A greedy graph search algorithm based on changepoint analysis for automatic QRS complex detection.

Area of Science:

• Biomedical Signal Processing
• Cardiovascular Disease Diagnostics
• Machine Learning in Healthcare

Background:

• Electrocardiogram (ECG) is crucial for non-invasive cardiovascular disease investigation.
• Accurate delineation of ECG fiducial points, especially the R-peak, is fundamental for ECG analysis.
• Existing methods often require extensive preprocessing for reliable R-peak detection.

Purpose of the Study:

• To propose a novel Graph-Constrained Changepoint Detection (GCCD) model for ECG fiducial point delineation.
• To treat R-peak detection as a changepoint detection problem within a graphical model framework.
• To incorporate biological knowledge via constraint graphs for improved detection accuracy.

Main Methods:

• Developed the GCCD model, treating ECG fiducial point delineation as a changepoint detection problem.
• Utilized the sparsity of changepoints to detect abrupt changes in non-stationary ECG signals.
• Incorporated prior biological knowledge using manually defined or automatically learned constraint graphs.

Main Results:

• Manual constraint graph achieved 99.64% sensitivity, 99.71% positive predictivity, and 0.19% detection error rate.
• Automatically learned constraint graph achieved 99.76% sensitivity, 99.68% positive predictivity, and 0.55% detection error rate.
• The GCCD model demonstrated high accuracy in R-peak detection without preprocessing.

Conclusions:

• The proposed GCCD model offers a robust and accurate method for R-peak detection in ECG signals.
• Incorporating constraint graphs, whether manual or learned, significantly enhances detection performance.
• This approach simplifies ECG analysis by eliminating the need for preprocessing steps.