Early detection and diagnosis are critical for effectively treating Diabetic retinopathy (DR), a severe vision-threatening diabetes-related challenge. We introduced an innovative technique that employed algorithms for deep learning for the automatic identification of DR. The significance of the proposed model lies in its capacity to rapidly and accurately diagnose DR, enabling prompt medical intervention to prevent visual impairment. Here we implemented multiple pre-processing techniques, including Top-hat filtering, median filtering, CLAHE, and Gaussian filtering. These techniques notably improved the accuracy diabetic retinopathy detection, making a contribution to the medical image analysis field. The performance evaluation conducted on the dataset APTOS 2019 has yielded results regarding accuracy, sensitivity and also specificity. These findings highlight the efficiency of our technique in world applications for DR detection. For our experimentation we utilized the APTOS 2019 dataset consisting of 1299 image files for DR training and 279 image files, for DR testing.
Paper
Full text
Detection of diabetic retinopathy using deep learning methods
Semantic Scholar · Medicine · 2024
Abstract
Early detection and diagnosis are critical for effectively treating Diabetic retinopathy (DR), a severe vision-threatening diabetes-related challenge. We introduced an innovative technique that employed algorithms for deep learning for the automatic identification of DR. The significance of the proposed model lies in its capacity to rapidly and accurately diagnose DR, enabling prompt medical intervention to prevent visual impairment. Here we implemented multiple pre-processing techniques, including Top-hat filtering, median filtering, CLAHE, and Gaussian filtering. These techniques notably improved the accuracy diabetic retinopathy detection, making a contribution to the medical image analysis field. The performance evaluation conducted on the dataset APTOS 2019 has yielded results regarding accuracy, sensitivity and also specificity. These findings highlight the efficiency of our technique in world applications for DR detection. For our experimentation we utilized the APTOS 2019 dataset consisting of 1299 image files for DR training and 279 image files, for DR testing.