Fig. 1From: Performance of a 3D convolutional neural network in the detection of hypoperfusion at CT pulmonary angiography in patients with chronic pulmonary embolism: a feasibility studyThe convolutional neural network architecture was based on U-net with three max-pooling steps in the downsampling pathway and linear upsampling in the resolution recovery path. Convolutional layers (C) were followed by batch or group normalisation and exponential linear unit (ELU) activations. Skip connections with cropping were used between the matching resolution levels (dashed arrows). C Convolutional layer, BN Batch normalisation, ELU Exponential linear unit, GN Group normalisationBack to article page