Improving Automatic Coronary Stenosis Classification Using a Hybrid Metaheuristic with Diversity Control

Area of Science:

• Computational intelligence
• Medical image analysis
• Feature selection

Background:

• Traditional evolutionary computing methods may converge prematurely to local optima.
• The proposed strategy incorporates worst-case individuals to enhance search space exploration.
• Explicit diversity control prevents convergence issues in feature selection.

Purpose of the Study:

• To introduce a novel Hybrid Metaheuristic with explicit diversity control.
• To identify an optimal feature subset for coronary stenosis classification.
• To improve the robustness and accuracy of feature selection algorithms.

Main Methods:

• Utilized a dataset of 608 coronary stenosis images (positive and negative cases).
• Extracted 473 features (intensity, texture, morphological) from images.
• Employed a Support Vector Machine classifier with Accuracy and Jaccard Coefficient metrics.

Main Results:

• Achieved 0.92 Accuracy and 0.85 Jaccard Coefficient.
• Identified a 16-feature subset, achieving 0.97 discrimination from 473 initial features.
• Demonstrated superior classification performance compared to existing literature.

Conclusions:

• The Hybrid Metaheuristic with diversity control significantly improved coronary stenosis classification.
• The selected 16-feature subset shows promise for clinical decision-support systems.
• This method offers a more uniform feature selection frequency, reducing local optima risks.