A multi model deep net with an explainable AI based framework for diabetic retinopathy segmentation and classification

Diabetic Retinopathy (DR) is a serious condition affecting diabetes people caused by hemorrhage in the light-sensitive retinal area. DR sufferers should receive urgent therapy to avoid vision loss. The intelligent medical diagnosis system for DR is evolving as a result of the inclusion of Artificial Intelligence (AI) technology. The current AI-driven DR diagnosis techniques are hampered by numerous serious challenges, which undermine their performance. Model performance is generally degraded by low contrast, inhomogeneous lighting, and noise, rendering accurate results hard to achieve. As a result, the proposed study has created an Adaptive Gabor Filter (AGF) based on the Chaotic Map to improve filtering performance. The multi-folded features like Local Binary Patterns (LBP), Speeded-Up Robust Features (SURF), and Texture Energy Measurement (TEM) are extracted and fed into classification phase. The classification phase has the Attention layer, the dense block of DenseNet, and Optimized Gated Recurrent Unit (OGRU) based on a Self-Adaptive Northern Goshawk Optimization (SANGO) algorithm for enhancing classification performance. The system was evaluated using three datasets: DiaRetDB1, APTOS 2019, and EyePacs, which demonstrated its robustness and reliability. Furthermore, the Grad Cam in the suggested technique assures the effective implementation of segmentation and classification performance. Performance is demonstrated by the use of Intersection over Union (IoU), Accuracy, Precision, Recall, F1-Measure, Dice Similarity Coefficient (DSC), and other metrics. In addition, the five-fold categorization is used to analyse the outcome performance. The suggested model achieved an accuracy of 99.01% on the DiaRetDB1 dataset, 98.98% on the APTOS 2019 dataset, and 99.12% on the EyePacs dataset.