Corpus Callosum Probabilistic Subdivision based on Inter-Hemispheric Connectivity

1. Average Probabilistic Contour Model Corpus Callosum Probabilistic Subdivision based on Inter-Hemispheric Connectivity Martin Styner1,2, Ipek Oguz1, Rachel Gimpel Smith2, Carissa Cascio2, Matthieu Jomier1 1 Department of Computer Science, University of North Carolina at Chapel Hill, 2 Department of Psychiatry, University of North Carolina at Chapel Hill Shape analysis and Diffusion Tensor Image (DTI) tractography have become of increasing interest in neuroimaging. Here both are employed to compute a probabilistic subdivision model of the Corpus Callosum (CC) structure. The CC subdivision allows us to study the regional CC morphology regarding area measurements or Diffusion Tensor Imaging properties. As a first small scale application, we applied it to a small study of regional CC growth in pediatric healthy controls. Introduction The corpus callosum (CC) is the major commissural pathway between the hemispheres and plays an integral role in relaying sensory, motor and cognitive information from homologous region in the two hemispheres. It is of much interest in neuroimaging studies of normal development, schizophrenia and autism. The computation of CC regional areas is most commonly executed manually by relabeling an already segmented structure into subregions. These manual methods are time-consuming, not reproducible and subjective. The currently most widely applied subdivision scheme for the CC was originally proposed by Witelson[4] and is motivated by neuro-histological studies. We propose a novel automatic CC subdivision based on probabilistic boundaries based on inter-hemispheric connectivity from Diffusion Tensor Imaging[1] (DTI). Fig. 3: Left: Fibers of 4 selected lobes transformed back to MR image space. Middle: Schematic visualization of probability computation. Right: Contour probability maps of 5 training cases. Results The probability maps of all 5 cases in the training population show a high similarity across all cases and the subdivision model. The largest variability is present in the occipital-temporal lobe section. Alternatively, we also computed the hard decision maps, which resulted in a high decision variability in all cases and the final probability map. The application of the subdivision model shows that the occipital-temporal lobe region has a low probability in a relatively large region. The resulting probabilistic area is relatively large (21% for the shown case). A hard decision model would highly underestimate this area. Fig. 1: Corpus callosum in an MR image (left) with Witelson subdivision[5] and its neuro-histological motivation (right). This subdivision scheme is sensitive to alignment and/or manual labeling. Methods/Data/Material Our subdivision is based on a training population of 5 pediatric cases (age 2-4y). We first compute for each case the automatic lobe subdivision and CC segmentation. The CC is segmented as a 2D contour on the midsagittal plane based on a deformable shape model trained on over 200 cases (both adult and pediatric). The lobe subdivision uses a fluid deformable registration of a pediatric lobe atlas to all cases Fig. 4: Left: Final probabilistic subdivision model. Right: Sample subdivision case with relative area noted below. The subdivision was applied to a small study of CC growth in 3 healthy subjects from age 2 to 4. The results show CC growth over its full length, and its main growth in the region associated with anterior frontal lobe connections. This region experienced a relative growth of 26% (posterior-frontal:24%, parietal:15%, occipital-temporal:15%). Conclusions We developed a novel probabilistic CC subdivision based on inter-hemispheric connectivity. Applied to a study of healthy growth from age 2 to 4, we showed that the main growth is in CC regions associated with frontal lobe connections. Fig. 2: Left: Midsagittal MR slices and the automatically segmented corpus callosum using deformable shape models. Middle: 3D view of the lobe subdivision (yellow=frontal, purple=parietal, red=occipital, green=temporal lobe). Right: Set of inter-hemispheric connectivity from a sample DTI dataset. In-vivo assessment of the inter-hemispheric pathways through the CC is difficult, but can be approximated using DTI and Tractography[2,3]. The lobe subdivision serves as an initialization for the DTI Tractography. This leads to a set of inter-hemispheric DTI fiber tracts for each lobe set. In the next step we compute a distance-weighted probabilistic subdivision of the CC contour from the location of all tracts. The resulting subdivisions are averaged to produce the final CC subdivision model consisting of probabilistic contour maps that assign to each contour point the probabilities to belong to any of the connectivity based subdivisions. The probabilities are propagated to the whole CC object using a Danielsson distance transform based label map. Our method is fully automatic and its results are more stable than commonly applied schemes such as the Witelson subdivision. Fig. 5:Relative growth curves of CC subdivision regions. Data from 3 healthy subjects along mean curves from age 2 to age 4. A: Regional growth relative to the overall CC growth. B: Regional growth relative to the size of the corresponding region at age 2. Original brain images for the corpus callosum were provided by BIOMORPH consortium (EU BIOMED 2) and the UNC Autism center REFERENCES [1]. Fillard, P., Gilmore, J., Lin, W., Piven, J., Gerig, G.: Quantitative analysis of white matter fiber properties along geodesic paths. MICCAI. 2879 in Lecture Notes in Computer Science (2003) 16–23 [2]. Xu, D., Mori, S., Solaiyappan, M., van Zijl, P., Davatzikos, C.: A framework for callosal fiber distribution analysis. NeuroImage 17 (2002) 11311143 [3]. Gee, J., Zhang, H., Dubb, A., Avants, B., Yushkevich, P., Duda, J.: Anatomy-based visualizations of diffusion tensor images of brain white matter. Vis. and Image Processing of Tensor Fields. (2005) [4]. Witelson, S.: Hand and sex differences in the isthmus and genu of the human corpus callosum. a postmortem morphological study. Brain 3 (1989) 799–835 [5]. Narr, K., Thompson, P., Sharma, T., Moussai, J., Cannestra, A., Toga, A.: Mapping morphology of the corpus callosum in schizophrenia. Cereb Cortex 1 (2000) 40–9 May 2005, UNC Radiology Symposium