Automated classification of patients with coronary artery disease using grayscale features from left ventricle...

Area of Science:

• Medical imaging analysis
• Cardiovascular disease diagnostics
• Machine learning in healthcare

Background:

• Coronary Artery Disease (CAD) poses a significant mortality risk.
• Manual interpretation of echocardiography for CAD is prone to variability and errors.
• Automated detection methods are needed for efficient and reliable CAD diagnosis.

Purpose of the Study:

• To develop and present a computer-based data mining technique for classifying normal and CAD-affected echocardiography images.
• To reduce inter-observer variability and interpretation errors in CAD detection.
• To create an automated system for easier CAD diagnosis in clinical settings.

Main Methods:

• Extraction of multiple grayscale features (fractal dimension, spectral entropies, texture, LBP, wavelet) from 800 echocardiography images (400 normal, 400 CAD).
• Feature selection using t-test to identify discriminating capabilities.
• Evaluation of various supervised classifiers with feature combinations.
• Development of a novel HeartIndex for objective image classification.

Main Results:

• The Gaussian Mixture Model (GMM) classifier, using nine selected features, achieved 100% accuracy, sensitivity, specificity, and positive predictive value.
• The developed HeartIndex provides a single, highly discriminative numerical value for automated CAD classification.
• The proposed method demonstrates superior performance in differentiating normal and CAD cases.

Conclusions:

• The developed computer-based technique effectively classifies normal and CAD cases from echocardiography with high accuracy.
• The novel HeartIndex facilitates automated and objective CAD detection, improving clinical workflow.
• This approach holds promise for widespread implementation in hospitals and clinics for early CAD diagnosis.