Hybrid CNN-Swin Transformer for Glaucoma Screening in Fundus Images

In addressing the complexities of medical image analysis, our research leverages the AI Glaucoma Screening JustRAIGS dataset to tackle a significant challenge, which involved an accurate detection of the glaucoma presence in fundus photographs. So, we proposed a CNN-Swin Transformer model trained on the JustRAIGS dataset. The EfficientNet model is placed as the backbone of our proposed architecture to extract relevant feature maps from the input images to be passed to the Swin transformer model, which will learn dependencies between the extracted features. Our study achieves exceptional outcomes through an intricate methodology underscored by the application of the CLAHE modification histogram, boasting competitive auc and accuracy scores of 98 % and 98 % respectively, on the test dataset. Our performance in classification is achieved through a combination of advanced models, careful data preprocessing, and innovative augmentation strategies.