Detection of Diabetic Retinopathy Using Deep Learning

Eyes are the main organs in the human body for sight and are among the most important sensory organs, using a large portion of brain processing power. Diabetes Mellitus (DM) has been a long-lasting condition that devastates the way the body processes glucose in various organs and tissues. DM is a chronic condition that eventually affects the body as a result of excess sugar in various part of the body. It has the potential to cause complications in microvascular areas, such as cerebral hemorrhage and coronary heart disease . The worrisome concern is retinopathy, which can lead to vision loss in severe cases. Due to the irreversible nature of the disease phases, timely screening and detection are vital. This research sets itself apart from previous studies by offering new insights and addressing important limitations in diabetic retinopathy (DR) detection. While many studies focus solely on the accuracy of their models, we took a broader approach. We evaluated our model using multiple metrics, such as F1 score, AUC, precision, recall, and sensitivity. This comprehensive evaluation ensures our model is not only accurate but also reliable in various real-world scenarios. To improve our model's performance, before deploying the pre-trained model, we perform data augmentation on the images in order to make the images ready and effective for training. We then used a pre-trained model called MobileNetV2; this model was already trained on a large dataset of images, giving it a strong starting point. We then fine-tuned this model specifically for DR detection, making it even more effective for medical diagnosis