681.06: Automatic cortical labeling system for neuroimaging studies of normal and disordered speech Jason A. Tourville  Frank H. Guenther  Department of Speech, Language, and Hearing Sciences, Department of Biomedical Engineering, Center for Computational Neuroscience and Neural Technology (CompNet) Background The SpeechLabel Cortical Labeling System Volume-based inter-subject averaging methods fail to account for cortical variability, resulting co-regis- Regions are plotted on the Freesurfer fsaverage surface template for the left hemisphere. Each cortical hemisphere is divided into 63 regions of interest based on individual anatomical landmarks. Superior temporal, in- tration of functionally distinct regions and a loss of statistical power. This is particularly problematic for ferior frontal, precentral, and medial frontal regions are divided into functionally meaningful subunits for neuroimaging studies of speech. See table 1 for region abbreviation key. neuroimaging studies of speech because functionally distinct regions in the peri-Sylvian region that are distant along cortical surface are proximal in the volume. Individual region of interest (ROI)-based analy- Inated Surface Pial Surface sis analysis of cortical responses signicantly improves statistical power (Nieto-Castanon et al., 2003) but Lateral Medial

pdPMC requires lengthy, tedious labeling by an expert. dMC Lateral dSC Medial mdPMC dMC pdPMC dSC SPL SMA adPMC preSMA pMFg midMC SFg midPMC aSMg pSMg PCN pCGg midCGg SFg aMFg vSC Ag pIFs vMC

vPMC PO pCO aIFs PT dIFovIFo aCO pFO OC OC dIFt Hg aCG FP pINS pSTg FP vIFt aFO pdSTs MTO aINS PP pvSTs Lg FOC ITO SCC FMC aSTg pMTg pPHg adSTsavSTs TP pITg Dorsal Ventral aMTg aPHg aITg pITg ITO aITg aFO pFO TP aTFg pTFg aINS TOFg Ventral pPHg aPHg FOC FP OC Anatomical variability after volume-based normalization and co-registration. Left: Mean Lg voxel overlap of superior temporal ROIs at various group sizes. Little of no overlap is FMC SCC seen with groups of 9 subjects. Right: The alignment of major sulci. Sulci were drawn on 2 individual surfaces (top) and then overlaid on the same surface (bottom), aligned to the Sylvian ssure. Overlap of automatic and manually edited labels Anatomical landmarks PFS: Mean PO: 88% +/- 10%, Range: 46-100; PWS: Mean PO: 88% +/- 9%, Range: 52-100 Freesurfer’s (http://surfer.nmr.mgh.harvard.edu/) individual cortical surface-based inter-subject averag- Bounding sulci are shown on the mean surface curvature of the inated fsaverage surface template for Good overall overlap b/w auto and edited labels and no dierence b/w groups (2-tailed paired t-test of mean over- ing (Fischl et al., 1999) improves upon volume-based methods (eg., Ghosh et al, 2010) but packaged cor- the left hemisphere. Red (positive curvature) indicates a sulcal region; green (negative curvature) indi- laps for each region: p > .41). But are specic regions of low overlap, especially subdivisions of the inferior frontal tical labeling atlases fail to distinguish putative functionally distinct areas that underlie speech process- cates a gyral regions. Additional dividing landmarks include sulcal intersections and the anterior com- region (see plot below). es. Here, we describe a cortical labeling system tailored for neuroimaging studies of speech that was missures . White stars indicate landmarks that are note easily viewed on the average template. See used to generate a Freesurfer-based labeling atlas. We applied the atlas to T1 volumes of 18 uent Table 2 for landmark abbreviation key. PO of speech regions with PO < 75% in one of the groups subject groups. speakers and 20 persons who stutter to assess the performance of the automatic atlas.

