Lesion Segmentation in Retinal Images for Diabetic Retinopathy Diagnosis Using Weighted Gradient Assistance

For screening and grading diabetic retinopathy to be reliable, a computer-aided system must accurately detect lesions from retinal fundus photographs.This study presents a novel and reliable approach to automatically identifying three primary anomalies in retinal images: microaneurysms (MA), hemorrhages (HM), and hard exudates (HE).The key innovation of this method lies in the use of weighted gradients at multiple scales and orientations, combined with a hybrid feature set that integrates both intensity and shape features to precisely differentiate between lesion types.The proposed method begins with image pre-processing to enhance the quality of retinal photographs.Next, weighted gradients assistance (WGA) is computed using a combination of the Robinson Compass Mask (for handling multiple orientations) and the Gaussian Filter (for managing different scales).This step significantly reduces false positives by accurately identifying potential lesion regions.A total of 17 features (5 intensity-based and 12 shapebased) are extracted from each candidate region, enabling a comprehensive representation of the lesions.Finally, a Random Forest classifier is employed to predict the type of lesion (MA, HM, or HE) based on these features.The effectiveness of the proposed method is evaluated using the IDRiD and DIARETDB1 benchmark datasets.The results demonstrate the method's ability to achieve high sensitivity while maintaining a low rate of false positives.Specifically, the method achieved an average sensitivity of 0.6200 on the IDRiD dataset and 0.9612 on the DIARETDB1 dataset.The novelty of this approach lies in its weighted gradient mechanism, which improves lesion detection accuracy, and its hybrid feature set, which combines intensity and shape characteristics to enhance classification performance.This method offers a robust and efficient solution for automated diabetic retinopathy screening, reducing the workload of clinicians and improving diagnostic consistency.