CNN for Parapapillary Atrophy Detection and Auxiliary Annotation in Complex Dataset

One of the primary approaches for diagnosing diseases such as high myopia and glaucoma is the detection of parapapillary atrophy. Computer-assisted detection of parapapillary atrophy has medical significance and can greatly increase the diagnostic efficiency of related diseases. This article is aiming at finding the optimal performance algorithm, analyzing the benefits of deep learning algorithms for parapapillary atrophy detection. Furthermore, we explore the model's usage in auxiliary correction labeling. Four convolutional neural network models were used in our experiments. We studied the performance of classification models on a small clean data set and a large complex data set. The best results were finally obtained using EfficientNetB5, and, accuracy rates of 98.28% and 91.88% were achieved for the two data sets respectively after optimization. Large-scale complex data sets are normal in real-world clinical applications, and the obtained performance proves the feasibility of using the algorithm in real-world situations. The advantages of this approach are demonstrated by comparing it with conventional methods The algorithm can also be applied to correct manual annotation and preliminary study shows its effectiveness.