Swin-MMC: Swin-Based Model for Myopic Maculopathy Classification in Fundus Images

Myopic maculopathy is a highly myopic retinal disorder that often occurs in highly myopic patients, serving as a major cause of visual impairment and blindness in numerous countries. Currently, fundus images serve as a prevalent diagnostic tool for myopic maculopathy. However, its efficacy relies on the expertise of clinicians, making the process labor-intensive. Thus, we propose a model specifically designed for the image classification of myopic maculopathy, named Swin-MMC, based on the Swin Transformer model architecture, which achieves outstanding performance on the test dataset. To achieve a finer-grained classification of myopic maculopathy in fundus images, we have innovatively and for the first time proposed the use of enhanced ArcFace loss in medical image classification. Then, based on the Swin-MMC model, we introduce a weak label strategy that effectively mitigates overfitting. Our approach achieves significantly improved results on the test dataset and can be easily used for various datasets and classification tasks. We conduct a series of experiments in the MMAC2023 challenge. In the testing phase, our average performance metric reaches 86.60%. In the further testing phase, our model's performance improves to 88.23%, ultimately securing the championship in the MMAC2023 challenge. The codes allowing replication of this study have been made publicly available at https://github.com/LuliDreamAI/MICCAI_TASK1 .