Fig. 5: Validation loss per epoch for benchmarking six deep-learning architectures.

Validation loss (y axis) versus number of epochs (x axis) for six deep-learning algorithms (VGG-11, VGG-16, VGG-19, ResNet-34, ResNet-50, and ResNet-101) during model benchmarking. The plot demonstrates the performance of each algorithm in terms of loss minimization, with the objective of identifying the model with the lowest validation loss for the given task of classifying knee OA cases and controls. The same training (n = 436) and validation set (n = 110) were used for model benchmarking as was used to train DL-binary.