Enhanced Diabetes Detection with Deep Learning: An Iris Image Analysis Approach

Diabetes is a chronic and progressive condition that, if not diagnosed early, can lead to serious health complications, often we will see that patients learn they have diabetes only years after its emergence this points out the significance of precise and early identification. Thus, diabetes is a global health issue that requires timely detection. Traditional methods for diagnosing diabetes often involve expensive procedures, which can lead to delays in diagnosis. Using iris images for diabetes detection through deep learning algorithms offers a more accurate and cost-effective alternative. In general, a person with diabetes may experience changes in the iris, such as alterations in pattern, texture, or pigmentation. To identify these altered features, high-resolution observation is required. For this purpose, we use advanced imaging techniques and deep learning algorithms to analyze and interpret the changes in iris characteristics. CNN's, known for their powerful feature extraction capabilities, effectively captured and analyzed complex details in iris patterns. This approach utilizes several types of algorithms available in deep learning like Convolutional Neural Networks (CNNs), MobileNetV2, InceptionV3, ResNet50 and CNNs along with transformers. All these algorithms are widely known for their efficiency in image analyzing applications. Thus, we have chosen to implement the same procedural approach but for different algorithms. Finally, we had found an efficient algorithm in such a way that it gives the highest accuracy for the same dataset as compared with the other algorithms. It proved that CNNs are effective in identifying features and patterns in images and it gave the accuracy of about 93.91 % showing its effectiveness. Thus, ultimate aim of this approach is to provide an accurate, non-invasive and cost-effective way of diabetes disease diagnosis.