Fig. 9

Performance of Multiple Models. A and B represent the accuracy and loss of the different segmentation networks on the training set. C and D represent the accuracy and loss of the different segmentation networks on the validation set. In the four graphs A, B, C, and D, we can infer that DeepHipp outperforms other algorithms. During the testing process, FCN and Unet_3D accuracy is almost zero. The results of SegNet are unstable, approximately 0.2. PSPNet and DeepHipp can reach 0.8315 and 0.8363 respectively. Graph E represents the change in learning rate during the training process. F shows the training accuracy using different loss functions of DeepHipp. Using the dice coefficient can achieve the best results