5/3/16 Problem Statement AUTOMATED HIPPOCAMPUS • Goal: automated segmentation of hippocampus in SEGMENTATION OF BRAIN MRI brain MRI images Input Output hippocampus IMAGES Group 7 Yingchuan Hu Chunyan Wu Li Zhong May 1st, 2014 Cornell University Hippocampus Anatomy Clinical Significance • Epilepsy is the most common serious brain disorder The hippocampus is located in the medial temporal lobe of worldwide. the brain. 1 Temporal lobe • Prevalence of epilepsy worldwide (WHO ) • 7 sufferers in every 1,000 people • 3 new sufferers in every 10,000 people each year Frontal lobe Occipital lobe • People with epilepsy are at increased risks for status epilepticus (life-threatening) • One continuous, unremitting seizure lasting longer than five minutes or recurrent seizures without regaining consciousness between seizures for greater than five minutes. • Prevalence of status epilepticus in US (NIH2) • 195,000 new patients of status epilepticus each year • 42,000 deaths caused by status epilepticus each year 1. World Health Organization http://www.who.int/mental_health/neurology/epilepsy/en/ *Modified from a scan of a plate of “Posterior and inferior cornua of left lateral ventricle exposed 2. National institute of Health http://www.ninds.nih.gov/disorders/epilepsy/detail_epilepsy.htm from the side” in Gary’s Anatomy Clinical Research Issues for Segmentation • Hippocampal volume reduction >10% of “normal” size indicates epilepsy.[1-4] • Low contrast to neighboring brain • “normal”: structures • People with the same age • Bilateral hippocampus comparison • Personal changes in more than 1 year • Over 90% sensitivity + 98% specificity for amgydala hippocampus [5-7] MRI image measurement diagnosis. • No clear boundary between hippocampus 1. Cook MJ, et al. (1992). Brain 115:1001–1015. 2. Jack, C. R., et al. (1990). Radiology 175:323–429. 3. Jackson, G. D., et al. (1994). Neurology 44:42–46. and amygdala 4. Bronen RA, et al. (1995). AJNRAmJ Neuroradiol16:1193–1200. 5. Oppenheim C, et al. (1998). AJNRAmJNeuroradiol 19:457–463. 6. Jack CR. (1995). Neuroimaging Clinics of North America 5:597–622. 7. Tien RD, et al.(1993). Radiology 189:835–842 8. Baulac M., et al. (1998). Ann Neurol 44:223-33 1 5/3/16 Previous work 99% confidence interval Algorithm Overview Dataset Method Result5 Somasundaram’s algorithm1: Hybrid generative/ Same as our [0.61-0.67] Tu 1 2008 discriminative model, Issues: no clear boundary & low contrast dataset Dice Coeff PCA Morpholog Middle- Fuzzy Noise Thresholdi Largest cial block logic edge Connected 2 removal ng Aljabar Same as our Multi-atlas model, [0.71-0.79] operations selection detection Component 2007 dataset label fusion classifier Dice Coeff 518 cases Atlas registration, [0.763-0.902] Van3 2008 (20 manually voxel classification Dice Coeff marked) and graph cuts Fiorina4 Adaptive threshold, [0.73-0.75] 56 cases 2012 probability map Dice Coeff 1. Tu, Z. et al. (2008). Brain anatomical structure segmentation by hybrid discriminative/generative models. IEEE Transactions on Medical Imaging, 27, 495-508. 2. Aljabar, P. et al. (2007). Classifier selection strategies for label fusion using large atlas databases. Medical Imaging Computing and Computer-Assisted Intervention, 10, 523-531. 3. van der Lijn, F. et al. (2008). Hippocampus segmentation in MR images using atlas registration, voxel classification, and graph cuts. Neuroimage, 43(4), 708-720. 4. Fiorina, E. et al. (2012, June). Fully automated hippocampus segmentation with virtual ant colonies. In Computer- Based Medical Systems (CBMS), 2012 25th International Symposium on (pp. 1-6). IEEE. 5. Confidence Interval Calculator 1. Somasundaram et al. Segmentation of hippocampus from human brain MRI using mathematical morphology and fuzzy logic. In Emerging Trends in Science, Engineering and Technology (INCOSET), 2012 International Conference on (pp. 269-273). IEEE. Middle- Noise Min/max Threshold Edge Largest Our algorithm based on Somasundaram: block connected removal filtering ing detection selection component Middle- Noise Min/max Threshold