Fig. 1.

CNN architecture for kidney function estimation based on kidney sonographic images. a Proposed neural network architecture included 33 residual blocks (100 convolution layers in total) as a CNN-based feature extractor and three fully connected layers of 512, 512, and 256 neurons as a regressor for eGFR prediction. Feature maps are colored in blue and the regressor is specified in yellow. The dropout probability was set at 0.5. (b) Components of the first residual block in the CNN