Parametric-based feature selection via spherical harmonic coefficients for the left ventricle myocardial infarction...

Area of Science:

• Cardiology and Medical Imaging
• Machine Learning in Healthcare
• Biomedical Engineering

Background:

• Computer-aided diagnosis (CAD) for heart disease is advancing with machine learning.
• Accurate detection of myocardial infarction (MI) is crucial for patient outcomes.
• Existing methods for cardiac diagnosis have limitations in precision and scope.

Purpose of the Study:

• To develop a novel parametric-based feature selection method for MI classification.
• To utilize 3D spherical harmonic (SH) shape descriptors of the left ventricle (LV).
• To investigate the hypothesis that SH coefficients of LV endocardial shapes can distinguish MI patients from healthy subjects.

Main Methods:

• Parametric-based feature selection using 3D SH coefficients of LV endocardial shapes.
• SH parameterization, expansion, and registration were performed.
• Machine learning classifiers (SVM, K-NN, RF) were trained and tested with LOOCV after PCA, evaluating different phases and alignment procedures.

Main Results:

• The proposed method achieved high performance, with SVM reaching 97.50% accuracy, 95.00% sensitivity, 100.00% specificity, and 97.56% F-score.
• The combination of SH coefficients and machine learning demonstrated robustness.
• The method outperformed conventional point-based techniques and current clinical measures, showing generalizability in dilated cardiomyopathy detection.

Conclusions:

• The novel parametric-based feature selection using SH coefficients is effective for MI classification.
• The study highlights the strength of combining SH coefficients with machine learning for cardiac diagnosis.
• This approach offers a robust and accurate alternative for screening myocardial infarction and other cardiac conditions.