Edge Largest block connected removal filtering ing detection selection component filter size w1 filter size w2 block control points P1,P2 before min/max filtering after min/max filtering filter size w1 filter size w2 block control points P1,P2 Enhanced image fe = f + fp - fv Properties of min/max filtering: • Min/max Filtering: a combination of min and max filtering to enhance image contrast • min(): min filter max(): max filter f: original image • More enhancement on “small” peaks Extract valleyspeaks • Separate the hippocampus region from neighboring Original image f à ff1 == min(max(f))max(min(f)) à valleyspeaks ffp == ff - -f 1f regions 2 enhancedv 2 original • Control enhanced peaks by filter size • Use proper size of filter to choose peaks for enhancement Middle- Noise Min/max Threshold Edge Largest block connected removal filtering ing detection selection component Hypothesis filter size w1 filter size w2 block control points P1,P2 • The algorithm can segment the hippocampus- amygdala region in all cases to 0.7 DC • Hippocampus is located in P2 the medial temporal lobe à • The algorithm can segment hippocampus and only consider the middle part amygdala for 3.0T cases to 0.7 DC of the sagittal view à defined as “middle-block” • Middle-block • Rectangle defined by two control • Evaluation function: P1 points P1(x1,y1), P2(x2,y2) • Dice Coefficient (DC): 2(A ∩ B) / (A + B) • x1=image width / 3 • A: Ground truth B: Segmentation result • x2=2x1 • y1=image height / 2 – 20 • y2=image height / 2 + 20 2 5/3/16 Dataset Experiment • Data Set1 (with manual markings) • For C1 • Training set: 8 C1 (5 E & 3 N) • C1: • Testing set: 7 C1 (5 E & 2 N) • GE 1.5T Coronal T1W MR • For C2 • voxel size=0.78x0.78x2mm image dim=256x256x124 • Training set: 5 C2 • C2: • Testing set: 5 C2 • GE 3.0T Coronal T1W MR • Parameters tuning • voxel size=0.39x0.39x2mm image dim=512x512x112 Middle- Noise Min/max Threshold Largest block connected removal filtering ing selection component • 15 C1 • 10 epilepsy(E) & 5 non-epilepsy(N) Filter size w2 • 10 C2 (E) Block control points P1, P2 1.Jafari-Khouzani, K. et al. (2011). Dataset of magnetic resonance images of nonepileptic subjects and temporal Mean filter size w1 lobe epilepsy patients for validation of hippocampal segmentation techniques. Neuroinformatics, 9(4), 335-346. Changes Results-parameters Local block • Mean filter deleted filter size w block control points P1,P2 1 size w • Reason: Data with low resolution, lose edge information when 2 Erosion- using mean filtering Min/max Middle-block Largest Thresholding Connected filtering selection Region Component growing • Thresholding method changed • Optimal Parameters • From Balanced Histogram Thresholding (BHT) to local w: image width, h: image height thresholding • For Dataset C1(1.5T): Local thresholding • Reason: w1 w2 P1 P2 • Low contrast between hippocampus 3 20 (w/3,h/2-20) (2w/3,h/2+20) and neighboring regions • Non-uniform intensity for interested local block regions • For Dataset C2(3.0T): local mean w1 w2 P1 P2 • Erosion-Region growing added 4 23 (w/3,h/3) (2w/3,h/3+60) • Reason: gain a more accurate boundary Image 1. Local variations Results-evaluation Discussion inside hippocampus 2. Errors caused by transferring from • Poor outcomes: coronal to sagittal • HF015 ground truth our result view Dice Coefficient for C1(1.5T) Dice Coefficient for C2(3.0T) 0.9 0.8 Dice Coefficient: 0.8 0.7 0.23 0.7 0.6 0.6 0.5 0.5 0.4 Dice Coefficient 0.4 Dice Coefficient 0.3 0.3 • Good outcomes: 0.2 0.2 0.1 0.1 • HF009 ground truth our result 0 HF009 HF010 HF011 HF012 HF013 HF014 HF015 0 HF021 HF022 HF023 HF024 HF025 Dice Coefficient: Mean dice coefficient : 0.6 Mean dice coefficient : 0.672 0.78 3 5/3/16 Summary • Min/max filtering and local filtering can correct non-uniform backgrounds and enhance boundaries • Parameters for MR imaging highly affect the performance of image analysis algorithms • Our method is sensitive to middle-block selection • Our result didn’t meet our expectation but close picture from: www.timothy-carter.com 4 .