ABSTRACT
This chapter evaluates a multi-atlas segmentation system using an online atlas selection approach to choose a subset of optimal atlases for multi-atlas segmentation. The atlas selection is performed in two phases. In the first phase, the correlation coefficient of the image content in a cubic region is used to rank the atlases in an atlas pool. A subset of atlases based on this ranking is selected, and deformable image registration is performed to generate deformed contours and deformed images. In the second phase, Kullback-Leibler divergence is used to measure the similarity of local intensity histograms between the new image and each of the deformed images, and the measurements are used to rank the previously selected atlases to further identify a subset of optimal atlases. The deformed contours from these optimal atlases are fused together using a modified simultaneous truth and performance level estimation algorithm to produce the final segmentation. The approach is validated with promising results using a public grand challenge dataset for thoracic auto-segmentation.
