Secure and Trustworthy AI for Efficient Diabetic Retinopathy Screening with Deep Learning Model

It is typically known that image processing is a kind of signal processing from where the input is the image, which may be a photo or even a video and the output, may be the parameters, which are the characteristics of the input image. Further, this image process technique allows various types of algorithms, which are of a wider range to be solicited to the data, which the input and supports in avoiding many problems is including the removal of noise, promoting enhanced future extraction, etc… In the technique of image processing, some of the artificial intelligent-based learning models, especially the deep learning-based models play a vital role, the deep planning models have a significant impact on the image processing techniques. It helps scientists to get rid of the main complications in object detection, segmenting the images, and image classification. When compared with the other traditional models, the detection process in deep learning is done more precisely as the deep learning algorithms identify and locate the things to be detected in the image more exactly. The proposed study deals with the detection of diabetic retinopathy, which is known as the illness of the eye caused by uncontrollable blood sugar levels. If this is untreated, it can lead to damage to the retina, which results in permanent blindness. Some of the deep learning techniques play a crucial role in the detection of diabetic retinopathy. These techniques detect the disease earlier, which helps in saving the patients from blindness. This proposed study uses a Swin transformer with a dense capsule network (Swin-CapsNet) for detecting and curing diabetic retinopathy earlier, as it helps protect the eye which serves as the most important part of the human body. Additionally, the efficiency of the work is enhanced by comparing the proposed Swin transformer with a dense capsule network (Swin-CapsNet) with the other existing models such as ResNet-50, VGG18, and BI-LSTM with transformer for better understanding. After the comparison, the proposed model is verified by using the evaluation metrics namely, accuracy precision, recall, and F1 score. Moreover, the work of the proposed model is proven significantly, as it has achieved the highest range of accuracy of about 96.5%, a precision of about 95.7%, and recall of about 96.1% in the detection of vision-threatening disease, known as diabetic retinopathy earlier, which crucially saves thousands of people from the vision deficiency.