80 PWS Methods cs sfrs itps cgs PFS pocs * sbps 70 SpeechLabel System: Each hemisphere is divided into 63 cortical regions based on individual anatom- prcs cas pos ical landmarks. Superior temporal, inferior frontal, precentral regions are divided into multiple gyral csts1 ifrs pals and sulcal components. ls phls csts2 60 aals csts3 aocs ccs crs hs

SLaparc Atlas Generation: Image segmentation, cortical surface reconstruction, surface-based ahls % Overlap (Dice) 50 ftts sts sros co-registration, and initial surface labeling was performed on T1 MRI volumes from 17 neurologically crs cos AC its normal subjects (9 Female, Age: 20-43, Mean: 29) using Freesurfer v5.0. T1 data aqcuisition: Siemens 40 TIM Trio 3T scanner, MPRAGE sequence, 1mm voxel size. its rhs aals ots **

Left vIFt Left aIFs Surface labels were edited to conform to the SpeechLabel system (editing was overseen by a trained ahls crs Left vIFo Left SMA Left dIFo Left dIFt Left pIFs rhs cos Right vIFo Right dIFo Right vIFtRight aFO Right dIFt Right aIFsRight pIFs neuroanatomical expert). The labeled surfaces were used as a training set to generate the SLaparc atlas Left pdPMC using standard Freesurfer tools (Fischl et al. 2004). olfs Conclusions Atlas Application: The atlas was used to automatically label 18 additional uent speakers (PFS; 4 The SpeechLabel system divides the cortex into regions that are tailored for neuroimaging studies of Female, Age 19-35, Age Range: 19-35) and 20 matched persons who stutter (PWS; 5 Female, Age: speech production. 18-47, Mean: 27; SSI-4: 13-43, Median: 26 ). [See POSTER 681.12 for further analysis of these data] Table 2: Landmarks The SLaparc Freesurfer atlas automatically applies the system to T1 MRI volumes with relatively little To assess the performance of the automatic atlas, labels were corrected by a trained expert to conform aals anterior ascending ramus of the lateral user input/anatomical expertise, providing cortical ROIs at a huge time savings: Table 1: Regions ahls anterior horizontal ramus of the to the SpeechLabel system and the automatic and manually corrected labels were compared in terms of aocs anterior occipital sulcus Fully manual operator editing time: 8-16 hours/brain percent overlap. Cortical regions were mapped to the each subjects T1 volume for these comparisons. aCGg anterior cingulate gyrus pCGg posterior cingulate gyrus cas callosal sulcus aCO anterior central PCN cortex ccs Fully automatic operator editing time: .5-1 hours/brain adPMC anterior dorsal pCO posterior central operculum cgs 2 |X ∩ Y| cos collateral sulcus Region % overlap, PO, given by Dice coecent: PO = adSTs anterior dorsal pdPMC posterior dorsal premotor cortex Auto+manual cleanup time: 1-2 hours/brain |X|+|Y| aFO anterior frontal operculum pdSTs posterior dorsal superior temporal sulcu crs circular insular sulcus Ag pFO posterior frontal operculum cs aIFs anterior pIFs posterior inferior frontal sulcus csts1 caudal superior temporal sulcus, rst segment Freesurfer tools provide an easy means to create surface and volume-based cortical AND white Labeled fsaverage: The Freesurfer fsaverage surface template was labeled by SLaparc and manually aINS anterior insula pINS posterior insula csts2 caudal superior temporal sulcus, second segment aITg anterior pITg posterior inferior temporal gyrus csts3 caudal superior temporal sulcus, third segment matter regions of interest . edited as needed. aMFg anterior pMFg posterior middle frontal gyrus ftts rst transverse temporal sulcus aMTg anterior pMTg posterior middle temporal gyrus hs Heschl's sulcus aPHg anterior PO parietal operculum ifrs inferior frontal sulcus 2 Useful applications for neuroimaging research: Acknowledgments aSMg anterior PP planum