Functional Segmentation in 3D Angiography

For computer-based diagnostics on high-resolution 3D angiography, precise modeling of anatomy is inevitable for facilitating reliable results. Thereby the process of classification normally covers pre-processing steps, a segmentation task and the classification itself generally utilizing a priori knowledge. The quality of the applied a priori model, e.g. an atlas for cortical-center classification, is deciding for the accuracy of the classification. Anatomical variations of vascularization-dependent partitioning in succession of vessel bypasses can hardly be considered by deforming an atlas. We present a classification algorithm utilizing alternately applied dilation kernels for iteratively assigning the tissue to classify according to the distance from the sustaining vessel systems to improve the expressiveness and validity of the analysis and models for medical diagnostics. The discussed classification strategy has been successfully applied to Coui-naud’s liver lobe classification on contrast enhanced CT data and is currently being evaluated for vascularization-dependent classification of the brain into Brodmann’s areas in the context of neurosurgery.