Related Experiment Video
Updated: Aug 29, 2025
08:20
Author Spotlight: AI-Driven Trypanosome Species Detection from Microscopic Images
Published on: October 27, 2023
1.6K
A New Approach for Detecting Fundus Lesions Using Image Processing and Deep Neural Network Architecture Based on YOLO
Carlos Santos1,2, Marilton Aguiar2, Daniel Welfer3
1Computer Center, Federal Institute of Education, Science and Technology Farroupilha, Alegrete 97555-000, Brazil.
Sensors (Basel, Switzerland)
|September 9, 2022
Summary
This study introduces a novel computational approach for early Diabetic Retinopathy detection. The method effectively identifies fundus lesions, improving diagnostic accuracy and patient outcomes.
Area of Science:
- Ophthalmology
- Computer Vision
- Medical Imaging
Background:
- Diabetic Retinopathy (DR) is a leading cause of vision loss, characterized by fundus lesions like microaneurysms and hemorrhages in early stages.
- Early detection of these lesions is crucial for preventing severe vision impairment and guiding treatment.
- Automated detection of DR lesions faces challenges including lesion variability, image quality issues, and difficulties in identifying small objects with deep learning.
Purpose of the Study:
- To develop and evaluate a novel computational approach for assisting in the medical diagnosis of Diabetic Retinopathy fundus lesions.
- To overcome challenges in automated DR lesion detection using advanced image processing and deep learning techniques.
Main Methods:
- The study employed a combination of image processing techniques, data augmentation, transfer learning, and deep neural networks.
- The proposed approach was implemented using the YOLOv5 model within the PyTorch framework.
- Training and testing were conducted on the public DDR and IDRiD Diabetic Retinopathy datasets.
Main Results:
- The proposed approach achieved an mAP of 0.2630 (IoU limit 0.5) and an F1-score of 0.3485 in the validation stage on the DDR dataset.
- In the test stage on the DDR dataset, the approach yielded an mAP of 0.1540 (IoU limit 0.5) and an F1-score of 0.2521.
- Experimental results indicate superior performance compared to existing methods for Diabetic Retinopathy lesion detection.
Conclusions:
- The developed computational approach shows promise for enhancing the early diagnosis of Diabetic Retinopathy.
- The integration of image processing, data augmentation, and deep learning (YOLOv5) effectively addresses challenges in automated fundus lesion detection.
- This method offers a valuable tool for DR screening and treatment planning, potentially improving patient vision outcomes.
