Enhancing Mild Diabetic Retinopathy Detection: A Comparative Study of CLAHE-Preprocessed and Unprocessed Fundus Images Using a Minimal CNN Model

Abstract

Early identification of Diabetic Retinopathy (DR) is crucial for preserving vision in individuals with diabetes, as this condition is a leading cause of sight loss among diabetic patients. Timely detection enables effective treatment interventions. Hence, we proposed a Minimal Convolutional Neural Network (MCNN) model for detecting mild DR symptoms using fundus images. Utilizing publicly available datasets from Kaggle and Messidor, the research applies Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing to enhance image quality. The MCNN is then trained on both CLAHE-processed and unprocessed versions of the same images. The research evaluates CLAHE preprocessing's effect on mild DR detection by analyzing the model's performance across both datasets, seeking to quantify any accuracy improvements. This approach leverages modern machine learning techniques to potentially improve early diagnosis of DR, addressing a critical need in ophthalmological practice and diabetic care.