FIGURE 1.
Study design. Patient underwent magnetic resonance imaging (MRI) before surgery. Slices with tumor were selected and segmented as rectangular region of interest (ROI) after images format conversion. Then, data were split into training, validation and test dataset. The data used in five-fold cross-validation were augmented 14 times, and keep the test dataset unchanged. Finally, AlexNet and GoogLeNet were explored and evaluated by training from scratch and fine-tuning from pre-trained model.