FocusNetv2: Imbalanced large and small organ segmentation with adversarial shape constraint for head and neck CT images

Radiotherapy is a treatment where radiation used to eliminate cancer cells. The delineation of organs-at-risk (OARs) vital step in radiotherapy planning avoid damage healthy organs. For nasopharyngeal cancer, more than 20 OARs are needed be precisely segmented advance. challenge this task lies complex anatomical structure, low-contrast organ contours, and the extremely imbalanced size between large small Common segmentation methods that treat them equally would generally lead inaccurate small-organ labeling. We propose novel two-stage deep neural network, FocusNetv2, solve challenging problem by automatically locating, ROI-pooling, segmenting organs with specifically designed localization sub-networks while maintaining accuracy segmentation. In addition our original FocusNet, we employ adversarial shape constraint on ensure consistency estimated shapes prior knowledge. Our proposed framework extensively tested both self-collected dataset 1,164 CT scans MICCAI Head Neck Auto Segmentation Challenge 2015 dataset, which shows superior performance compared state-of-the-art head neck OAR methods.