polare ihs interhemispheric sulcus i. individually derived ROIs for group comparative analyses, e.g., morphometric analyses, func- This research was supported by NIDCD R01 grants DC007683 and DC002852 (PI: FG). We would like to thank Bobbie Holland for editing aSTg anterior pPHg posterior parahippocampal gyrus itps its brain labels and Jennifer Segawa and Shanqing Cai for providing the T1 volumes. aTFg anterior temporal fusiform preSMA presupplementary motor area avSTs anterior ventral superior temporal sulcus pSMg posterior supramarginal gyrus locs tional analysis, diusion tensor analyses, that is not dependent upon inter-subject image volume dIFo dorsal , pars opercularis pSTg posterior superior temporal gyrus ls lateral sulcus References dIFt dorsal inferior frontal gyrus, pars triangularis PT olfs averaging [See POSTER 681.12 and POSTER for examples] Fischl, B., Sereno, M.I., Dale, A.M., (1999). Cortical Surface-Based Analysis II: Ination, Flattening, and a Surface-Based Coordinate System. NeuroImage, dMC dorsal pTFg posterior temporal fusiform ots occipitotemporal sulcus 9(2):195-207. dSC dorsal somatosensory cortex pvSTs superior temporal sulcu pals posterior ascending ramus of the lateral sulcus ii. provides a common substrate for localizing eects within a surface-based template Fischl, B., Salat, D.H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., van der Kouwe, A., Killiany, R., Kennedy, D., Klaveness, S., Montillo, A., FMC frontal medial cortex SCC subcallosal cortex pcs paracentral sulcus FOC frontal orbital cortex SFg phls posterior horizontal ramus of the lateral sulcus But, accuracy of automatic classier needs improvement in some key speech processing regions, in- Makris, N., Rosen, B. and Dale, A.M. (2002). Whole brain segmentation: automated labeling of neuroanatomical structures in the FP frontal pole SMA pis primary intermediate sulcus . Neuron, 33(3): 341-355. Hg Heschl's gyrus SPL pocs postcentral sulcus cluding medial and inferior frontal areas. Why? Bounding landmarks in these areas are highly vari- Fischl, B., van der Kouwe, A., Destrieux, C., Halgren, E., Segonne, F., Salat, D.H., Busa, E., Seidman, L.J., Goldstein, J., Kennedy, D., Caviness, V., ITOg inferior temporal occipital gyrus TOFg temporal occipital pos parietooccipital sulcus prcs precentral sulcus Makris, N., Rosen, B. and Dale, A.M. (2004). Automatically parcellating the human . Cerebral Cortex, 14(1): 11-22. Lg TP temporal pole able and boundaries are not well captured by surface curvature (only aects Application (i)!). mdPMC middle dorsal premotor cortex vIFo ventral inferior frontal gyrus, pars opercularis rhs Satrajit S. Ghosh, Sita Kakunoori, Jean Augustinack, Alfonso Nieto-Castanon, Ioulia Kovelman, Nadine Gaab, Joanna A. Christodoulou, midCG middle cingulate gyrus vIFt ventral inferior frontal gyrus, pars triangularis sbps subparietal sulcus Christina Triantafyllou, John D.E. Gabrieli, Bruce Fischl (2010). Evaluating the validity of volume-based and surface-based brain image midMC middle motor cortex vMC ventral motor cortex sfrs As we continue to rene and expand the labeling system, need to develop piece-wise labeling tem- registration for developmental cognitive neuroscience studies in children 4 to 11 years of age. NeuroImage 53 (2010) 85–93 midPMC middle premotor cortex vPMC ventral premotor cortex sros superior rostral sulcus MTOg middle temporal occipital gyrus vSC ventral somatosensory cortex sts superior temporal sulcus Nieto-Castanon, A., Ghosh, S.S., Tourville, J.A. and Guenther, F.H. (2003). Region of interest based analysis of functional imaging data. Neu- ti temporal incisure plates based on better understanding of region/sulcal variability. roimage, 19(4): 1303-1316.