Fig. 4.

Process of fine-tuning VGG16. The images were preprocessed and split into training, validation, and test sets. Then, we performed 4-fold cross-validation, fine-tuned the VGG16 with each training sample, and evaluated the performance using receiver operating characteristic curves