Sequential Probability Assignment for Outlier Detection in Heartbeat Timings.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Inaccurate heart rate and heart rate variability estimates arise from artifacts, noise, and physiological outlier beats.
• The proliferation of wearable devices necessitates robust methods for identifying erroneous heartbeat detections due to motion artifacts and poor sensor contact.

Purpose of the Study:

• To propose a sequential probability assignment procedure for detecting outlier heartbeats.
• To develop a flexible time-varying point process model capable of capturing changes in both mean and variance of interbeat intervals.

Main Methods:

• A time-varying point process model estimating a two-parameter exponential family distribution per time index.
• Formulation of a maximum likelihood problem with a Kullback-Leibler regularizer at each time step.
• Testing of inverse Gaussian, gamma, and log-normal distributions, with inverse Gaussian showing the best fit for interbeat intervals via Kolmogorov-Smirnov statistic.

Main Results:

• The inverse Gaussian distribution demonstrated the best fit for interbeat interval data from clinical electrocardiogram (ECG) data.
• The proposed model successfully detected outlier heartbeats in both simulations and clinical data.
• The sequential probability assignment procedure proved effective in identifying statistically unlikely heartbeat timings.

Conclusions:

• The developed outlier detection method enhances the reliability of heart rate and heart rate variability measurements, particularly in noisy environments.
• This technique is crucial for improving the accuracy of wearable health monitoring devices.
• The model's ability to identify ectopic beats and arrhythmic events contributes to more precise clinical relevance.