pattern only activation). Subjects included 3 normal
55 adults ranging in age from 36-54 years with normal visual acuity (20/20 or better)
and normal stereoacuity (40 sec of arc or better). Look OKN generated significantly
more FMRI activation than did stare OKN. These preliminary results suggest that
look OKN involves more brain sites than stare OKN. These results are consistent
with our previous study. The anatomical correlates of look vs stare OKN will be
discussed.
5.2 Introduction
An issue of concern in FMRI studies of OKN is the lack of specification of whether
voluntary OKN or involuntary OKN was assessed (see Leguire et al. 1991 [196] for
discussion of the different types of OKN) as well as paradigm differences. Voluntary
OKN is characterized by large amplitude (e.g., > 50va) and low frequency (e.g.,
< 1Hz) alternating voluntary pursuit and saccadic eye movements. Involuntary OKN is characterized by low amplitude (e.g., < 5ova) and high frequency (e.g., 1 − 5Hz)
involuntary alternating eye movements. [Note: The typical frequency of involuntary
stare, OKN is about 3Hz; higher than a normal subject can voluntarily move their
eyes in terms of saccadic or pursuit movements (about 2.25Hz)] [197]. A detailed
review of OKN literature may not answer the question of which type of OKN was
utilized in FMRI studies, particularly if not all subject OKN waveforms or details
of the OKN characteristics were published. However, explicit differentiation of the
slow versus fast phase of OKN has been made, specifically in the subdivisions of the
slow phase [67, 69], by eliciting the direct component [216]. But the same for look
versus stare has not. Dieterich et al. 2003 [67] used a rotating drum that contained
“colored figures”, thus a stimulus more inclined to evoke voluntary, “look” OKN. They
56 also reported that the OKN amplitude ranged from 2 − 13o visual angle, suggesting against purely involuntary OKN. More recently, Konen et al. 2005 [69] suggested that looking passively at the stimulus may not maintain purely involuntary OKN.
As a consequence, the literature is ambiguous regarding the type of OKN evoked in
FMRI studies. Konen et al. 2005 [69], also reported that the more reflexive the eye movements the weaker the cerebral activity. In the present paper, the response of both voluntary and involuntary OKN in normal subjects was studied. Our hypothesis was that the two forms of OKN would yield different anatomical correlates, and that voluntary OKN would activate more brain sites because of its less reflexive nature.
We chose only one direction (right to left), as Bense et al. 2006 [71], found that there was no direction dependent activation in cortical eye fields, but there was asymmetry in the paramedian visual cortex areas. Also they found stronger activation in the hemisphere contralateral to slow OKN phase (pursuit). Because the amplitude of pursuit involved in look is greater than in stare, look thus should have stronger activation than stare in the contralateral hemisphere.
5.3 Materials and methods
5.3.1 Subjects
Subjects included three normal adults ranging in age from 36-54 years with normal visual acuity (20/20 or better) and normal stereoacuity (40 sec of arc or better), Tables
5.1 5.2. All subjects were given a written explanation of the study and signed a consent form. For each subject LogMAR visual acuity (ETDRS), contrast sensitivity function (VCTS 6500) and stereoacuity (Randot) were measured and assessed as previously described [198], utilizing alternative forced-choice procedures.
57 5.3.2 Scanning sequences
FMRI was performed with a 3.0T GE Medical Systems Signa Excite and the
BOLD sensitive T2* weighted echo-planar (EPI) sequence [199] using an 8 channel array head RF coil. Scanning protocol included a screening brain MR scan, includ- ing sagital T1-weighted and axial T2-weighted scans, to exclude any anatomic brain abnormality (e.g., Arnold - Chiari malformation) [200]. The locations of activa- tion and deactivation clusters were defined with respect to MNI coordinates and anatomical landmarks [201–203]. These sites were identified with the Talairach and Tournoux [204] atlas. The FMRI acquisition parameters were: TE = 35ms;
TR = 3.0s; flipangle = 90o single shot; full k-space; 96 96 acquisition matrix with a field of view (FOV) = 24cm, generating an in-plane resolution of 1.875mm2 with a max total of 27 axial slices. A total of 120 volumes (time points) were acquired. For anatomic imaging, we used a three-dimensional volume spoiled gradient-echo pulse sequence (0.469mm2) in the axial plane and obtain 1.3mm thick slices. Example values of the parameters were: TR = 5.852ms; TE = 2.1ms; matrix = 352x224; flipangle = 45o. An additional Axial Fast Relaxation Fast Spin Echo sequence
(FRFSE) T2 weighted anatomic MR image (0.234mm2) was acquired. For the T2 weighted anatomic MR images, the image sets slice thickness was the same as their respective functional EPI scans. Example values: TR = 3.4s; TE = 99.268ms; matrix = 320x256; flipangle = 90o.
5.3.3 Optokinetic Stimulation Protocol
Visual stimuli were programmed in the Python programming language and pre- sented on a Toshiba laptop. Visual stimuli were delivered within the bore of the MRI
58 scanner using an LCD projector (Sony VPL-X1000) fitted with a custom lens (Buhl
Optical, Pittsburgh, PA). This projected onto a rear-projection screen mounted on
the head coil and visible to the subject through a tilted mirror placed above the sub-
ject’s head [126]. The screen was placed 20 − 23cm from subjects’ eyes. The field of
views were 48o in the horizontal and 15o in the vertical direction. Head motion was restricted by firm cushions packed around the head and by use of a head strap. Visual stimuli followed a standard block design (Figure 5.1) [72]. During the stimulation condition, a vertical sine wave grating of 1.21 c/deg drifted right to left (7Hz) at 35% contrast and duration of 20s. During the resting condition, the subject was told to look at the stationary grating with 1.21 c/deg.
Look OKN and stare OKN were generated under identical stimulus conditions.
The only difference between look and stare OKN conditions was the subject instruc- tions. Subjects were ether instructed to stare at the drifting grating stripes but not to follow them (involuntary OKN) or to actively pursue the drifting grating stripes from the center of the display to the edge of the display (about 100 out and back again in repeat fashion, voluntary OKN). These instructions were given so that the subjects do not simply look passively [71] at the screen which may elicit a mixture of look and stare nystagmus [69]. Thus we wanted to activate either look or stare
OKN separately. An eye movement practice session (look versus stare OKN) was per- formed so that subjects understood directions prior to FMRI data collection. Also, we had someone in the MRI room instructing the subjects on what to do for the next set of stimuli (look or stare) between the scans. This extra step was taken because eye movements were not recorded during FMRI due to the lack of MRI compatible
59 measuring equipment, which makes it harder to verify exactly what the subjects were doing.
5.3.4 Data Analysis
Multi-stage data analysis was carried out using FEAT (FMRI Expert Analysis
Tool) Version 5.4 (FMRIB Software Library, www.FMRIb.ox.ac.uk/fsl). This multi- stage process involved analyzing individual level statistics followed by higher/group level analysis. Slice-timing correction using Fourier-space time-series phase-shifting, motion correction using MCFLIRT [205], spatial smoothing using a Gaussian ker- nel of FWHM 5mm, mean-based intensity normalization of all volumes, and high- pass temporal filtering (Gaussian-weighted LSF straight line fitting, with σ = 30s) were applied as pre-statistics processing. Post processing motion correction included mean voxel displacements: absolute (each time point with respect to the reference image = 0.14mm) and relative (each time point with respect to the previous im- age = 0.04mm). Before further analysis, these relative displacements were com- pared across the subjects to assess image motion. For this study, the subjects for higher level analysis had an average and maximum relative motion displacement of
0.03mm and 0.05mm, respectively. Independent Component Analysis (ICA)-based exploratory data analysis was carried out using MELODIC [206], in order to in- vestigate the possible presence of activation or unexpected artifacts. Time-series statistical analysis was carried out using FILM with local autocorrelation correc- tion [207]. Z (Gaussianized T/F) statistic images were thresholded using clusters determined by an average Z > 5.1 and a (corrected) cluster significance threshold of P < 0.05 [208–210]. Registration to high resolution and/or standard images was
60 carried out using FLIRT [205, 211]. The first level of statistical analysis, individual subject and session level, was carried out using a general linear modeling (GLM) ap- proach [212]. The input stimulus function was convolved with a gaussian (σ2.8s and peak lag 5s) to yield the regressor for the general linear model. The purpose of this convolution is to form the hemodynamic response function (HRF) which will blur and delay the original waveform to match the difference between the input function
(i.e., stimulus waveform) and the output function (measured FMRI HRF). A single contrast, on-state (e.g., OKN) versus off-state (e.g., fixation) was formed. Group statistics (the higher level analysis) was carried out using FLAME (FMRIB’s Local
Analysis of Mixed Effects) [213]. A paired analysis was applied on a voxel-wise basis between the stimulus-evoked BOLD signal change in the on-state and the off-state to yield a statistical parametric map. We analyzed the BOLD [63] MRI signal activation for the following (OKN look - stationary; OKN stare - stationary; OKN look - OKN stare; OKN stare - OKN look) and FMRI activation maps were obtained accordingly.
5.4 Results
The typical voxel activation is illustrated in Figure 5.2 Unless otherwise noted, look OKN showed more activated voxels and higher intensity activation. Look OKN showed more activation sites compared to stare OKN, with many overlapping sites.
Note, we will not discuss the results of each subject in detail except to show the typical activation trend seen across the individuals is similar (Figure 5.3). In this
figure the activation of look voluntary (red) and stare involuntary (blue) overlayed from one subject. This trend was seen across the other subjects as well. We see more cerebellar and higher brain activation with look OKN. P − value < 0.05, z cluster
61 threshold = 5.1. As a result of taking a group average (P − value < 0.05) of look and stare OKN separately, look OKN generated significantly more FMRI activation than stare OKN (Table 5.3). Tables 5.4 5.5 5.6 5.9 5.10 5.8 5.7 are the comprehensive lists subdivided into cortical lobes (µ±σ where more than one coordinate is available per correlate) and anatomical correlates across entire brain for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. Figure 5.4 illustrates the average for look voluntary OKN (blue) and stare involuntary OKN (red). Overall more brain activation is observed for look versus stare OKN with many sites of overlap. Occipital lobe activation from look include: lingual gyrus, cuneus, middle occipital gyrus, inferior temporal gyrus, fusiform gyrus. Occipital lobe activation from stare include: lingual gyrus, cuneus, middle occipital gyrus, inferior occipital gyrus, fusiform gyrus with a P −value < 0.05, z cluster threshold = 4.0. In Figure 5.5 a two sample paired t-test is performed for look voluntary - stare involuntary OKN (blue) and stare involuntary - look voluntary
OKN (red). More brain sites are observed solely for look OKN but not for stare OKN with a P − value < 0.05, z cluster threshold = 4.0.
