SNet-RA: Designing an Advanced Diabetic Retinopathy Detection Framework using Shuffle Network with a Region Attention Mechanism

The Diabetic Retinopathy (DR) is a retinal condition, that can harm the blood vessels in the eye and it is the main reason for vision impairment or blindness. The intricate structure of the eye makes manual identification of DR as laborious and prone to human mistakes. The DR substantially affects ophthalmic healthcare because of the high diagnostic costs and widespread prevalence. The DR cases are increased because of poor access to healthcare. Numerous automated methods have been designed to identify DR in fundus images. However, these methods are not very good at capturing the intricate characteristics of the DR. Hence, it is critical to resolve issues raised by the traditional DR detection paradigm. So, a novel deep learning-based DR detection framework is proposed in this work. The main novelty of the proposed work is the innovative combination of the Shuffle Network with Region Attention which reduces the computational time on the DR detection by focusing on the specific parts of the images. Next, the gathered images are passed to the DR detection phase. Here, the developed Shuffle Network with Region Attention (SNet-RA) mechanism is utilized to detect the DR. In this phase, a DR-detected outcome is obtained from the developed SNet-RA. Later, different experiments are executed to verify the efficiency of the suggested DR detection technique over multiple performance measures.