MRI quantification of cranial nerves such as anterior visual pathway (AVP) inMRI is challenging due to their thin small size, structural variation along itspath, and adjacent anatomic structures. Segmentation of pathologically abnormaloptic nerve (e.g. optic nerve glioma) poses additional challenges due tochanges in its shape at unpredictable locations. In this work, we propose apartitioned joint statistical shape model approach with sparse appearancelearning for the segmentation of healthy and pathological AVP. Our maincontributions are: (1) optimally partitioned statistical shape models for theAVP based on regional shape variations for greater local flexibility ofstatistical shape model; (2) refinement model to accommodate pathologicalregions as well as areas of subtle variation by training the model on-the-flyusing the initial segmentation obtained in (1); (3) hierarchical deformableframework to incorporate scale information in partitioned shape and appearancemodels. Our method, entitled PAScAL (PArtitioned Shape and AppearanceLearning), was evaluated on 21 MRI scans (15 healthy + 6 glioma cases) frompediatric patients (ages 2-17). The experimental results show that the proposedlocalized shape and sparse appearance-based learning approach significantlyoutperforms segmentation approaches in the analysis of pathological data.
展开▼
