Threshold-based system for noise detection in multilead ECG recordings.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Informatics

Background:

• Electrocardiogram (ECG) signal quality is crucial for accurate diagnosis.
• Common noise types can compromise the reliability of ECG recordings.
• Automated methods are needed to assess ECG signal integrity.

Purpose of the Study:

• To develop and validate a system for detecting common noise types in 12-lead ECG recordings.
• To evaluate the diagnostic reliability of ECG episodes based on estimated noise corruption levels.
• To identify primary sources of ECG quality disruption across different frequency bands.

Main Methods:

• Implementation of criteria for estimating noise corruption in specific frequency bands (e.g., QRS dynamics, artifacts, baseline drift, power-line interference, high-frequency noise).
• Time-domain processing of ECG series using 13 adjustable amplitude and slope criteria thresholds.
• System training and validation using an annotated set of 1000 ECGs from the PhysioNet database (CinC Challenge 2011).

Main Results:

• The system effectively identifies various noise sources, including missing signal, artifacts, baseline drift, power-line interference, and electromyographic noise.
• Two optimized implementations demonstrated high performance: one prioritizing sensitivity (Se=98.7%, Sp=80.9%) and the other specificity (Sp=97.8%, Se=81.8%).
• Receiver operating characteristic analysis showed excellent performance with an area of 0.968.

Conclusions:

• The developed system reliably detects common ECG noise, enabling better assessment of diagnostic signal quality.
• The system offers flexible configuration for application-specific noise level requirements.
• The high-sensitivity and high-specificity implementations provide valuable tools for real-time ECG quality monitoring and validation.