Towards Automated Keratoconus Screening Approach using Lateral Segment Photographed Images

Keratoconus (KC) is a common noninflammatory ocular disorder affecting mostly the younger generation of age 30 and below. Typically, KC patients will have symptoms of bulging eye cornea resulting from the conical displacement either heading outwards or downwards. This condition is a serious matter since it can affect the affected person's visual ability. As such, this paper intends to describe the developmental work involving the use of the pre-trained VGGNet-16 model and a conventional convolutional neural network to detect KC automatically. This experiment uses a total of 2000 KC, and 2000 healthy lateral segment photographed images (LSPI) extracted from videos captured from the side view of 125 patients using a smartphone camera. Three hyperparameters are fine-tuned during the training phase to generate the best model. From the results, we observe that the VGGNet-16 model with a learning rate of 0.0001, batch size of 16, and the epoch number of 70 is the best model. The proposed pre-trained model achieves an accuracy of 95.75%, the sensitivity of 92.25%, and specificity of 99.25%. In conclusion, the study proves that the proposed VGGNet-16 model which is lightweight and has compact architecture is suitable for the automated KC screening approach using LSPI.