5.5 Discussion
These results are in agreement with Konen et al. 2005 [69]. Sites common to both look and stare OKN along with sites exclusive to either look or stare OKN are listed.
The mean results show areas specific to look OKN such as parahippocampal gyrus, postcentral gyrus, inferior semi-lunar lobule, cerebellar tonsil, pyramis, and uvula.
The latter five areas are specific to the cerebellum. The declive, also located in the cerebellum, is active during both look and stare OKN. Sites activated in both look
62 and stare included: cingulate gyrus (BA 32), cuneus (BA 19), fusiform gyrus, inferior parietal lobule, inferior temporal gyrus, insula, middle occipital gyrus (BA 19), middle temporal gyrus (BA 21), superior temporal gyrus, supramarginal gyrus (BA 40), middle frontal gyrus (BA 6,10,11), inferior frontal gyrus (BA 47), inferior occipital gyrus, superior frontal gyrus (BA 6), lingual gyrus (BA 18), precentral gyrus (BA
6), precuneus, declive. Whereas sites activated only with look were: culmen, tuber, pyramis, fastigium, lentiform nucleus (putamen), thalamus, parahippocampal gyrus
(hippocampus) , postcentral gyrus (BA 40), inferior semi-lunar lobule, cerebellar tonsil, uvula.
As far as repeatability, there were common sites between this study and our pre- vious one ( [72], for review). Areas common in look and stare for both of the studies included: middle frontal gyrus (BA 6,10), inferior frontal gyrus (BA 47), inferior oc- cipital gyrus, superior frontal gyrus, lingual gyrus (BA 18), precentral gyrus (BA 6), precuneus, declive. And sites only for look were: postcentral gyrus (BA 40), inferior semi-lunar lobule, cerebellar tonsil, uvula. In the frontal lobe, the inferior frontal gyrus activated more in the right hemisphere in both look and stare. The middle frontal gyrus was active in both look and stare with right hemisphere dominance in look and left hemisphere dominance in stare. The precentral gyrus activated more in the left hemisphere in both look and stare in agreement with recent work [72].
The superior and frontal gyrus activated unilaterally (right hemisphere) only in look while the medial frontal gyrus activated unilaterally (right hemisphere) only in stare.
We observed cingulate gyrus activation in stare OKN contrary to our previous find- ings [72] in which the cingulate gyrus activated in look with dominance in the left hemisphere.
63 In the limbic lobe, the parahippocampal gyrus activated for look with right hemi- spheric dominance. In the occipital lobe, we saw a repeat activation of the cuneus unilaterally (right hemisphere) only in look. As well as with the lingual gyrus for both look and stare, but this time with left hemisphere dominance for both and more voxel activation for stare. The middle occipital activated in look and stare with right hemisphere dominance in look. In the parietal lobe, the precuneus activated uni- laterally in both look (left hemisphere) and stare (right hemisphere). The inferior parietal lobule activated unilaterally (left hemisphere) for look. The superior parietal lobule activated unilaterally (right hemisphere) for stare. The supramarginal gyrus activated in look and stare unilaterally (left hemisphere) with more voxels activated in stare. In look the postcentral gyrus activated unilaterally (right hemisphere).
To further confirm our hypothesis we performed a 2 sample paired t-test in order to do a direct comparison of voluntary look and involuntary stare OKN. In essence, the test subtracts activation of one stimulus from another. Note, stare - look OKN results yielded no significant brain sites meaning that all the areas activated during stare OKN are also activated during look OKN. We did however see common areas for look OKN in the parietal lobe such as the superior parietal lobule, precuneus and supramarginal gyrus, with the addition of inferior parietal lobule and postcentral gyrus solely from this study as opposed to our previous findings, as a result of the optimized paradigm. Look OKN significantly activated the declive, inferior semi- lunar lobule and the pyramis whereas stare did not. The trend in results is consistent with our previous study of look vs stare OKN. The overlap between sites for look and stare OKN can be explained by a possible mixture of look and stare as noted by [69]. It was also shown [66] that small field horizontal OKN as well as voluntary
64 saccadic eye movements activate areas of both cerebellar hemispheres including the superior semilunar lobule, simple lobule, quadrangular lobule and inferior semilunar lobule. In addition, activation was found in the middle cerebellar peduncle, dentate nucleus, culmen (medially), and uvula of the cerebellar nuclei. Fixation during OKN suppressed activation in the uvula and culmen. Dieterich et al. 1998 [65] also found
OKN to activate subcortical areas including the caudate nucleus, putamen, globus pallidus and paramedium thalamus. Fixation increased activity in the frontal eye field
(FEF) and anterior cingulate gyrus. We saw activation in the cerebellum specifically in the inferior semilunar lobule [66,71].
In FMRI analysis, many controlled and uncontrolled variables may affect results.
Two examples are subject head motion and subject attention. While these confounds can be minimized during the experiment, variation in experimental setup, post pro- cessing techniques, statistical significance and threshold levels can also affect results.
One example variation in the presented study is the scanning paradigm. Our previous study did not show as many activated voxels as this study, mainly due to the higher resolution images acquired in the current protocol. At the same time, the number of subjects included in each group average may work to clarify or obscure activated areas depending on how well the subject followed the paradigm.
5.6 Conclusion
These preliminary results confirm our previous suggestion that look OKN involves more brain sites than stare OKN. There were minor differences as a result of the visual paradigm; however the increased resolution yielded more areas of activation.
As suggested before, future FMRI studies on oculomotor systems i.e. OKN, should
65 specify OKN type as well as increase the resolution for more efficient localization of
anatomical correlates.
Figure 5.1: The Experimental task design. The counter phase gratings was intro- duced. The contrast for all gratings was 35%. The temporal frequency was 7Hz. The counter phase frequency and moving grating frequency were the same.
66 Figure 5.2: The visual paradigm and response from one voxel in the brain of one subject showing high correlation between block design and physiological response. All the subjects showed a similar correlation pattern. P − value < 0.05, z cluster threshold = 5.1.
67 Figure 5.3: The activation of look voluntary (red) and stare involuntary (blue) over- layed from one subject. This trend was seen across the other subjects as well. We see more cerebellar and higher brain activation with look OKN. P − value < 0.05, z cluster threshold = 5.1.
68 Figure 5.4: The average for look voluntary OKN (blue) and stare involuntary OKN (red). Overall more brain activation is observed for look versus stare OKN with many sites of overlap. Occipital lobe activation from look include: lingual gyrus, cuneus, middle occipital gyrus, inferior temporal gyrus, fusiform gyrus. Occipital lobe activation from stare include: lingual gyrus, cuneus, middle occipital gyrus, inferior occipital gyrus, fusiform gyrus. p − value < 0.05, z cluster threshold = 4.0.
Figure 5.5: The 2 sample paired t-test for look voluntary - stare involuntary OKN (blue) and stare involuntary - look voluntary OKN (red). More brain sites are ob- served solely for look OKN but not for stare OKN. P − value < 0.05, z cluster threshold = 4.0.
69 Visual Age Stereoacuity Acuity Subject Years Gender Race RE LE Min of arc.
A 54.6 MW -0.12 -0.12 20
B 38.4 MW -0.14 -0.14 20 70 C 36.26 FA 0 -0.12 20
µ = 43.1 -0.1 -0.1 20
σ = 10 0.1 0 0
Table 5.1: The visual, stereo acuity and demographics of subjects that participated in the study. CSF CSF
RE LE
Subject 1.5 3 6 12 18 1.5 3 6 12 18 WMCS WMCS
A 1.845 2.23 2.097 1.944 1.176 1.845 2.23 2.097 1.944 0.845 13.31 12.12
B 1.544 2.23 2.097 1.944 1 2.23 2.342 2.097 2.097 1.415 3.43 3.8 71 C 1.544 2.23 2.415 1.944 1.176 2.079 2.342 2.415 2.23 1.954 3.48 4.53
µ = 1.6 2.2 2.2 1.9 1.1 2.1 2.3 2.2 2.1 1.4 6.7 6.8
σ = 0.2 0 0.2 0 0.1 0.2 0.1 0.2 0.1 0.6 5.7 4.6
Table 5.2: The contrast sensitivity functions of subjects that participated in the study. OKN Anatomical Correlates Cingulate Gyrus, Cuneus, Fusiform Gyrus, Inferior Parietal Lobule, Inferior Temporal Look, Stare Gyrus, Insula, Middle Occipital Gyrus, Middle Temporal Gyrus, Superior Temporal Common Gyrus, Supramarginal Gyrus, Middle Frontal Gyrus, Inferior Frontal Gyrus, Inferior sites Occipital Gyrus, Superior Frontal Gyrus, Sub-Gyral, Lingual Gyrus, Precentral Gyrus, Precuneus, Declive.
