Robust, real-time generic detector based on a multi-feature probabilistic method.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Robust real-time event detection in physiological signals during ambulatory monitoring is challenging, especially with artifacted data.
• Existing methods struggle with noisy and low signal-to-noise ratio (SNR) electrocardiogram (ECG) signals.

Purpose of the Study:

• To propose and evaluate an original, generic multi-feature probabilistic detector (MFPD) for real-time QRS complex detection.
• To assess the MFPD's performance on both clean and highly artifacted ECG signals under noisy conditions.

Main Methods:

• Developed a multi-feature probabilistic detector (MFPD) using binary Bayesian probability for each feature and Kullback-Leibler divergence for fusion.
• Evaluated the MFPD on the MIT-BIH arrhythmia database and a benchmark noisy database created from the MIT-BIH noise stress test database.
• Compared MFPD performance against wavelet-based, spatiotemporal, and UNSW detectors.

Main Results:

• The MFPD achieved the lowest standard deviation in sensitivity and positive predictivity (+P) across both benchmark databases.
• For artifacted ECG with low SNR, MFPD demonstrated superior performance (87.48% sensitivity, 89.39% +P) compared to the best reference method (UNSW: 88.30% sensitivity, 86.06% +P).
• Performance was comparable for low-to mildly artifactual ECG signals.

Conclusions:

• The proposed generic MFPD structure is effective for real-time QRS detection in challenging, artifacted physiological signals.
• MFPD shows significant promise for improving long-term ambulatory monitoring applications, particularly for signals with high levels of noise and artifacts.