An efficient adaptive histogram based segmentation and extraction model for the classification of severities on diabetic retinopathy

Diabetic retinopathy (DR) is an important retinal disease, which occurs commonly among diabetic patients. This disease severely injures the basic vision of the eye and results in blindness in several cases, which could be eliminated by earlier detection and medication. The existence of many classes in DR makes the diagnosis process difficult. To resolve this process, this paper introduces a new segmentation based classification model to classify the DR images effectively. The proposed model involves three main processes, namely, preprocessing, segmentation, feature selection and classification. The proposed method undergoes preprocessing and contrast-limited adaptive histogram equalization (CLAHE) model is applied for segmentation. AlexNet architecture is applied as a feature extractor to extract the useful set of feature vectors. Finally, softmax layer is utilized to classify the images into different stages of DR. The validation takes place using the publicly available Kaggle dataset. The experimental outcome indicates that the presented model achieves maximum classification rate with an accuracy of 95.86%, sensitivity of 92.00%, and specificity of 97.86% respectively.