72 Culmen, Tuber, Pyramis, Fastigium, Lentiform Nucleus, Thalamus, Parahippocampal Look only Gyrus, Postcentral Gyrus, Inferior Semi-Lunar Lobule, Cerebellar Tonsil, Uvula, sites Extra-Nuclear. Stare only Superior Parietal Lobule. sites
Table 5.3: The group mean anatomical sites exclusively seen for either look or stare as well as common to both with P − value < 0.05. Occipital Lobe R/L x y z z-score
Cuneus
R 10.67±7.57 -82±2.0 20±15.1 6.02±0.61
L -17±1.67 -84±2.83 25.33±5.01 4.97±0.31
Lingual Gyrus
R 14±11.31 -74±5.66 1±4.24 6.17±0.05
L -17±4.24 -61±4.24 -3±4.24 5.44±0.42
Middle Occipital Gyrus - - L 2±8.85 5.02±0.61 47.33±7.55 74.67±5.47 Inferior Temporal Gyrus
L -51±9.9 -70±2.83 2±0.0 4.75±0.04
Table 5.4: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the occipital lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
73 Frontal Lobe R/L x y z z-score
Inferior Frontal Gyrus
R 52±5.16 13±12.85 21±14.91 6.57±0.64
L -53±4.24 19±15.56 9±32.53 5.28±0.67
Medial Frontal Gyrus
R 8.5±4.43 -3.5±3.0 57±2.0 5.52±0.15
L -2 6 46 5.59
Middle Frontal Gyrus
R 36±9.93 14.5±24.89 40±16.89 5.34±0.88
L -24.4±6.07 5.6±28.26 43.6±31.6 5.04±0.19
Precentral Gyrus
R 40.5±6.61 -7.5±1.0 49.5±4.73 5.54±0.11
L -36 -10 46 4.48
Sub-Gyral
R 23.33±1.15 0.67±5.77 48.67±6.11 5.74±1.46
L -18 -2 54 4.18
Superior Frontal Gyrus
R 30±8.49 21±18.38 42±16.97 4.73±0.49
Table 5.5: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the frontal lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
74 Parietal Lobe R/L x y z z-score
Precuneus
R 13±5.29 -69±12.49 52±10.71 4.97±0.23 - - L 51.33±10.07 5.74±0.42 14.67±11.55 62.67±17.01 Inferior Parietal Lobule
R 41±1.41 -53±1.41 46±2.83 4.98±0.30
L -44±22.63 -39±12.73 46±11.31 5.33±1.12
Sub-Gyral
L -28 -48 42 5.27
Superior Parietal Lobule
R 31±9.9 -58±5.66 52±0.00 5.70±0.08
L -27.33±4.62 -62.67±12.7 49.33±7.57 5.49±0.42
Postcentral Gyrus
R 48 -20 46 5.78 - L -39.33±9.87 57.33±15.01 5.04±0.79 28.67±23.86 Supramarginal Gyrus
R 40 -44 34 4.47
Inter-Hemispheric
0 -4 24 4.7
Table 5.6: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the parietal lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
75 Temporal Lobe R/L x y z z-score
Inferior Temporal Gyrus
R 62 -50 -14 5.23
L -46±8.49 -39±49.5 -19±24.04 4.55±0.05
Superior Temporal Gyrus
R 57±12.73 -12±19.8 -1±7.07 4.90±0.22
L -34 8 -40 4.47
Middle Temporal Gyrus
L -59±1.41 -55±18.38 3±9.9 5.06±0.33
Fusiform Gyrus
L -54 -64 -14 4.15
Table 5.7: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the temporal lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
Limbic Lobe R/L x y z z-score
Parahippocampal Gyrus
L -26 -54 -4 5.2
Uncus
L -17±18.38 -23±26.87 0±53.74 4.66±0.30
Cingulate Gyrus
R 6 4 46 6.01
Table 5.8: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the limbic lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
76 Posterior Lobe R/L x y z z-score
Declive
L* -20±5.66 -64±8.49 -15±4.24 5.24±0.24
Inferior Semi-Lunar Lobule
R* 20 -78 -38 4.81
L* -30±2.83 -71±1.41 -39±1.41 4.83±0.40
Pyramis
L* -26 -72 -32 4.74
Table 5.9: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the posterior lobe for look voluntary - stare involuntary OKN using 2 sample paired t-test with P − value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
Anterior Lobe R/L x y z z-score
Culmen
R* 16 -64 -8 6.49
L* -14±0.00 -54±5.66 -9±1.41 5.79±0.10
Table 5.10: The comprehensive list of the talairach coordinates (meanstdev where more than one coordinate is available per correlate) and anatomical correlates across the anterior lobe for look voluntary - stare involuntary OKN using 2 sample paired t- test with P −value < 0.05, z cluster threshold = 4.0. R/L (right or left hemispheres). *Cerebellum activation.
77 CHAPTER 6
FUNCTIONAL MAGNETIC RESONANCE IMAGING (FMRI) OF VISUALLY GUIDED SACCADES AND SMOOTH PURSUIT AT 3T
6.1 Abstract
The purpose of this study is to identify the anatomical correlates of saccadic and
pursuit voluntary predictive eye movements based on FMRI at 3T. Long term goal
is to assess abnormal oculomotor function in patient populations by utilizing whole
brain activation imaging as opposed to localized mapping. The identification of brain
sites involved in oculomotor control are still being established. Specifically, relat-
ing anatomical correlates of saccade and pursuit to optokinetic nystagmus (OKN).
Normal oculomotor eye movement (EM) sites allow comparison with abnormal EM.
FMRI was undertaken utilizing a 3.0T GE system scanning the whole head and the
BOLD technique. Data analysis was carried out using FEAT (FMRI Expert Analysis
Tool) Version 5.4 (FMRIB’s Software Library, www.fmrib.ox.ac.uk/fsl). Saccadic and
pursuit eye movements were elicited using a white dot (0.5cm in diameter and average
visual dot size of 0.66o ) moving horizontally (19o from center in each direction) at
0.5Hz with binocular viewing. Subjects included 7 normal adults ranging in age from
18-54 years with normal visual acuity (20/20 or better) and normal stereoacuity (40
78 sec of arc or better). The average activation across the seven subjects with P < 0.05 showed more activation sites with saccadic eye movements (χ2 = 12.37, P < 0.001).
The culmen in the cerebellum was activated with both saccades and pursuit; how- ever not in the same location. Activation was found in the tuber and uvula of the cerebellum only with saccades.
6.2 Introduction
Saccade and smooth pursuit eye movements (SPEM) have been studied exten- sively with FMRI. Specifically, groups have looked at saccades [78, 80, 81, 84, 85], pro-antisaccades [82, 86], SPEM [88, 89] as well as a combination of saccades and
SPEM [68, 77, 83]. However all except two of these studies have been limited to a
1.5 Tesla magnet. In the case where a 3T magnet was used [83] only six slices of the brain were acquired with a gap of 1mm between slices. The other study [86] used 4T, however they looked at the relation between pro and anti saccades. In our study we used a 3T magnet to run functional scans for both saccade and pursuit eye movements and acquired virtually a full brain scan with no gaps in between.
6.3 Materials and Methods
6.3.1 Subjects
Subjects included 7 normal adults ranging in age from 18-54 years with normal visual acuity (20/20 or better) and normal stereoacuity (40 sec of arc or better).
The study was approved by the IRB at Children’s Hospital, Columbus, OH. All subjects were given a written explanation of the study and signed a consent form. For each subject LogMAR visual acuity (ETDRS), contrast sensitivity function (VCTS
79 6500) and stereoacuity (Randot) were measured and assessed as previously described
utilizing alternative forced-choice procedures [198].
6.3.2 Scanning sequences
FMRI was performed with a 3.0T GE Medical Systems Signa Excite and the
BOLD sensitive T2* weighted echo-planar (EPI) sequence [199] using an 8 channel
array head RF coil. Scanning protocol included a screening brain MR scan, including
sagittal T1-weighted and axial T2-weighted scans, to exclude any anatomic brain ab-
normality (e.g., Arnold - Chiari malformation) [200]. The locations of activation and
deactivation clusters were defined with respect to the Montreal Neurological Institute
(MNI) coordinates and anatomical landmarks [201–203]. These sites were identified
with the Talairach and Tournoux atlas [204]. The FMRI acquisition parameters were:
TE = 35ms; TR = 1.5s; flipangle = 90o single shot; full k-space; 64 64 acquisi-
tion matrix with a field of view (FOV) = 24cm, generating an in-plane resolution of
3.75mm2 with a max total of 23 axial slices. A total of 120 volumes (time points)
were acquired. For anatomic imaging, we used a three-dimensional volume spoiled
gradient-echo pulse sequence (0.469mm2) in the axial plane and obtained 1.3mm thick slices. Example values of the parameters were: TR = 5.852ms; TE = 2.1ms; matrix = 352x224; flip angle = 45o. An additional Axial Fast Relaxation Fast
Spin Echo sequence (FRFSE) T2 weighted anatomic MR image (0.234mm2) was ac-
quired. For the T2 weighted anatomic MR images, the image sets slice thickness
was the same as their respective functional EPI scans. Example values: TR = 3.4s;
TE = 99.268ms; matrix = 320x256; flipangle = 90o.
80 6.3.3 Saccade and Pursuit Stimulation Protocol
Visual stimuli were programmed in Macromedia Flash and presented on a Toshiba laptop. Visual stimuli were delivered within the bore of the MRI scanner using an
LCD projector (Sony VPL-X1000) fitted with a custom lens (Buhl Optical, Pitts- burgh, PA). Stimuli were projected onto a rear-projection screen mounted on the head coil and visible to the subject through a tilted mirror placed above the subject’s head [126]. The screen was placed 20 − 23cm from subjects’ eyes, depending on head size. The fields of view were 30o visual angle in the horizontal and 11o visual angle in the vertical direction. Head motion was restricted by firm cushions packed around the head and by use of a head strap. Visual stimuli followed a standard block design
(Figure 6.1) [72]. During the ON stimulation condition, a a white dot (0.5cm in diameter and average visual dot size of 0.66o ) moved horizontally (19o from center in each direction) at 0.5Hz on a black screen with binocular viewing. During the OFF condition, the subject was presented a the same white dot however it was centered and stationary. There were two ON conditions, saccade and pursuit. The pursuit condition was a smooth motion from right to left and the saccadic condition was a ballistic movement from right to left. The ON-OFF cycle was repeated three times for each subject and for saccade and pursuit separately.
6.3.4 Data Analysis
Multi-stage data analysis was carried out using FEAT (FMRI Expert Analysis
Tool) Version 5.4 (FMRIB Software Library, www.FMRIb.ox.ac.uk/fsl). This multi- stage process involved analyzing individual level statistics followed by higher/group
81 Figure 6.1: The experimental task design for 1 cycle. The OFF condition consisted of a white fixation dot for 30sec. The ON condition consisted of the white dot moving back and forth at 0.5Hz for 30sec. For pursuit the motion was smooth and for saccade the motion was ballistic. The cycle repeated three times for pursuit and saccade separately.
level analysis. Motion correction using MCFLIRT [205], mean-based intensity normal-
ization of all volumes by the same factor, and highpass temporal filtering (Gaussian-
weighted LSF straight line fitting, with σ = 30s) were applied as pre-statistics pro-
cessing. No spatial smoothing of the data was performed. Post processing motion
correction included mean voxel displacements: absolute (each time point with re-
spect to the reference image = 0.14mm) and relative (each time point with respect to the previous image = 0.04mm). Before further analysis, these relative displace- ments were compared across the subjects to assess image motion. For this study, the subjects for higher level analysis had an average and maximum relative motion displacement of 0.03mm and 0.05mm, respectively. Independent Component Anal-
ysis (ICA)-based exploratory data analysis was carried out using MELODIC [206],
in order to investigate the possible presence of activation or unexpected artifacts.
