Automatic classification for corneal ulcer using a modified VGG network

Deep learning techniques are more and more used for automatic classification of corneal ulcer. A modified VGG framework for corneal ulcer classification was proposed to realize feature fusion by observing feature graphs in this work. Before the modified VGG model training, the data preprocessing methods of masks, AHE, normalization and data augmentation to corneal ulcer images were done. The weighted categorical cross entropy was selected as the loss function during the modified model training. Experiments showed that the modified VGG network had fewer parameters and better performance compared with the traditional CNN network. Finally, the modified VGG network achieved classification accuracy of 88.89%, sensitivity of 71.93%, precision of 92.27%, and F1 score of 71.39%.