Diabetes has become a widespread disease affecting approximately 382 million people globally, with projections estimating a rise to 592 million by 2025. One of the significant complications associated with diabetes is Diabetic Retinopathy (DR), a condition that damages the blood vessels in the retina of the eyes. It affects around 34.6% of individuals with diabetes worldwide. In the Khyber Pakhtunkhwa (KP) province of Pakistan, DR causes blindness in 4% of the population, with type I diabetes being the most prevalent cause. In this region, 30% of the population has diabetes, with 1.6% of them having type II diabetes. Among diabetic patients, approximately 30% have DR, and nearly 2% of these patients have developed complete blindness. Early detection of DR is essential to prevent or minimize its severe consequences.
The objective of this project is to develop a robust and efficient DR detection system using advanced deep learning techniques, with a particular focus on the Efficient-Net model. The early diagnosis and accurate classification of DR can significantly contribute to the management and prevention of vision loss in diabetic patients.
The retina, consisting of sensitive nerve cells, is impacted by diabetes, and DR can be categorized based on its severity. Non-Proliferative Diabetic Retinopathy (NPDR) can cause severe deterioration of diabetic patients' eye health. Initial symptoms include the appearance of tiny, round, and red-colored spots known as microaneurysms (MAs). Hemorrhages (H) may occur when blood vessels or MAs weaken and release blood, appearing as dot-like or blot-like red lesions. Hemorrhages and MAs are sometimes grouped together as HMA. Early detection of DR is challenging as it is often asymptomatic.
The proposed method for DR detection utilizes preprocessing, feature extraction, and machine learning techniques, with a specific focus on deep learning models like the Efficient-Net model. Deep learning has shown superior performance compared to traditional classifiers.
Among the evaluated models, the deep learning technique using the Efficient-Net model demonstrated outstanding performance, achieving an impressive F-score of 90.90. This result surpassed other classifiers, including SVM (F-score of 57.52), K-NN (F-score of 50.12), Naïve Bayes (F-score of 45.10), and Decision Tree (F-score of 42.85). This highlights the effectiveness of deep learning in accurately classifying diabetic retinopathy.
The implementation of an Efficient-Net-based deep learning model provides a promising approach for the early detection and management of diabetic retinopathy. Utilizing this system could help reduce the likelihood of diabetic patients losing their eyesight due to DR. Future research can explore other deep learning models or combinations of techniques to further enhance the accuracy and efficiency of diabetic retinopathy detection.
By leveraging advanced deep learning techniques, this project aims to make a valuable contribution to the medical field, specifically in diagnosing diabetic retinopathy at an early stage and providing better care to those affected.