Cortical sulcal bank segmentation via geometric similarity based graph partition

Development of imaging predictor of brain cytoarchitecture has been of significant interest in neuroimaging. Cortical geometric characteristic has been shown to be a good predictor of cortical cytoarchitecture. In this paper, a novel method based on sulcal geometric features is proposed for the extraction of sulcal banks, which are the cortical regions bounded by sulcal fundi and adjacent gyral crest lines. Given parcellated sulcal basins, we apply two graph partition techniques including the normalized cuts and the graph cuts to partition a sulcal basin into two opposing sulcal banks. Particularly, we designed novel geometric similarity metrics and cost functions to adopt these two graph partition algorithms specifically for our applications. As a test bed application, we applied this method to extract the anterior and posterior sulcal banks of the central sulci on over 400 cortical surfaces and achieved promising results. In addition, we applied this method to study the cortical thickness of central sulci and found that the cortical thickness of the anterior sulcal bank is significantly thicker than that of the posterior sulcal bank, suggesting that the segmented sulcal banks can differentiate cortical thickness layout, which is believed to be an indicator of brain cytoarchitecture. Finally, the asymmetry and longitudinal changes of cortical thickness are analyzed using the OASIS database and reasonable results are obtained.