Detection of Diabetic Retinopathy (DR) Severity from Fundus Photographs using Conv-ViT

In recent years, there has been a significant rise in the global prevalence of diabetes, affecting individuals across all age groups. Among those with longstanding diabetes, a common complication known as Diabetic Retinopathy (DR) poses a substantial risk, leading to vision impairment. DR stands as a major concern due to its widespread impact on individuals with diabetes worldwide. Traditionally, the assessment of DR severity has relied on manual evaluation by ophthalmologists using Fundus photography-based retina images, a process prone to time constraints and inconsistency. This paper proposes an automated approach for categorizing the severity stages of DR, aiming to alleviate the limitations of manual diagnosis. Our methodology employs a hybrid feature extraction technique, integrating Convolutional and Vision Transformer models, to infer the severity of DR from fundus photographs. The experiments are conducted using the publicly available APTOS-2019 blindness detection dataset, facilitating comprehensive evaluation and validation of our proposed method.