Related Experiment Video
Updated: Jun 6, 2025
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
7.4K
Interpretable diagnostic system for multiocular diseases based on hybrid meta-heuristic feature selection.
Raveenthini M1, Lavanya R1, Raul Benitez2
1Department of Electronics and Communication Engineering, Amrita School of Engineering, Coimbatore, Amrita Vishwa Vidyapeetham, India.
Computers in Biology and Medicine
|November 30, 2024
Summary
A new computer-aided diagnosis framework accurately detects multiple common ocular diseases like Age-related Macular Degeneration (AMD) and Diabetic Retinopathy (DR) from fundus images. This explainable AI approach aids early detection and diagnosis, improving patient outcomes.
Area of Science:
- Ophthalmology and Medical Imaging
- Artificial Intelligence in Healthcare
- Machine Learning for Disease Detection
Background:
- Four common ocular diseases—Age-related Macular Degeneration (AMD), Cataract, Diabetic Retinopathy (DR), and Glaucoma—cause significant vision loss.
- Early detection is crucial for managing these conditions, but manual diagnosis is time-consuming and costly, especially for mass screenings.
- Computer-aided diagnosis (CAD) systems can assist ophthalmologists, with a generic framework for multiple diseases offering substantial benefits.
Purpose of the Study:
- To develop and evaluate a single, generic computer-aided diagnosis (CAD) framework for detecting multiple common ocular diseases.
- To utilize non-linear handcrafted features and advanced machine learning techniques for improved diagnostic accuracy.
- To incorporate explainable AI methods for enhanced model transparency and clinical trust.
Main Methods:
- Extraction of non-linear handcrafted features from fundus images.
- Hybrid feature selection using JAYA algorithm (JA) followed by Harris Hawks Optimization (HHO).
- Training an extreme gradient boosting (XGB) model for classification, with Shapley Additive Explanations (SHAP) for model interpretability.
Main Results:
- The proposed generic framework achieved high performance metrics: 93% accuracy, 91.3% sensitivity, 96.4% specificity, 90.4% precision, and 90.8% F1 score.
- This system is the first to use non-linear features within a generic framework for detecting multiple specified ocular diseases.
- SHAP analysis provided insights into feature impact, demonstrating the model's explainability and transparency.
Conclusions:
- The developed explainable AI model offers an efficient and transparent method for diagnosing multiple ocular diseases from fundus images.
- This approach can assist physicians in clinical practice, reducing subjectivity associated with manual assessments.
- The framework's ability to handle multiple diseases in a unified system represents a significant advancement in ocular disease diagnostics.
