Figure 2.

The deep learning model architecture. An input image of 512 × 512 is fed as input to the model (gray). The model used in this study was inspired by U-Net and has two main parts; the encoder (green) and the decoder (red). In the encoder, starting from the initial number of filters as 16, the number of filters was doubled after each subsequent convolutional block followed by a max pooling operation. The decoder part restores the dimensionality of the segmentation mask by concatenating the features from the encoder by application of transposed convolutional operations followed by convolutions, batch normalization and ReLu operations. The segmentation mask of the same resolution is obtained as the model output (blue). Abbreviations: LAD = left anterior descending, LCx = left circumflex, RCA = right coronary artery.