Diabetes can cause diabetic retinopathy (DR), an eye condition that can ultimately lead to blindness. The DR is rising as a result of the rising prevalence of diabetes worldwide. Routine eye check-ups at the hospital are suggested to maintain the health of a diabetic eye. The greatest method to prevent complications with DR is early diagnosis. Therefore, the availability of trustworthy screening techniques that are simple to use is essential for the early detection of DR. The main goal of this study is to develop a compact machine-learning (ML) DR screening tool that delivers accurate results without the use of a powerful computer or specialized software. A secondary goal is to evaluate the usability of smartphone-based fundus imaging as a low-cost alternative for fundus image acquisition. Using MATLAB software and the GLCM features derived from 560 previously recorded fundus images, we constructed a DR classifier that had a 96.45 % training accuracy and a 95.99% test accuracy. A user can feed fundus photos received from a smartphone into the user-friendly graphical user interface (GUI) we created to do early screening in less than a second. We expect the classifier to work much better if we combine the suggested model with a large number of fundus images taken with the image-capturing device.
Diabetes can cause diabetic retinopathy (DR), an eye condition that can ultimately lead to blindness. The DR is rising as a result of the rising prevalence of diabetes worldwide. Routine eye check-ups at the hospital are suggested to maintain the health of a diabetic eye. The greatest method to prevent complications with DR is early diagnosis. Therefore, the availability of trustworthy screening techniques that are simple to use is essential for the early detection of DR. The main goal of this study is to develop a compact machine-learning (ML) DR screening tool that delivers accurate results without the use of a powerful computer or specialized software. A secondary goal is to evaluate the usability of smartphone-based fundus imaging as a low-cost alternative for fundus image acquisition. Using MATLAB software and the GLCM features derived from 560 previously recorded fundus images, we constructed a DR classifier that had a 96.45 % training accuracy and a 95.99% test accuracy. A user can feed fundus photos received from a smartphone into the user-friendly graphical user interface (GUI) we created to do early screening in less than a second. We expect the classifier to work much better if we combine the suggested model with a large number of fundus images taken with the image-capturing device.