Time-series statistical analysis was carried out using FILM with local autocorrelation
82 correction [207]. Z (Gaussianized T/F) statistic images were thresholded using clus- ters determined by an average Z > 5.1 and a (corrected) cluster significance threshold of P < 0.01 [208–210]. Registration to high resolution and/or standard images was carried out using FLIRT [205,211].
The first level of statistical analysis, individual subject and session level, was car- ried out using a general linear modeling (GLM) approach [212]. The input stimulus function was convolved with a gaussian (σ2.8s and peak lag 5s) to yield the regressor for the general linear model. The purpose of this convolution was to form the hemo- dynamic response function (HRF) which will blur and delay the original waveform to match the difference between the input function (i.e., stimulus waveform) and the output function (measured FMRI HRF). A single contrast on-state (e.g., movement) versus off-state (e.g., no movement) was formed.
Group statistics (the higher level analysis) was carried out using FLAME (FM-
RIB’s Local Analysis of Mixed Effects) [213]. A paired analysis was applied on a voxel-wise basis between the stimulus-evoked BOLD signal change in the on-state and the off-state to yield a statistical parametric map. We analyzed the BOLD [63]
MRI signal activation for the following subtraction pairs: saccade - fixation (off- state); pursuit - fixation (off-state); saccade - pursuit; and, pursuit - saccade. FMRI activation maps were obtained accordingly.
6.4 Results
The average activation across the seven subjects with P < 0.05 showed more activation sites with saccadic eye movements (χ2 = 12.37, P < 0.001). The culmen in the cerebellum was activated with both saccades and pursuit; however not in the
83 same location. Activation was found in the tuber and uvula of the cerebellum only with saccades. Figure 6.2 illustrates the number of subjects that showed activation for saccades and pursuits and is split into frontal, parietal, occipital and cerebellar regions. In the cerebellum the declive favored the saccadic eye movement with seven subjects showing activation with saccades and only two with pursuit. The rest of the regions in the Figure 6.2 were balanced between saccade and pursuit. However, this was not the case for then entire brain. In general more activation was seen for saccades then for pursuit as seen in Figure 6.3 which illustrates the typical trend for all seven subjects for P < 0.05.
The group averages were similar to the individual level results. Figure 6.4 is the center slice from a three plane view of the group mean activation across the seven subjects for both saccade and pursuit with P < 0.05 and a minimum Z threshold of 2 and maximum of 11.5. The cerebellar activation is more for saccade than for pursuit.
A closer look at the lower at the cerebellum shows more areas of activation for saccades versus pursuit, Figure 6.5. Here the axial view of several lower brain slices of the group mean activation across the seven subjects are seen for both saccade and pursuit with P < 0.05 and a minimum Z threshold of 4 and maximum of 11.5. Overall there were seventeen areas associated solely with saccades, one solely for pursuit and twelve areas of overlap, Table 6.1.
6.5 Discussion
In our study at 3T we found that the precuneus was only activated with saccades as opposed to previous studies which associate it with both saccades and pursuit [68].
84 Figure 6.2: The chart illustrates the number of subjects that showed activation for saccades and pursuits and is split by regions of interest for P < 0.05. On the y-axis are the number of subjects that showed activation for that region. The red indicating saccades and the blue indicating pursuit.
85 Figure 6.3: The activation differences between saccade and pursuit for one subject. This trend was typical for all seven subjects with P < 0.05.
Figure 6.4: The center slice from a three plane view of the group mean activation across the seven subjects for both saccade and pursuit with P < 0.05 and a minimum Z threshold of 2 and maximum of 11.5.
86 Anatomical Correlate Saccade Pursuit Both Inferior Frontal Gyrus X Superior Frontal Gyrus X Anterior Cingulate X Cingulate Gyrus X Cuneus X Inferior Occipital Gyrus X Middle Occipital Gyrus X Middle Temporal Gyrus X Inferior Parietal Lobule X Precuneus X Supramarginal Gyrus X Insula X Lateral Ventricle X Lentiform Nucleus X Fusiform Gyrus X Superior Temporal Gyrus X Declive X Thalamus X Middle Frontal Gyrus X Precentral Gyrus X Parahippocampal Gyrus X Lingual Gyrus X Inferior Temporal Gyrus X Middle Temporal Gyrus X Culmen X Cerebellar Tonsil X Inferior Semi-Lunar Lobule X Pyramis X Tuber X Uvula X
Table 6.1: The activation sites from the group averages of saccade and pursuit eye movements for P < 0.05. Overall there were seventeen areas associated solely with saccades, one solely for pursuit and twelve areas of overlap
87 Figure 6.5: The axial view of several lower brain slices of the group mean activation across the seven subjects for both saccade and pursuit with P < 0.05 and a minimum Z threshold of 4 and maximum of 11.5.
We also report that the uvula and tuber which are areas within the cerebellar vermis are activated in both saccadic and pursuit eye movements. The only other time the uvula and tuber were reported was in a saccade study [217] where the cerebellar vermis was broken down into uvula, tonsils, tuber, and folium/declive) activation in saccades.
Thus we may be the first to report these sites for SPEM. The only area found unique to SPEM was the thalamus which has been previously associated with SPEM [89].
The declive was reported previously to activated for visually guided saccades [218]. In that study activation is reported in the superior semilunar lobule, we however report activation in the inferior semilunar lobule which was observed for both saccades and pursuit. The culmen has been seen activated for optokinetic nystagmus (OKN) [69] as well as in our OKN study [72], but we also found the culmen involved in both saccades and pursuit. Unlike many studies we report more anatomical areas of activation as
88 a result of performing whole brain scans. This approach is a more accurate then to isolate one area and ignore the rest of the brain.
6.6 Conclusion
When whole brain imaging is employed with 3T FMRI, a significantly greater num- ber of activated areas are found with saccadic as opposed to pursuit eye movements, with a number of overlapping sites responsible for both types of eye movements. The average activation across the seven subjects showed more activation sites with sac- cadic eye movements. The culmen in the cerebellum was activated with both saccades and pursuit; however the location was both posterior and superior in saccades. The declive was activated only in saccades. Finally we report uvula and tuber activation in both saccade and pursuit. Other areas activated are in agreement with previous studies.
89 CHAPTER 7
COMPARISON OF AXIAL, SAGITAL, AND CORONAL IMAGING FOR SIMPLE FINGER TAPPING EXPERIMENT: AN FMRI CASE STUDY
7.1 Abstract
Traditional FMRI studies scan exclusively in the axial plane and has thus been a standard. The purpose of this chapter is to compare activation maps of a standard control condition (finger tapping) using three different imaging planes, namely axial, sagital and coronal. The hypothesis is that all three scans should yield the same results. The significance of which will justify the standard axial scans. However, we found that the scanning plane does indeed make a difference in the results as will be discussed.
7.2 Introduction
In this chapter we introduce a new FMRI scanning protocol that uses the infor- mation of all three scanning planes in order to better localize and anatomical areas of activation. This activation can otherwise be missed if only using one scanning plane. The use of additional scanning planes will increase the time the subject is in the scanner, however this can be compensated for by reducing the number of volumes
90 acquired. The potential of this technique is great for the FMRI community. Specif- ically, when trying to localize areas of activation as well as increasing the statistical power on the individual level.
7.3 Materials and Methods
7.3.1 Subjects
In a standard FMRI control trial to localize anatomical areas; one male subject
(age 27) was imaged while performing the same task in the axial, sagital and coronal planes. The study was approved by the IRB at Children’s Hospital, Columbus, OH.
7.3.2 Scanning sequences
FMRI was performed with a 3.0T GE Medical Systems Signa Excite and the
BOLD sensitive T2* weighted echo-planar (EPI) sequence [199] using an 8 channel array head RF coil. Scanning protocol included a screening brain MR scan, includ- ing sagittal T1-weighted and axial T2-weighted scans, to exclude any anatomic brain abnormality (e.g., Arnold - Chiari malformation) [200]. The locations of activation and deactivation clusters were defined with respect to the Montreal Neurological In- stitute (MNI) coordinates and anatomical landmarks [201–203]. These sites were identified with the Talairach and Tournoux atlas [204]. The FMRI acquisition pa- rameters were: TE = 35ms; TR = 3s; flip angle = 90o single shot; full k-space;
9696 acquisition matrix with a field of view (FOV ) = 24cm, generating an in-plane resolution of 1.875mm2 and slice thickness of 5mm with a max total of 30 axial, 34 coronal, and 28 sagitall slices respectively. A total of 120 volumes (time points) were acquired. A total of scan time for each plane was 6 minutes and 15 seconds. For anatomic imaging, we used a three-dimensional volume spoiled gradient-echo pulse
91 sequence (0.469mm2) in the axial plane and obtained 1.3mm thick slices. Example
values of the parameters were: TR = 5.852ms; TE = 2.1ms; matrix = 352x224;
flipangle = 45o. Additional axial, coronal and sagitall Fast Relaxation Fast Spin
Echo sequences (FRFSE) T2 weighted anatomic MR image (0.234mm2) were ac-
quired. For the T2 weighted anatomic MR images, the image sets slice thickness
was the same as their respective functional EPI scans. Example values: TR = 3.4s;
TE = 99.268ms; matrix = 320x256; flipangle = 90o.
7.3.3 Stimulation Protocol
The subject was instructed to begin and stop finger tapping every 30 seconds.
Head motion was restricted by firm cushions packed around the head and by use
of a head strap. The protocol followed a standard block design (Figure 4.1). The
ON-OFF cycle was repeated six times for each scanning plane.
