Related Experiment Video
Updated: Jul 25, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.3K
An automated unsupervised deep learning-based approach for diabetic retinopathy detection.
Huma Naz1, Rahul Nijhawan2, Neelu Jyothi Ahuja2
1Department of Computer Science, School of Computer Science, University of Petroleum and Energy Studies, Dehradun, 248007, India. huma.naz@ddn.upes.ac.in.
Medical & Biological Engineering & Computing
|October 23, 2022
Summary
This study introduces a novel hybrid deep learning approach for early diabetic retinopathy (DR) detection. The automated system achieves 98.6% accuracy in identifying retinal abnormalities, outperforming existing methods.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic retinopathy (DR) is a leading cause of blindness globally, affecting 35-60% of individuals with diabetes.
- Early detection of retinal abnormalities like microaneurysms is crucial for preventing vision loss.
Purpose of the Study:
- To develop an automated, hybrid unsupervised and deep learning technique for detecting diabetic retinopathy.
- To enhance the accuracy and generalizability of DR detection systems.
Main Methods:
- A modified fuzzy clustering method (MdFCM) was developed by revising k-means and fuzzy clustering.
- A modified convolutional neural network (CNN) was proposed, integrating MdFCM and extracted features.
- The algorithm was evaluated on three diverse datasets (DIARETDB1, APTOS, Liverpool) against existing clustering methods.
Main Results:
- The proposed hybrid algorithm achieved a high accuracy rate of 98.6% in detecting diabetic retinopathy.
- The system demonstrated superior performance compared to state-of-the-art algorithms across multiple datasets.
- The novel approach integrates unsupervised and deep learning for robust DR detection.
Conclusions:
- The developed hybrid unsupervised and deep learning system offers a promising tool for early diabetic retinopathy detection.
- This research pioneers a novel methodology for DR detection, combining clustering and deep learning.
- The system's high accuracy and generalizability suggest its potential for clinical application.
