Sequential Markov chain Monte Carlo filter with simultaneous model selection for electrocardiogram signal modeling.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Biology

Background:

• Statistical electrocardiogram (ECG) models are crucial for automated cardiac disease classification, but often require extensive preprocessing and user-defined parameters.
• Existing models struggle with the inherent variability of ECG morphologies (P, QRS, T waves) across individuals and diseases.
• A need exists for adaptive ECG modeling techniques that can handle data variations and reduce reliance on manual initialization.

Purpose of the Study:

• To propose and evaluate a novel ECG modeling technique using the sequential Markov chain Monte Carlo (SMCMC) filter.
• To enable simultaneous model selection and adaptive representation based on ECG data characteristics.
• To utilize estimated model parameters for automated classification of cardiac arrhythmias.

Main Methods:

• Development of an ECG modeling technique employing the sequential Markov chain Monte Carlo (SMCMC) filter.
• Implementation of simultaneous model selection, allowing adaptive representation choice based on data.
• Utilizing estimated model parameters as features for a classification task.

Main Results:

• The proposed SMCMC filter effectively tracks diverse ECG morphologies, including intermittent beats.
• The algorithm demonstrates robustness in adapting to different ECG signal characteristics.
• Classification accuracy was achieved for distinguishing normal sinus rhythm from four types of arrhythmia using estimated model parameters.

Conclusions:

• The SMCMC filter offers a powerful and adaptive approach for statistical ECG modeling.
• This method reduces the need for a priori information and manual parameter tuning in ECG analysis.
• The developed technique shows significant potential for improving automated cardiac disease diagnosis and classification.