A Machine-Learning Approach for Detection and Quantification of QRS Fragmentation

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Fragmented QRS (fQRS) on an electrocardiogram (ECG) indicates myocardial scarring.
• Current fQRS scoring is manual, time-consuming, and subjective.
• There is a need for automated, objective methods for fQRS assessment.

Purpose of the Study:

• To develop and evaluate an automated method for detecting and quantifying fQRS.
• To utilize variational mode decomposition (VMD) and phase-rectified signal averaging (PRSA) for feature extraction.
• To compare the performance of different machine learning classifiers for fQRS analysis.

Main Methods:

• QRS complexes were segmented using VMD.
• Ten VMD- and PRSA-based features were computed.
• Support vector machine (SVM), K-nearest neighbors (KNN), Naive Bayesian (NB), and TreeBagger (TB) classifiers were employed and compared using 12-lead ECG data from 616 patients.

Main Results:

• The automated method achieved high AUC values for fQRS detection: 0.95 (SVM), 0.94 (KNN), 0.90 (NB), and 0.89 (TB).
• The quantified fQRS score demonstrated a clear correlation with expert rater scores, validating its accuracy.
• The method successfully detected and quantified fQRS in continuous ECG signals.

Conclusions:

• The proposed automated method effectively detects and quantifies fQRS from ECG signals.
• This novel approach provides an objective and efficient tool for assessing QRS fragmentation and myocardial scarring.
• The VMD and PRSA-based feature extraction combined with machine learning offers a promising advancement in ECG analysis.