Skip to main content
. 2021 May 31;11(6):492. doi: 10.3390/jpm11060492

Figure 3.

Figure 3

3D view of a case from the UMCG head and neck dataset. (a–m) Ground truth, 2D U-Net, 2.5D U-Net, 3D U-Net, RUnetSeg, 2D SegU-Net, 2.5D SegU-Net, 3D U-Net, RSegUnetSeg, 2D Att U-Net, 2.5D Att U-Net, 3D Att U-Net, and RAttUnetSeg. We use cyan to indicate the correctly segmented mandible compared to the ground truth. Pink represents the regions that were missed by the algorithms, while yellow indicates the segmented regions that are not the mandible. The red circles indicate the coronoids of the mandibles that are often missed by the traditional EDCNNs, and the yellow circles indicate parts of the mandible body.