Dynamic Regulation of Level Set Parameters Using 3D Convolutional Neural Network for Liver Tumor Segmentation

Figure 3

Our baseline 3D CNN consists of four layers with kernels for feature extraction, leading to a receptive field of size . The classification layer is implemented as convolutional with 2 fully connected layers, which enable dense-inference. When the network is fed an input of , it predicts three classes simultaneously, one for each shift of its receptive field over the input. Number of FMs, convolutional filters, and their sizes are depicted as , respectively.