Fig. 1From: Convolutional neural network-based automatic heart segmentation and quantitation in 123I-metaiodobenzylguanidine SPECT imagingArchitecture of CNN used to segment lungs and liver. Convolution layers do not use padding. Input shape to network is 72 × 72 × 72 pixel cube; output shape is 8 × 8 × 8 pixel cube.Back to article page