Diabetic Retinopathy Detection Leveraging Convolutional Neural Network Architectures

Diabetic retinopathy (DR) is a prevalent cause of vision impairment and blindness worldwide, affecting millions annually due to damage to the small blood vessels in the retina from prolonged high blood sugar levels. In 2020, WHO reports that the number of adults worldwide with DR was estimated to be 103.12 million; by 2045, the numbers are projected to increase to 160. 50 million. Early detection is vital to prevent vision loss, and leveraging deep learning, particularly convolutional neural networks (CNNs), has shown promise in automating detection. However, to improve effectiveness, personalized approaches are needed due to variability in retinal images. Technologies like fundus photography and OCT (Optical Coherence Tomography) aid in detection, and CNNs analyse images for signs of DR, aiding in early diagnosis. The PrecisionEye project proposes fine-tuning CNNs to adapt to individual patient characteristics, enhancing performance, and uses transfer learning approach. Integration of patient-specific clinical information are employed, resulting in improved accuracy, sensitivity, and specificity compared to standard CNN approaches. This framework shows efficiency in advancing the automation of DR diagnosis and personalized healthcare interventions for individuals at risk of vision loss due to diabetes.