Figure 1

Block diagram of the proposed framework for knowledge-based ECG noise filtering using a machine learning (ML) model and the evaluation of possible filtering options. The raw ECG signal is passed to the feature extraction module, and noise-agnostic filtering is applied simultaneously. A ML model is used for detecting noise based on extracted features, and then noisy segments are simultaneously filtered (noise-presence filtering) and passed to the noise classification module. The ML-based noise classification module classifies ECG segments and feeds them into the noise-profile filtering module. Finally, the benefit of the proposed framework is evaluated by extracting biomarkers (QRS complex and Q-T interval) after three different filterings: (i) noise-agnostic filtering, (ii) noise-presence filtering, and (iii) noise type-based filtering.