Fig. 4From: Preoperative prediction of lymph node metastasis using deep learning-based featuresSchematic representation of adapted VGG-16 network for DLB feature extraction. Each layer is defined by corresponding color legend at the bottom of the figure. Input was a 3D volume for a single slice of the three kinetic maps. Features were extracted from the last fully connected layer (yellow), resulting in a feature vector of 1000 features for each 2D slice. Conv: Convolutional layer; ReLu: Rectified linear unitBack to article page