Abstract:Autosegmentation of the knee joint in magnetic resonance (MR) images is significant for clinical requirements. However, it is challenging due to that the segmentation targets have dramatically different sizes. In this study, an endtoend DRD UNet is proposed, which is based on the deep learningframework. Theresidualmodule isused asthebasic modulein theUNetmodel, whichincreasestheabilityof reusingfeature maps. Theparalleldilated convolution modulesareusedtoachieve differentreceptivefields,which can overcomethe limitations of single receptive field in the UNet model and effectively improve the segmentation capability with targets of different sizes. The multioutput fusion deep supervision module is designed to directly utilize the feature maps of different levels. In this way, the information complementarity is obtained, the consistency and accuracy of the segmented regions are improved. The proposed algorithm is evaluated by using the public OAIZIB data set. The average segmented surface distance is 02 mm, the root mean square surface distance is 043 mm, the Hausdorff distance is 522 mm, the average dice similarity coefficient (DSC) is 9305%, and the volume overlap error is 386%. Compared with the conventional UNet and other currently available models, the proposed DRD UNet has better segmentation accuracy.