7.3.4 Data Analysis
Multi-stage data analysis was carried out using FEAT (FMRI Expert Analysis
Tool) Version 5.4 (FMRIB Software Library, www.FMRIb.ox.ac.uk/fsl). This multi-
stage process involved analyzing individual level statistics followed by higher/group
level analysis. Motion correction using MCFLIRT [205], mean-based intensity normal-
ization of all volumes by the same factor, and highpass temporal filtering (Gaussian-
weighted LSF straight line fitting, with σ = 30s) were applied as pre-statistics pro- cessing. No spatial smoothing of the data was performed. Post processing motion correction included mean voxel displacements: absolute and relative. Independent
Component Analysis (ICA)-based exploratory data analysis was carried out using
92 MELODIC [206], in order to investigate the possible presence of activation or un- expected artifacts. Time-series statistical analysis was carried out using FILM with local autocorrelation correction [207]. Z (Gaussianized T/F) statistic images were thresholded using clusters determined by an average Z > 5.1 and a (corrected) clus- ter significance threshold of P < 0.05 [208–210]. Registration to high resolution and/or standard images was carried out using FLIRT [205,211].
The first level of statistical analysis, individual subject and session level, was car- ried out using a general linear modeling (GLM) approach [212]. The input stimulus function was convolved with a gaussian (σ2.8s and peak lag 5s) to yield the regres- sor for the general linear model. The purpose of this convolution was to form the hemodynamic response function (HRF) which will blur and delay the original wave- form to match the difference between the input function (i.e., stimulus waveform) and the output function (measured FMRI HRF). A single contrast on-state (e.g., finger tapping) versus off-state (e.g., no finger tapping) was formed. A paired analysis was applied on a voxel-wise basis between the stimulus-evoked BOLD signal change in the on-state and the off-state to yield a statistical parametric map. We analyzed the BOLD [63] MRI signal activation for the following subtraction pairs: ON - OFF.
FMRI activation maps were obtained accordingly.
7.4 Results
We expected the same sites would be activated among all three planes. However, different planes yielded different activation maps. For instance, high activation was seen in the dentate nucleus in both the coronal and sagital planes but no evidence of dentate nucleus activation was found in the axial plane. Brodmann area 47 was
93 detected in the sagital scan, but not in the other two planes. Table 7.1 lists the different activation maps depending on the slice (plane) selection.
Anterior, frontal, limbic, occipital, parietal, posterior and temporal lobes were activated in all three planes, however medulla, pons and midbrain activation was only seen in the axial and/or sagital plane scans Figure 7.1. Within the brainstem several regions were detected in the sagittal scan, but not in the other two planes Figures
7.2 7.3. High activation was seen in the dentate nucleus in both the axial and sagital planes but no evidence of dentate nucleus activation was found in the coronal plane.
With respect to cerebellar activation, axial and sagital were consistent.
7.5 Discussion
FMRI experiments usually perform a finger tapping control condition as a ref- erence in regards to apriori anatomical coordinates. The majority of FMRI studies usually image in the axial plane or in some cases at an oblique angle. The findings as shown in Table 7.1 and in the images highlight the importance of the use of different plane selections to localize activated areas and, further, may explain differences in activated areas among different FMRI studies if different plane selections are em- ployed. Finger tapping was chosen to test a new paradigm because of resulting high signal activation and apriori knowledge of anatomical coordinates. Some explanation of differences may have been due to susceptibility artifacts, gradient non-uniformities, motion/pulse artifact, and even post processing algorithms. In ideal subject all planes would yield same regions, however; nasal cavity, fillings, etc. distort some planes more than others. We expected the same sites would be activated among all three planes.
However, different planes yielded different activation maps.
94 7.6 Conclusion
The resulting activation maps from FMRI studies are influenced by the imaging plane chosen. Also, there is information that can be acquired from one scan orien- tation but not in another. Thus this leads us to to modify our FMRI paradigm and scanning protocol in order to have all three planes for FMRI analysis. In order to do this we need to develop an algorithm to combine images from different orientation scans and this will be discussed next.
95 Site Axial Coronal Sagital BA 5,34,35,37,41,46, Caudate Body, Hippocampus, Hypothalmus, Pulvinar X - X BA 4,28,30,32,38,39,42,47, Caudate Head, Lateral Globus Pallidus, Substania Nigra, Ventral Posterior Lateral Nucleus X
96 BA 10,17,19 XX BA 3,9,11,18,20,22,27,31,36, Amygdala, Corpus Callosum, Putamen, Ventral Lateral Nucleus X X BA 8,23, Dentate Nucleus XX BA 6,7,21,24,40 XXX
Table 7.1: The anatomical correlates from the three plane scans. Figure 7.1: The functional activation maps from the axial, coronal and sagital ac- quired EPI volumes.
97 Figure 7.2: The brainstem and cerebellar activation for all three planes from slice number 29 from the anatomical image of the brain.
Figure 7.3: The brainstem and cerebellar activation for all three planes from slice number 22 from the anatomical image of the brain.
98 CHAPTER 8
TWO AND THREE DIMENSIONAL IMAGE COMBINATION FOR MRI ACQUIRED USING DIFFERENT SCANNING PLANES
8.1 Abstract
The purpose of this work was to introduce a novel method for using the original information from more than one magnetic resonance imaging (MRI) volume in order to increase the resolution based on a priori knowledge of voxel resolution. This paper was motivated by MRI imaging of the brain, however is valid for all anatomy as well as other imaging modalities. The procedure for validating the MRI data combination algorithm was performed using a Shepp-Logan phantom in both 2 and 3 dimensions
(2-D and 3-D) of varying resolution matrices of 64, 96 and 128 as well as on MRI images. The squared error (SE) and mean squared error (MSE) were computed.
8.2 Introduction
Current magnetic resonance imaging (MRI) resolution enhancements focus on sophisticated pre-acquisition techniques such as scanning protocol [219–223]. At times the scanning planes are also changed within the scanning protocol, i.e. changing from the traditional axial plane to either sagital, coronal or oblique. This work exploits
99 the concept of acquiring images from more than one plane by their combination.
In essence it is fusing the data with the aim of both increasing the resolution for visual and data processing with focus on unimodal data using unions based on a priori knowledge of voxel resolution. This is in contrast to common fusion algorithms which mainly focus on visualization of either multiresolution or multimodal images
[224–226]. The combination of MRI data from three slice orientations will allow both visualization and data analysis of smaller areas that otherwise may be missed by using only one slice orientation. The validation pipeline assumes that the images are properly orientation and aligned, Figure 8.1, thus encompassing the combine sets and interpolation blocks.
Figure 8.1: Project pipeline.
8.3 Background
8.3.1 Significance
The ability of using one slice orientation over another is useful especially if the anatomical area of interest is known a priori. To date there does not seem to be any work that attempts to combine the information of three MRI slice orientations
100 using brain images. This proposal aims at taking advantage of the different slice orientations and combining them into one volume in order to have a higher resolution image. This can be achieved since the typical pixel spacing is usually AxAxZ*A where
Z is some factor. For example, having three volumes with voxel resolution of 2x2x4
(axial), 2x4x2 (sagital) and 4x2x2 (coronal) allows for the combination of all three and thus the resulting resolution would increase and the new volume would have a voxel dimension of 2x2x2.
8.3.2 Methods
A Shepp-Logan phantom was used to test the accuracy of the method in both
2 and 3 dimensions (2-D and 3-D). In 2-D the phantom was subsampled in the x- direction and in the y-direction in order to generate test subphantoms, Figure 8.2.
In 3-D the phantom was subsampled in three directions in order to simulate axial, coronal and sagital scans as seen the Figure 8.3. The combination and interpolation of the different resolution images were coded in MATLAB. The combination and interpolation of the different resolution images were coded in MATLAB.
Figure 8.2: The Shepp-Logan phantom and subsampled test phantoms in x and y directions.
101 Once the images/volumes were subsampled in 2-D/3-D, they were rescaled to a common grid. Afterward, the original pixels/voxels from each volume were mapped onto a new combined volume. Note the combination can be done with either two or three plane directions. In this paper data was used from all three planes. Combination and interpolation of the volumes were performed in an iterative fashion in order to fill in the gaps in between pixels/voxels in the combined data sets. Both the combination and interpolation are dependent on the original pixel/voxel resolutions that the images were acquired. The implementation of this routine is discussed next.
Figure 8.3: The 3-D subsampling in three planes and the combination of original voxels spacing from each plane into a new volume.
8.3.3 2 dimensional phantom (2-D)
Sample code for 2-D image interpolation for pixel spacing of 1x4 is illustrated in the Appendix A. In this case there are three iteration passes and three different weighting factors applied to the nearest neighbors accordingly. This can be extended
102 for different pixel spacing and for 3-D volumes. In this routine the subphantoms
(64x16 and 16x64 with pixel resolutions of 1x4 and 4x1) were remapped to a common grid and then an iterative process of interpolation based on the nearest neighbor(s) was performed after the initial combination of both planes, Figure 8.4. The unique feature of this process was that after the interpolation, the original data may have been averaged, thus the last two steps of the routine reinserted the original data from both planes in their proper grid points based on their original pixel resolution.
This way the original data was preserved and non interpolated information from both planes was used. Table 8.1, summarizes the iterations and weighting factors needed for other pixel spacing.
Dimension Number of Iterations Weighting Factors 2x4 2 (1/3,2/3) 1x4 3 (1/4,2/4,3/4) 2x6 4 (1/5,2/5,3/5,4/5) 2x7 5 (1/6,2/6,3/6,4/6,5/6)
Table 8.1: The iteration scheme for 2-D interpolation of combined images for several different pixel dimensions. The number of iterations and weighting factors are de- pendent on the difference between the small and large pixel dimensions, i.e. spacing. Note this algorithm is also valid for floating point resolution.
8.3.4 3 dimensional(3-D) phantom
The next step was to extend the routines for use on 3-D volumes. Three volumes are first remapped to a common size, then they are resampled based on the original volumes’ voxel dimensions, Figure 8.6. Figure 8.7 illustrates subsequent steps after combination of the three volumes where the same algorithm of preserving the original
103 Figure 8.4: The step by step iterative reconstruction process for 1x4 and 4x1 pixel spacings for subsampled test phantoms resulting in a 1x1 pixel dimension and 64x64 matrix.
