Diagnosis of neurodegenerative diseases based on multi-modal hemodynamic classification of the brain

Accurate diagnostic assessment of metabolic processes in nuclear medicine diagnostic imaging, e.g. SPECT and PET, rely on specific localization of physiological activity. The major step in precise staging of neuro-degenerative diseases is robust, patient-specific classification of the brain. In this work a vascularization-based classification strategy for MRI datasets of the brain is introduced to handle variability of patient's anatomy. The vascularization-based classification utilizes skeletonization in combination with m-adjacency to construct a hierarchical vessel tree from binary pre-segmentations. Based on the vessel topology, the brain voxels are classified with respect to a minimal distance criterion from the vessel branches they are assigned to. This blood-supply oriented approach shows proper segmentation of respective anatomical regions of the human brain. Results are validated on T1-weighted brainweb database.