An Artificial Multi-Channel Model for Generating Abnormal Electrocardiographic Rhythms

Area of Science:

• Computational physiology and signal processing.
• Artificial intelligence in medical diagnostics.

Background:

• Existing artificial models for multi-channel ECG generation lacked the capability to simulate abnormal heart rhythms.
• Accurate simulation of cardiac electrical activity, including arrhythmias, is crucial for developing and testing diagnostic algorithms.

Purpose of the Study:

• To generalize previous ECG generation models to enable the simulation of abnormal cardiac rhythms.
• To incorporate realistic beat-to-beat variations, heart rate variability, and specific phenomena like T-Wave Alternans (TWA).

Main Methods:

• Utilized a three-dimensional vectorcardiogram (VCG) formulation with Gaussian kernels to model normal cardiac dipoles.
• Introduced abnormal beats as new or perturbed dipoles, controlled by a hidden Markov model (HMM) coupled to heart rate.
• Simulated QT-HR hysteresis, respiration-induced morphology changes, and TWA using VCG perturbations and HMMs, mapping VCG to clinical ECG leads via a Dower-like transform.

Main Results:

• Successfully generated multi-channel ECG signals with simulated abnormal rhythms, including TWA.
• Demonstrated realistic beat-to-beat morphology changes and QT-HR hysteresis.
• Modeled TWA magnitude and its HR-dependent probability, mimicking physiological associations.

Conclusions:

• The generalized artificial model effectively simulates normal and abnormal cardiac rhythms, including TWA, with physiological realism.
• The model provides a valuable tool for developing and validating ECG analysis algorithms, particularly for arrhythmia detection.
• Generated ECGs with calibrated TWA were successfully included in the PhysioNet/CinC Challenge 2008 dataset.