104 information was extended to 3-D. The 3-D phantom used here was 64x64x64, the sub- phantoms had voxel dimensions of 1x1x4, 1x4x1, 4x1x1 simulating the three planes and the slices shown were 32x32x32. Notice the last step remapped the original data in all three planes. To further test the method in 3-D, three typical matrix sizes were simulated, namely, 64, 96 and 128. The ideal phantom, three plane combined sub- phantoms and the final reconstructed volume are illustrated in the surface rendering in Figure 8.8. A closer look at this method highlights the preservation of the original data from the three planes, in which the last two steps, the data is remapped. The axial, coronal and sagital views of the center slices for 64, 96 and 128 sized phantoms are seen, Figure 8.9.
Figure 8.5: The three volumes orientated in different planes are the phantoms simu- lating isotropic in plane resolution for coronal, axial and sagital directions.
105 Figure 8.6: This is an illustration of the first steps of the original information method on a 128x128x128 phantom. Each subphantom is upsampled to a common grid (top). Then each is subsampled based on original pixel spacing to extract original informa- tion from each volume (middle).
106 Figure 8.7: The extension of the original information algorithm to 3-D on a 64x64x64 phantom. The last steps remap the original data in x, y, and z planes (bottom row).
107 Figure 8.8: The ideal phantom, combined subphantoms after remapping and final reconstructed volume (left to right) for 64, 96, and 128 volume sizes (top to bottom).
108 Figure 8.9: The iterative process of preserving original information in the last two steps of algorithm for 64, 96 and 128 (top to bottom) for axial, coronal and sagital (left to right) center slices.
109 Quantitative Assessment
There are two ways to assess the accuracy of this method, one which has been
illustrated is visually. A more consistent approach is thus needed. The 3-D squared
error (SE) was computed between the ideal phantom (reference) and the reconstructed
volume and the results are seen visually,
2 SEijk = Tijk − Teijk (8.1)
where i,j,k represent the direct comparison of each coordinate location. The mean squared error (MSE) was also computed for phantoms of sizes 64, 96 and 128 in order to assign a value and compare the results.
n m l 1 X X X MSE = SE (8.2) n · m · l ijk i j k Here n,m,l are the number of points in the x,y,z directions respectively for the reconstructed volume.
8.4 Results
The SE of the 64x64 phantom is illustrated for three different slice coordinates.
Note the SE for the image on the far right, coordinate position (32,32,32) is zero as result of preserving the original information at this grid coordinate, Figure 8.10.
The MSE for all three phantoms (64, 96, 128) is tabulated in Table 8.3 for both major steps of the algorithm. The first row is solely after combination of all three planes and the second row is after interpolation of the remaining gaps. The MSE is inversely proportional to the original phantom size. Figure 8.11, illustrates the SE and thus the accuracy of the algorithm. The interpolation routine still has room for
110 improvement at the edges. This is evident in Figure 8.12 in which the worst case scenarios are shown for all three phantom sizes. Note the SE at the original grid locations is always zero even though not illustrated here.
Table 8.2: MSE MSE #64 #96 #128 MSE (combined) 0.016599 0.017324 0.017496 MSE (interpolated) 0.008584 0.006873 0.005359
Table 8.3: MSE for 3-D phantoms of sizes 64, 96 and 128. The MSE between the ideal phantom and the combined subphantoms without interpolation (top row) as well as after interpolation (bottom row). We expect a decrease from row 1 to 2 as well as a decrease from 64 to 128 in the second row.
This method was also implemented on real data in both 2-D and 3-D. The first 2-
D data sets were anatomical images with dimensions of 128x512 and 512x128. After combining the two images the resulting image was 512x512, Figure 8.13. The 3-
D data set used for the validation of the original information preservation method were anatomical volumes in the axial (128x128x32), coronal (128x32x128) and sagital
(32x128x128) planes, Figure 8.14. Each was mapped to 128x128x128 grid then subsampled preserving original data and combined. Figure 8.15, shows a closer look at the resulting volume beginning with the first step of the algorithm and the last two steps. Common interpolation methods stop at the middle row whilst this algorithm combines both interpolation and original information to reconstruct the volume.
111 Figure 8.10: The SE for the reconstructed volume. Slice position of 30x31x33, 31x31x31 and 32x32x32 (left to right).
Figure 8.11: The SE of 64, 96 and 128 size volumes between ideal and final recon- structed phantoms.
112 Figure 8.12: The SE of axial, coronal, and sagital (left to right) for 64, 96 and 128 (top to bottom). Slice coordinates chosen to highlight worst case for 64 (27,27,27), 96 (50,50,50) and 128 (63,63,63).
113 Figure 8.13: The original data (2-D anatomical images with dimensions of 128x512 and 512x128) remapped to a common grid of 512x512 (top) and then combined (mid- dle). The result of different interpolation points used (bottom).
114 Figure 8.14: 3-D anatomical volumes in the axial (128x128x32), coronal (128x32x128) and sagital (32x128x128) planes (top). Each was mapped to 128x128x128 grid (sec- ond from top) then subsampled (second from bottom) preserving original data and combined (bottom).
115 Figure 8.15: The top row is the first step of the algorithm and the bottom are the last two. Common interpolation methods stop at the middle row whilst this algorithm combines both interpolation and original information to reconstruct the volume and thus stopping at the bottom row. Note top row is not the same slice positions as the bottom two.
116 8.5 Discussion
A questioned may be raised; why not just interpolate the volume to increase the resolution? This is a good question; however the purpose of this method was to use information from different planes that otherwise can not be found in one plane orientation. Thus, by scanning in two or three different planes we are adding new information, specifically since the in plane resolution (xy) is usually higher than the slice thickness (z). After the combination of the original information from each plane then indeed interpolation is used, but based on the new information. This is a crucial point because it is known that in fusing different images, they are not always optimal and thus the need for fusion. In this method if only interpolation is used there would be no gain of additional information. Especially, if more data processing will be performed on the output volume and not just used for visualization. With regards to interpolation, two different methods were used, one based on nearest neighbors and the other was trilinear, however only the first was presented. Another interpolation scheme can also be used if more optimization is needed. In order to reduce the error at the edges Gaussian smoothing can be introduced after the reconstruction. Another point of differentiation between this method and common fusion algorithms is that it is not necessarily fusing the data; however, it is placing the original data in their proper grid locations based on the voxel resolution. Inherently, there will be overlap between the points and this can be dealt with in two ways. One way is by fusing the points from the three planes and using weighting factors. The other way is simply choosing one voxel over the other based on a priori decision. The latter was used for this paper. Thus, currently not worrying about which plane takes precedence over
117 other however there will be cases where careful consideration of this point is needed.
The results will slightly change depending on overlap order.
8.6 Conclusion
The method introduced in this paper was not overly complicated yet no one has implemented it as far as our knowledge and search through databases for this specific application. This ability to combine information will enrich the data sets by both enhancing the image for visualization and the data for further data processing.
118 CHAPTER 9
USING FMRI AND FNIRS FOR LOCALIZATION AND MONITORING OF VISUAL CORTEX ACTIVITIES
9.1 Abstract
The purpose of this study was to design a near infrared spectroscopy (NIRS) sensor head to continuously monitor visual cortex activation. Visual cortex activation regions as a result of eye movements were localized using functional magnetic resonance imaging (FMRI). Once the region was determined we placed the NIRS sensor head on that region and emulated the same task perfomed in the FMRI experiment. The eye movement chosen for our current validation study was saccades. One subject was instructed to move their eyes in a saccadic fashion for 30 seconds then fixate for 30 seconds. We were able to see changes in oxyhemoglobin and deoxyhemoglin using the NIRS design. These preliminary results suggest that NIRS can be used as a monitoring tool, guided by FMRI in patients who may have visual disorders.
9.2 Introduction
Studying brain functional activities is an area that is experiencing rapid growth in the field of biomedical imaging. Magnetic resonance imaging (MRI) is the most
119 common modality used in functional imaging. However, just like with any other modality, it has some drawbacks, including the lack of portability and expense. That is why functional researchers are turning to new modalities either to simultaneously use or run standalone. Near infrared spectroscopy (NIRS) is one of these modalities because of its high temporal resolution. Currently there have been and are efforts to combine NIR with MRI [58–60,62,187,227]. We introduce the concept of continuous monitoring of the visual cortex subsequent to an FMRI scan rather than simultane- ously. A. Clinical Significance Functional near-infrared spectroscopy (FNIRS) offers the ability to monitor brain activation by measuring changes in the concentration of oxy- and deoxy-hemoglobin (Hb) by their different spectra in the near-infrared range [162, 171, 172, 176]. NIRS may be useful in detecting visual dysfunction ob- jectively and noninvasively in patients with visual disturbance. Specifically, Miki et al. [189] demonstrated that a decreased activation of the visual cortex in patients with optic neuritis, which can be seen with NIRS. Also, Wilcox et al. [190] has studied the relationship between object processing and brain function in human infants, and ob- served neural activation in the primary visual cortex and the inferior temporal cortex.
However, other than these few studies no thorough studies have been made related to the visual cortex and specifically visual disorders. Thus, we combined both FMRI and FNIRS technology to first localize the regions of the cortex that are normally highly activated in the subject, and then to repeat the same task using FNIRS by properly placing the sensor on the location determined by the FMRI scan.
120 9.3 Technology Review
9.3.1 Near Infrared Light (NIR)
For more than two decades, the single channel measurement technique of NIR
spectroscopy has been successfully used to measure the hemodynamic response to
brain activity in both adults and neonates [180, 184]. More recent studies using the
multi channel NIRS technique, NIR optical topography (OT), have improved spatial
and temporal resolution in both adults [194] and infants [182]. Near infrared light
ranges from 600 − 950nm and is strongly absorbed by two chromophores, namely, oxyhemoglobin (HbO) and deoxyhemoglobin (Hb). As a result of this interaction, the modified Beer-Lambert Law (MBLL), seen in Equation 9.1, can be used to quantify changes in chromophore concentrations [152,153].
I OD = − ln final = ∆CLB (9.1) Iinitial
where OD = ODfinal − ODinitial is the change in optical density, Ifinal and Iinitial
are the measured intensities before and after the concentration of the chromophores
change, ∆C. L is the distance between incident light and detected light, is the
extinction coefficient, and B is the differential path-length factor (DPF). The general
one chromophore equation can be further expanded for brain functional monitoring
applications where two chromophores need to be monitored, yielding Equation 9.2.
