Deep Learning for Diabetic Retinopathy with Color Fundus Image Analysis in AI for Medical Imaging

Abstract

This work analyzes clinical photos of diabetic retinopathy (DR), a common and preventable cause of vision impairment, using deep-learning methods. We have analyzed the computational efficiency of several concatenated CNN models, including DenseNet201, ResNet50, VGG19, and MobileNet V2. ResNet50 has a training time of 15.37 seconds, but MobileNet V2 has 78.22% validation accuracy. This shows a trade-off between accuracy and efficiency. MobileNet V2 completes a 5-fold cross-validation process with 77.4% accuracy and predicts DR based on AUC. The IRI clustering-based technique can locate damaged retinal areas almost 100% reliably. These results provide several relevant insights for healthcare professionals and AI researchers in DR detection. The future research, known as OmniVec, offers a highly accurate multimodal deep-learning model for image processing. OmniVec sets new records on the ImageNet benchmark and outperforms existing bests in visual, audio, and text domains and 3D data. During pre-training, OmniVec enables existing modalities to jointly exchange information, which in turn boosts performance on a wide range of benchmarks, and therefore it may be a beneficial model for other multimodal works in the future.