Self-organizing visualization and pattern matching of vectorcardiographic QRS waveforms.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• QRS morphology in electrocardiograms (ECG) is crucial for diagnosing ventricular depolarization abnormalities, including left bundle branch block (LBBB) and myocardial infarction.
• Current diagnostic methods rely on qualitative assessments of QRS morphology, potentially limiting accuracy and scalability.

Purpose of the Study:

• To investigate the efficacy of 3-dimensional vectorcardiogram (VCG) QRS loop pattern matching for improving patient grouping.
• To assess the potential of self-organizing algorithms for unsupervised learning and computer-assisted cardiac disorder detection.

Main Methods:

• A 93x93 patient-to-patient dissimilarity matrix was generated based on VCG QRS loop morphology.
• A self-organizing algorithm was employed to optimize node (patient) locations, preserving the dissimilarity matrix by minimizing network energy.
• The convergence and clustering performance of the algorithm were evaluated.

Main Results:

• The self-organizing algorithm successfully and automatically organized 93 patients into three distinct clusters: healthy controls, LBBB, and infarction.
• Patient location convergence was achieved and independent of initial node placement, indicating algorithm robustness.
• The spatial coordinates derived from the optimized network served as novel predictors for computer-assisted cardiac disorder detection.

Conclusions:

• Self-organizing pattern matching of VCG QRS loops offers a powerful approach for unsupervised learning and patient group identification.
• This method demonstrates significant potential for enhancing the computer-assisted detection of cardiac disorders, improving clinical decision-making.