λ λ λ λ ∆OD = HbO∆ [HbO] + Hb∆ [Hb] B L (9.2)
where λ indicates particular wavelength. After rearranging the mathematical terms the concentration changes can now be computed by measuring the change in optical density at two different wavelengths as seen in Equations 9.3 and 9.4.
121 λ2 ∆ODλ1/Bλ1 − λ1 ∆ODλ2/Bλ2 ∆ [Hb] = HbO HbO (9.3) λ1 λ2 λ2 λ1 HbHbO − HbHbO
λ1 ∆ODλ2/Bλ2 − λ2 ∆ODλ1/Bλ1 ∆ [HbO] = HbO HbO (9.4) λ1 λ2 λ2 λ1 HbHbO − HbHbO
The basic process is now to propagate two near infrared wavelengths through the tissue, one which is more strongly absorbed by oxyhemoglin and the other by deoxyhemoglobin at two different time periods and take the difference of these con- centrations. In this case all the other variables in Equations 9.3 and 9.4 are defined by the design. This is possibly the simplest way to do these measurements.
9.3.2 NIR Systems
Time domain (TD), frequency domain (FD), and continuous-wave (CW) systems are currently being used to solve the above implementation of NIRS [175]. TD systems
[151, 154, 228–230] introduce into tissue light impulses on the order of picoseconds which after passing through different layers such as skin, skull, cerebrospinal fluid
(CSF), and brain become broadened and attenuated [175]. The output of this pulse after being transmitted through the highly scattered medium is known as the temporal point spread function (TPSF) [228]. An advantage of TD systems over the latter two is that theoretically it can obtain the highest spatial resolution and can accurately determine absorption and scattering [175]. The disadvantages include lower temporal resolution as a direct result of trying to achieve an adequate signal to noise ratio, large dimensions, high cost due to the necessary ultrafast lasers, as well as the requirement
122 to mechanically stabilize the instrument [231]. In FD systems, [159, 232–234] the light source is continuously operating, but is amplitude modulated at frequencies on the order of tens to hundreds of megahertz. FD systems have the advantage of achieving higher temporal resolution than TD systems. One disadvantage is the noise in the measurements of scattering effects [175]. In CW systems, [174, 235–237] light sources emit light continuously at constant amplitude or modulated frequencies not higher than a few tens of kilohertz [175]. The most highly developed application of CW imaging technology is the study of hemodynamic and oxygenation changes in superficial tissues, and of the outer (cortical) regions of the brain in particular using optical topography [238]. An advantage of CW systems over the other two is their cost. However, the main disadvantage is the inability to uniquely quantify the effects of light scattering and absorption [163]. Other disadvantages are that intensity measurements are far more sensitive to the optical properties of tissues at or immediately below the surface than to the properties of localized regions deeper within the tissue [238]. This is due to the characteristic ’banana’ shape of the volume of tissue over which the measurement is sensitive, which is narrow near the source and detector and very broad in the middle [157,239]. Also the detected intensity is highly dependent on surface coupling, meaning if an optical fiber is moved slightly or pressed more firmly on the skin then it can result in a very large change in measurement [238].
9.4 Experimental Design
9.4.1 FMRI Scanning Parameters
FMRI was performed with a 3.0T GE Medical Systems Signa Excite and the
BOLD sensitive T2* weighted echo-planar (EPI) sequence [199] using an 8 channel
123 array head RF coil. Scanning protocol included a screening brain MR scan, includ- ing sagittal T1-weighted and axial T2-weighted scans, to exclude any anatomic brain abnormality (e.g., Arnold - Chiari malformation) [200]. The locations of activation and deactivation clusters were defined with respect to the Montreal Neurological In- stitute (MNI) coordinates and anatomical landmarks [201–203]. These sites were identified with the Talairach and Tournoux [204] atlas. The FMRI acquisition pa- rameters were: TE = 35ms; TR = 1.5s; flipangle = 90o single shot; full k-space;
64 64 acquisition matrix with a field of view (FOV) = 24cm, generating an in-plane resolution of 3.75mm2 with a max total of 23 axial slices. A total of 120 volumes
(time points) were acquired. For anatomic imaging, we used a three-dimensional vol- ume spoiled gradient-echo pulse sequence (0.469mm2) in the axial plane and obtained
1.3mm thick slices.
9.4.2 FMRI Data Analysis
Data analysis was carried out using FEAT (FMRI Expert Analysis Tool) Version
5.4 (FMRIB Software Library, www.fmrib.ox.ac.uk/fsl). Slice-timing correction us- ing Fourier-space time-series phase-shifting, motion correction using MCFLIRT [205], mean-based intensity normalization of all volumes by the same factor, and highpass temporal filtering (Gaussian-weighted LSF straight line fitting, with σ = 30s) were applied as pre-statistics processing. No spatial smoothing of the data was performed.
Post processing motion correction included voxel displacements: absolute (each time point with respect to the reference image = 0.14mm) and relative (each time point with respect to the previous image = 0.04mm). Independent Component Analy- sis (ICA)-based exploratory data analysis was carried out using MELODIC [206],
124 in order to investigate the possible presence of activation or unexpected artifacts.
Time-series statistical analysis was carried out using FILM with local autocorrelation
correction [207]. Z (Gaussianized T/F) statistic images were thresholded using clus-
ters determined by an average Z > 5.1 and a (corrected) cluster significance threshold
of P < 0.05 [208–210]. Registration to high resolution and/or standard images was
carried out using FLIRT [205,211].
9.4.3 FNIR Instrument and Sensor
We used a modified frequency-domain spectrophotometer (Oximeter, ISS, Cham-
paign, IL) at two different wavelength, 692nm and 834nm. Eight light sources were
used, four for each wavelength, and one photomultiplier tube (PMT) detector to col-
lect the light. The output signals from the PMT were sent to a computer for data
processing and the AC, DC, and of the wave were determined. The sensor consisted
of four source-detector pairs with four sources placed 2cm and 2.5cm away from the detector in a symmetric fashion, as shown in Figure 9.1.
Every two sources (692nm and 834nm) were coupled to allow for more accuracy and limit the number of positions to four. The sources and detector were held into place by a plastic molded strap with a slight bend in order to fit comfortably (using elastic straps) on the subject’s head.
9.4.4 Task
The subject was instructed to fixate eyes for 30 seconds (OFF condition) and then move eyes back and forth in a saccadic fashion for 30 seconds (ON condition). This same task was performed for the FMRI trial. The difference between these two states was calculated.
125 Figure 9.1: The actual sensor head design. This straps onto the head at the region determined from the FMRI trial. We used a symmetric sensor head design with one detector and eight sources. Sources 1-4 were 692nm and 5-8 were 834nm.
9.4.5 FNIR Data Analysis
The raw AC and DC data were analyzed. However we did not use the algorithm
provided by the frequency domain instrument, rather we derived our own compu-
tation based on the modified Beer-Lambert law. The raw data was filtered using a
moving average. The extinction coefficients for oxyhemoglobin and deoxyhemoglobin
were taken from a table. The differential path length factors were derived for both
wavelengths [240].
9.5 Results
After localizing the visual cortex through the FMRI activation we were able to suc-
cessfully monitor the corresponding changes in oxyhemoglobin and deoxyhemoglobin
as a result of the functional ON-OFF paradigm. Overall the results from the sources
from each side were symmetric as expected. The source-detector pairs at 2.5cm showed a stronger signal than the 2cm distance. This is expected because the 2cm
126 does not penetrate as deep as the 2.5cm distance and thus may not detect the signal as efficiently. The results of the trial are illustrated accordingly, Figure 9.2 9.3 9.4
9.5.
Figure 9.2: The relative change in concentration of oxyhemoglobin (HbO) and deoxy- hemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 4 and 8. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases.
9.6 Discussion
FMRI and NIR are complimentary to each other in that they allow for the mon- itoring and measurement of changes in the oxy to deoxy hemoglobin ratio, which corresponds to activation. While FMRI is well developed, NIR is still lagging behind with respect to imaging the visual cortex and is not ready for the clinical environ- ment. NIR measurement accuracy can vary depending on the design of the detector head. Constant optode distance is crucial; if head circumference changes even by
127 Figure 9.3: The relative change in concentration of oxyhemoglobin (HbO) and deoxy- hemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 1 and 5. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases.
Figure 9.4: The relative change in concentration of oxyhemoglobin (HbO) and deoxy- hemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 2 and 6. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases.
128 Figure 9.5: The relative change in concentration of oxyhemoglobin (HbO) and deoxy- hemoglobin (Hb) as a result of alternating between OFF-ON paradigm for sources 3 and 7. The characteristic pattern in the plot shows that as the HbO (blue) increases the Hb (green) decreases.
a fraction of a millimeter, the trends are significantly biased [192]. Strangman et al. [177] did a comprehensive study on factors affecting the accuracy of NIR concen- tration calculations and found that the wavelength selection and optode placement to be important factors in reducing error. A limitation of NIR is its low spatial resolution [156]. Statistical parametric mapping, as in FMRI, cannot be directly ap- plied to NIRS studies because there are too few measurement portions [180]. On the other hand compared to other imaging methods, optical approaches have an excellent temporal resolution [162, 195] that enables analysis in the frequency domain. NIR is also low cost, non-invasive, and portable. The use of functional MRI has proved to be a successful imaging modality but unfortunately, like many other modalities, it has its drawbacks. The main being cost and portability. Therefore, the devel- opment of novel, more selective, and noninvasive diagnostic techniques is a priority.
129 The inherent potential of this optical technique is that it is noninvasive and can be used during behavioral tasks in order to provide temporal and spatial information, making it ideal for infant research [190]. However, this technique is limited by its low spatial resolution. With all these in mind, it is clear that there is a significant clinical need for low cost, non-invasive, portable imaging modalities that combines both high spatial resolution and high physiological sensitivity to improve the clinical sensitivity and specificity for visual cortex (and long term motor cortex) diagnosis.
FNIRS has several advantages in comparison with other imaging methods, such as high flexibility, portability, low cost and biochemical specificity. Moreover, patients and children who might not stand the confined environment of functional magnetic resonance imaging (FMRI) experiments can be repetitively examined. Therefore, it is useful to further establish FNIRS as a method for functional imaging. NIRS may be useful in detecting visual dysfunction objectively and noninvasively in patients with visual disturbance, especially when used at the bedside [167, 189]. Specifically, Miki et al [189] demonstrated that a decreased activation of the visual cortex in patients with optic neuritis can be seen with NIRS. As for the field of evaluating visual func- tion via the cortex, this has not yet gained popularity. The use of FNIR and FMRI concurrently may lead to new findings in the area of functional imaging. Studies combining FMRI and FNIR are now emerging [58–60,62,187,227].
