Related Experiment Video
Updated: Jan 10, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
999
LightMG-Net: an efficient lightweight deep neural network for multiclass grading of retinal detachment using
Sonal Yadav1, R Murugan1, Balachandra Pattanaik2,3
1Bio-Medical Imaging Laboratory (BIOMIL), Department of Electronics and Communication Engineering, National Institute of Technology Silchar, Silchar, Assam, 788010, India.
Scientific Reports
|November 26, 2025
Summary
Early diagnosis of retinal detachment is crucial for vision recovery. A new AI model, LightMG-Net, accurately grades retinal detachment from fundus images, improving early detection and treatment success rates.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Retinal detachment is a significant cause of visual impairment, particularly in older adults.
- Early diagnosis and classification of retinal detachment are critical for successful vision restoration, with nearly 90% recovery rates if treated promptly.
- Accurate grading of retinal detachment is essential for timely and effective clinical intervention.
Purpose of the Study:
- To develop and validate a novel, optimized, lightweight multiclass model for grading retinal detachment from fundus images.
- To enhance the accuracy and efficiency of early-stage retinal detachment classification using artificial intelligence.
- To introduce the LightMG-Net model, incorporating handcrafted features and optimized deep learning for improved diagnostic capabilities.
Main Methods:
- Development of the LightMG-Net model, a lightweight convolutional neural network.
- Integration of image and feature-oriented handcrafted techniques for comprehensive feature analysis.
- Application of the Grey Wolf Optimization technique for automatic hyperparameter tuning of the neural network.
- Validation of the model on four public datasets: Retinal Image Bank, Cataract Image Dataset, Kaggle, and Eye Disease Retinal Image.
Main Results:
- The LightMG-Net model achieved high performance metrics: 95.42% classification accuracy, 95.10% sensitivity, 98.90% specificity, and an area under the curve of 0.9947.
- The model demonstrated superior performance compared to existing baseline methodologies in multiclass grading of retinal detachment.
- Experimental results confirm the effectiveness of the proposed approach in classifying retinal detachment from fundus images.
Conclusions:
- The developed LightMG-Net model offers a highly accurate and efficient solution for the multiclass grading of retinal detachment.
- Early and accurate classification of retinal detachment using AI can significantly improve patient outcomes and prevent vision loss.
- The proposed method shows promise for clinical application in aiding ophthalmologists with the diagnosis and management of retinal detachment.
