Abstract
Organs-at-risk (OAR) delineation in computed tomography (CT) is an importantstep in Radiation Therapy (RT) planning. Recently, deep learning based methodsfor OAR delineation have been proposed and applied in clinical practice forseparate regions of the human body (head and neck, thorax, and abdomen).However, there are few researches regarding the end-to-end whole-body OARsdelineation because the existing datasets are mostly partially or incompletelyannotated for such task. In this paper, our proposed end-to-end convolutionalneural network model, called \textbf{AttentionAnatomy}, can be jointly trainedwith three partially annotated datasets, segmenting OARs from whole body. Ourmain contributions are: 1) an attention module implicitly guided by body regionlabel to modulate the segmentation branch output; 2) a predictionre-calibration operation, exploiting prior information of the input images, tohandle partial-annotation(HPA) problem; 3) a new hybrid loss function combiningbatch Dice loss and spatially balanced focal loss to alleviate the organ sizeimbalance problem. Experimental results of our proposed framework presentedsignificant improvements in both S{\o}rensen-Dice coefficient (DSC) and 95\%Hausdorff distance compared to the baseline model.