9.7 Conclusion
We have demonstrated the feasibility of using both FMRI and FNIR in order to localize and monitor visual function. This has potential for a low cost continuous
130 monitoring of visual disorder via FNIR, after an initial diagnosis is made through
FMRI.
131 CHAPTER 10
DESIGN AND DEVELOPMENT OF A FUNCTIONAL TOOL USING NIRS FOR NON-INVASIVE DETECTION OF VISUAL CORTEX ACTIVITIES
10.1 Abstract
Near infrared imaging and spectroscopy has been used for non-invasive, continuous monitoring of visual cortex hemodynamic changes in response to visual stimulation.
However, inaccurate estimation of the cortex activation area and coupling fluctuations at the skin-sensor interface place a big challenge for measurement reliability and reproducibility. We developed a near infrared portable sensor head to minimize the coupling effects and improve the measurement reliability. This chapter will discuss the procedure of modifying a near infrared spectroscopy (NIRS) system, the design of the sensor head, the development of Labview code to build a graphical user interface, and the derivation for SpO2 calculation based on pulse oximetry principles and MBLL.
The long term goal is to evaluate abnormal visual function in patient populations and develop novel assessment tools. The final sensor head consists of four source
fibers (each combines two wavelengths: 690nm and 830nm) and two detection fibers mounted on a flexible rubber strip. Elastic straps were used to secure the sensor
132 head on the subjects head comfortably. A Labview interface was designed for data acquisition, analysis, and real time display.
10.2 Background
Functional near-infrared spectroscopy (FNIRS) offers the ability to monitor brain activation by measuring changes in the concentration of oxy- and deoxy-hemoglobin
(Hb) by their different spectra in the near-infrared range [162, 171, 172, 176]. NIRS may be useful in detecting visual dysfunction objectively and noninvasively in patients with visual disturbance. Specifically, Miki et al. [189] demonstrated that a decreased activation of the visual cortex in patients with optic neuritis, which can be seen with
NIRS. Also, Wilcox et al. [190] has studied the relationship between object processing and brain function in human infants, and observed neural activation in the primary visual cortex and the inferior temporal cortex. However, other than these few studies no thorough studies have been made related to the visual cortex and specific visual disorders. Thus the motivation for developing this modality for both simultaneous
FNIRS and FMRI as well as a standalone system in a clinical environment, Figure
10.1.
10.3 Methods and Development
In this section the development stages and components for achieving a clinical
NIRS device is discussed. The components are divided into the system, sensor head, software and algorithm development.
133 Figure 10.1: An overview for the motivation of the development and validation of NIRS alongside FMRI.
134 10.3.1 System
Originally a commercially available frequency domain system was used from ISS,
Figure 10.2; however, disadvantages of this system include the cost and the lack of
flexibility, hence a more economical solution was pursued. The alternative was a low cost more portable RS232C QualSys-2044 continuous wave prototype system with a baud rate of 19200, sampling frequency of 14Hz, and power of 30mW . The system consisted of four photodiode detectors and four sources (two at 690nm and two at
830nm), as shown in Figure 10.3.
Figure 10.2: The ISS frequency domain system originally used.
After initial tests it was determined that this new system needed to be modified.
The first modification was the coupling optimization of the detector fibers and the photodiode detectors to minimize signal loss (Figure 10.4).
135 Figure 10.3: The RS232C QualSys-2044 continuous wave prototype system with a baud rate of 19200, sampling frequency of 14Hz, and power of 30mW.
Figure 10.4: The first steps taken to modify the RS-232 CW system at the detector interface in order to increase coupling efficiency. The internal components and the final system are illustrated.
136 The next step was directly couple the source fibers to the laser diodes in the system. Again this was done with the same goal and fashion as the detector cou- pling Figure 10.5. At this point no further modification was performed on the CW prototype and the focus shifted to optimizing the sensor head.
Figure 10.5: The modification of the source coupling interface in order to reduce power loss.
10.3.2 Sensor Head
The sensor head design evolved from many versions. The last three are presented here and a fourth is already being designed. The first sensor head had four source and one detector slots. Both the source and detector slots were design to allow the user to adjust the sources and detector positions as seen in Figure 10.6. Each source probe contained two fiber lines, thus allowing both 690nm and 830nm at each slot position. In this design the source diameter used was 0.5mm for both 690nm and
830nm, and the detector diameter was 4mm. The source detector separations are tabulated in Table 10.1
The finished sensor head design is seen in Figure 10.7. After use of this sensor head it was determined that another detector should be added as well as a new material
137 Figure 10.6: The schematic for the first sensor head prototype included adjustable source and detector slots. Note: not to scale.
needed to be used for both comfort and reducing the affect of ambient light influence.
The second sensor head was then developed and included the same components as the
first design except the source detector spacing was changed (Figure 10.8, Table 10.2).
One other decision made for sensor head 2 was to fix the source detector separations as seen in the actual developed design, Figure 10.9.
Figure 10.7: Sensor head design 1 fabricated and tested.
138 Figure 10.8: The schematic for the second sensor head prototype no longer included adjustable source and detector slots. Note: not to scale.
Figure 10.9: Sensor head design 2 fabricated and tested.
139 Wavelength (nm) Source Minimum Distance from A (cm) 834 1 2.0 2 2.5 3 2.0 4 2.5 692 5 2.0 6 2.5 7 2.0 8 2.5
Table 10.1: The minimum source detector distances allowed for sensor design 1.
The third sensor head was also designed in a symmetric fashion with ten slots for source and detector probes to be inserted, Figure 10.10. An example combination would be to allow a total of sixteen sources and two detectors. The source detector distances can range from approximately 1.6 − 5.7cm. The major difference between this and the previous designs is the number of sources. For our trials we used only four sources and two detectors with diameters of 1mm and 4mm, respectively. Originally, this sensor head had only eight slots available but later two additional slots were added, Figure 10.11.
10.3.3 Software
The third component for the development of the NIRS system is the develop- ment of a Labview graphical user interface (GUI). The data collection was completed through the serial port of a Lenovo R61 laptop using a graphical user interface pro- grammed and designed with NI Labview 8.0. The software included one main GUI with twelve subroutines for various functionality. The detailed code hierarchy is seen
140 Figure 10.10: The schematic for the third sensor head prototype no longer included adjustable source and detector slots and the source diameters were increased to 1mm. Note: not to scale.
Figure 10.11: The different configurations for sensor head design number 3.
141 Wavelength (nm) Source Distance from A (cm) Distance from B (cm) 834 1 3.223 4.515 2 3.036 3.685 3 3.737 3.041 4 4.783 3.343 692 1 3.223 4.515 2 3.036 3.685 3 3.737 3.041 4 4.783 3.343
Table 10.2: The source detector distances for sensor design 2.
in Figure 10.12. Currently the GUI reads and writes data as well as allows for real time marker stamps and the ability to choose to write the data to a file or simply monitor, Figure 10.13. Other features include adjusting the power and the sampling rate. Signal post processing code is still being developed but not shown here.
10.4 Pre Trial test
Once the first three components are complete then pre trial tests can be made to test the functionality and feasibility of the entire system. In order to do this we used a blood pressure cuff to occlude blood flow and monitor the signal acquired and displayed by the GUI. The setup is seen in Figure 10.14.
The protocol was to occlude the flow four times and release. The first time the occlusion was light (five pumps) and the next three were heavy (ten pumps). The resulting signal was good and followed the protocol well, Figure 10.15.
Other steps were taken to assure proper functionality of the system such as running the system for one hour straight and testing for drifts and fluctuations. After these
142 Figure 10.12: The Labview code hierarchy.
143 Figure 10.13: The Labview graphical user interface.
Figure 10.14: The test setup for functionality evaluation of system, sensor head and software.
144 Figure 10.15: The resulting signal from complete setup from one source detector combination. The other combinations showed similar trends.
tests were complete the final experimental setup is prepared consisting of the modified
CW system, sensor head, labview GUI and the visual stimulus, Figure 10.16.
Figure 10.16: The experimental setup consists of the modified CW system, sensor head, labview GUI and the visual stimulus.
The stimulus is a blocked design visual paradigm was used to stimulate the pri- mary visual cortex. The design consisted of 30 seconds baseline followed by 30 seconds
145 of stimulation using a reversal checkerboard pattern flashing at 3.7Hz for a total du- ration of 150 seconds. The system hardware is still being developed and is currently under more performance evaluation.
10.5 Conclusion
We have introduced a portable CW NIR device with a sensor head. We have also introduced a Labview GUI interface that will allow the user to define acquisi- tion parameters and acquire data. Future work includes adding the further signal processing components to the GUI along with computational algorithm (discussed in next Chapter). The fourth sensor head design is already being prepared with newer wavelengths and more sources and detectors. Once complete it will be validated on normal subjects before evaluating abnormal visual function in patient populations in order to assess the potential of the system for the diagnosis of visual disorders.
146 CHAPTER 11
ALGORITHM SIMULATION
11.1 Introduction
In this chapter we first discuss the forward problem and review the heterogeneous diffusion equation and discuss two methods to derive a solution, namely, the Born expansion and the Rytov expansion, of light energy density changes [241, 242]. Sec- ondly, we discuss the inverse solution based on pulse oximetry (PO). Specifically, we develop and an analytical solution to calculate oxyhemoglobin, deoxyhemoglobin and
SpO2 based on AC and DC measurements. Finally, we present a model and simulation results. The following discussion is an adaptation based on [241,242].
11.2 The Heterogeneous Diffusion Equation
We are mainly concerned with the changes in the absorption coefficient (µa(r))
and assume that the reduce scattering coefficient (µs(r)) is constant. Thus (µa(r)) is
o subdivided into a background, µa and a homogeneous, δµa (r) components.
o µa (r) = µa + δµa (r) (11.1)
147 After substituting Equation 11.1 into the diffusion equation we attain the heteroge-
neous equation for the light energy density at r from a source at rs, Aδ (rs)
2 −D∇ Uac (r) + (−iw + vµa) U (r) = −Aδ (rs) (11.2)