AHO-MLCNN: archerfish hunting optimisation based modified lightweight CNN for diabetic retinopathy detection

Abstract

ABSTRACT This paper presents a novel automated model for the detection of diabetic retinopathy (DR), a common retinal disease caused by high blood sugar levels. Early detection of DR is crucial in preventing severe complications. The proposed model focuses on effectively classifying retinopathy and non-retinopathy cases using two fundus image datasets: DIARETDB0 and IDRiD. The DR detection module consists of two phases: preprocessing and classification. In the preprocessing phase, tasks such as resizing, normalization, and denoising are performed to enhance the accuracy of the classifier. The preprocessed fundus images are then fed into the proposed modified residual block lightweight CNN-based archerfish hunting optimizer (MRLCNN-AHO) approach, which accurately detects and classifies normal and abnormal cases. The experimental results demonstrate the efficiency of the proposed MRLCNN-AHO approach. It achieves an accuracy rate of approximately 97.8% and 97.5% for DIARETDB0 and IDRiD datasets, respectively. These results are compared with various existing methods, validating the effectiveness of the proposed approach. Automated DR detection models like the one proposed in this paper contribute to reducing processing time, cost, and effort associated with manual diagnosis. Early identification of mild-stage DR can significantly improve patient outcomes by enabling timely interventions and preventing the progression of the disease.