Research on classification method of high myopic maculopathy based on retinal fundus images and optimized ALFA-Mix active learning algorithm

AIM: To conduct a classification study of high myopic maculopathy (HMM) using limited datasets, including tessellated fundus, diffuse chorioretinal atrophy, patchy and macular minimize annotation costs, to optimize the ALFA-Mix active learning algorithm apply it HMM classification. METHODS: The optimized (ALFA-Mix+) was compared with five algorithms, ALFA-Mix. Four models, ResNet18, were established. Each combined four models for experiments on dataset. experiment consisted 20 rounds, 100 images selected per round. evaluated by comparing number rounds in which ALFA-Mix+ outperformed other algorithms. Finally, this employed six EfficientFormer, classify HMM. best-performing model among these as baseline achieve satisfactory results small RESULTS: outperforms algorithms an average superiority 16.6, 14.75, 16.8, 16.7 terms accuracy, sensitivity, specificity, Kappa value, respectively. This conducted classifying several advanced deep complete training set 4252 images. EfficientFormer achieved best value 0.8821, 0.8334, 0.9693, 0.8339, Therefore, combining 0.8964, 0.8643, 0.9721, 0.8537, CONCLUSION: reduces required samples without compromising accuracy. Compared more experiments. It effectively selects valuable In classification, enhances performance, further demonstrating effectiveness ALFA-Mix+.