Image Classification in Medical Diagnosis for Diabetic Retinopathy-Deep Learning

A major contributor to visual impairment, diabetic retinopathy (DR) necessitates early identification for successful treatment. Conventional CNNs, which are employed in DR diagnosis, sometimes have trouble with huge datasets, which results in overfitting and decreased accuracy. To overcome these constraints, a hybrid deep learning model that combines CNN, Inception V3, and VGG19 architectures is suggested. By enhancing the identification of microaneurysms, an early sign of DR, our approach improves feature extraction capabilities. Training and assessment are conducted using the APTOS 2019 dataset, which consists of retinal pictures classified into five groups. Techniques for augmenting data, such as different image reflections, improve resilience and avoid overfitting. When compared to current models, the hybrid model exhibits higher accuracy and provides more trustworthy predictions. The model's efficacy in early DR identification is demonstrated by improved performance measures including accuracy and recall, which may improve patient outcomes by enabling prompt diagnosis and care.