ELECTROCARDIOGRAM (ECG) noise filtering took a step forward with new research showing that tailoring signal processing to the presence and type of noise significantly improved the accuracy of key cardiac measurements.
Researchers developed and evaluated a knowledge-based ECG noise filtering framework that first detected whether noise was present, then identified the noise type, before applying a targeted filtering strategy. This adaptive approach reduced unnecessary signal distortion compared with conventional, noise-agnostic filtering methods that apply the same preprocessing regardless of signal quality.
Knowledge-Based ECG Noise Filtering Preserves Intervals
Electrocardiograms are central to cardiac monitoring across hospitals, clinics, and portable devices, but signal contamination from motion artefacts, muscle activity, or baseline wander remains a persistent challenge. Excessive or inappropriate filtering can alter clinically important features, particularly QT and QRS intervals, which are critical for diagnosing arrhythmias and assessing drug safety.
In this study, the knowledge-based ECG noise filtering framework was tested across seven datasets using synthetic noise. The authors compared three approaches: traditional noise-agnostic filtering, noise-presence filtering, and noise-profile filtering. Changes in QT and QRS intervals were used as markers of clinical signal distortion.
Noise-profile filtering performed best, achieving the smallest mean QT interval difference of 2.50 ms. This compared favourably with noise-presence filtering and substantially outperformed noise-agnostic filtering, which produced the largest deviations. QRS interval accuracy also improved progressively, with noise-profile filtering reducing differences to 4.28 ms.
To optimise detection, the team examined sampling frequency using kernel density estimation. A sampling rate of 500 Hz delivered the most reliable noise detection and classification performance. A hierarchical Adaboost model outperformed support vector machine, random forest, and ExtraTree classifiers, achieving high accuracy in both noise detection and noise classification.
Implications for ECG Analysis and Portable Devices
These findings are particularly relevant as ECG monitoring expands beyond controlled clinical environments into ambulatory and wearable technologies. More precise interval preservation supports reliable interpretation in settings where noise is unavoidable, such as home monitoring and remote care. Related developments in ECG analytics have also highlighted the growing role of intelligent preprocessing in improving downstream diagnostics.
The main limitation was reliance on synthetic noise, which may not capture the full complexity of real-world artefacts. However, the authors noted that the framework remained suitable for portable ECG systems and could be extended to other physiological signals through retraining.
Reference
Rahman S et al. Design and evaluation of a knowledge-based ECG noise filtering framework. Sci Rep. 2026; DOI:10.1038/s41598-025-32249-7.
