Diagnosis of Diabetic Retinopathy Using Hierarchical Feature Learning and Modified CNN with ACAF

Diabetic retinopathy (DR) remains one of the leading causes of vision loss and blindness today, making the need for accurate and reliable detection techniques paramount. This work presents a new method for feature extraction through combining two methods namely; vSLM and Wavelet CNN. The approach is complemented by a modified CNN model that is equipped with an AdaptiveConvolutional ActivationFunction (ACAF) for classification. As it was mentioned before, the vSLM technique creates detailed retinal maps, and thus can be effective for the detection of critical DR indicators such as microaneurysms, haemorrhages, and exudates. Wavelet CNN enhances the feature extraction procedure, taking into consideration both spatial and spectral features of the retinal images. The resolution provided by the proposed CNN model with ACAF demonstrates outstanding classification results with accuracy rates of 97.12% on DiaretDB0 dataset, 99. 12% on the public EyePACS Kaggle dataset and 99.15% on the DRIVE dataset and 98.12% on the CHASE dataset. These findings highlight the importance of using the feature extraction and classification approach for improving the accuracy of DR detection and classification. The findings point to severe advancements in the diagnostic tools of DR and the effect of these enhancements in the pragmatic aspect of identify and handling of DR.