Figure 2.

Schematic representation of the cross-validation. This figure shows how all images were divided for training, validation, testing, and concordance. First, 10,202 glomerular images were collected from the whole slide images (WSIs). Next, 100 glomerular images were chosen for the concordance data. Finally, the remaining 10,102 images were divided into 5 groups, 1 for the test data and the other 4 groups as the training data and validation data that were used for convolutional neural network training. ROC−AUC, receiver operating characteristic−area under the curve.