Pelvis segmentation using multi-pass U-Net and iterative shape estimation

In this report, an automatic method for segmentation of the pelvis in three-dimensional (3D) computed tomography (CT) images is proposed. The method is based on a 3D U-net which has as input the 3D CT image and estimated volumetric shape models of the targeted structures and which returns the probability maps of each structure. During training, the 3D U-net is initially trained using blank shape context inputs to generate the segmentation masks, i.e. relying only on the image channel of the input. The preliminary segmentation results are used to estimate a new shape model, which is then fed to the same network again, with the input images. With the additional shape context information, the U-net is trained again to generate better segmentation results. During the testing phase, the input image is fed through the same 3D U-net multiple times, first with blank shape context channels and then with iteratively re-estimated shape models. Preliminary results show that the proposed multi-pass U-net with iterative shape estimation outperforms both 2D and 3D conventional U-nets without the shape model.