Machine Learning Based Automated Segmentation and Hybrid Feature Analysis for Diabetic Retinopathy Classification...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Diabetic retinopathy (DR) is a leading cause of blindness globally.
• Accurate and early detection of DR is crucial for effective treatment.
• Current diagnostic methods can be time-consuming and subjective.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) for segmenting and classifying diabetic retinopathy (DR).
• To develop and validate a novel automated framework for DR analysis using retinal fundus (RF) images.

Main Methods:

• Utilized a dataset of 2500 RF images from 500 patients with varying DR stages.
• Developed a clustering-based automated region growing framework for image segmentation.
• Extracted texture features (histogram, wavelet, co-occurrence matrix, run-length matrix) and applied data fusion.
• Employed feature selection techniques and deployed five ML classifiers (SMO, Lg, MLP, LMT, SLg) with 10-fold cross-validation.

Main Results:

• Initial ML classifiers achieved accuracies ranging from 77.67% to 96.33%.
• After feature fusion and selection, classification accuracies significantly improved.
• The best performing ML classifiers (LMT, SLg) achieved accuracies of 99.73% on optimized features.

Conclusions:

• Machine learning methods demonstrate high potential for accurate diabetic retinopathy classification.
• The proposed automated framework and feature optimization techniques enhance diagnostic capabilities.
• This approach can aid in early and reliable detection of DR, potentially preventing vision loss.