DCNMST: A Deep Contrastive Network With Multiple Self-Supervised Tasks for Diabetic Retinopathy Grading Classification in Internet of Medical Things

Diabetic Retinopathy (DR) is a leading cause of vision loss, where early and accurate diagnosis is essential for effective treatment. In this study, we propose a novel Deep Contrastive Network with Multiple Self-Supervised Tasks (DCNMST) for DR severity classification within an Internet of Medical Things (IoMT) framework. The proposed model integrates reconstruction and denoising tasks to extract robust and generalizable features from unlabeled retinal images. Furthermore, a contrastive learning module is employed to enhance the discriminative power of the learned representations by aligning positive samples and separating negative ones in the feature space. Extensive experiments on real-world retinal datasets demonstrate that DCNMST achieves a precision of 0.948, recall of 0.948, and accuracy of 0.949, consistently outperforming state-of-the-art models such as ViT, ConvNeXt, and ResNet. These results confirm the effectiveness of our approach in improving DR diagnosis under limited labeled data conditions.