An empirical study of modified beat SQI based majority voting fusion method for heart-rate estimation in noisy multimodal cardiovascular signals

. Most existing heartbeat-detection algorithms rely heavily on cardiovascular signals, namely electrocardiogram (ECG) and arterial blood pressure (ABP), which are often corrupted by noise, leading to unreliable heart-rate estimates. Simultaneously recorded non-cardiovascular (NC) signals help with reliable heart-rate estimates when both cardiovascular signals are corrupted by noise. This study aims to: (i) propose a modified beat signal quality index-based majority voting fusion (MMVF) method to deal with extremely noisy cardiovascular signals; (ii) generate synthetic noise datasets from standard PhysioNet datasets by adding different types of ECG noises, i.e. baseline wander (bw), electrode motion (em), muscle artifact (ma), and realistic artificial ABP noises in clean or low-noise ECG and ABP signals, respectively; and (iii) analyze the effectiveness of the MMVF method for heart-rate estimation with different combinations of beat detectors. The modified beat signal quality index in the proposed method can identify the quality of the signal even when it contains long durations of noise. The MMVF method assigns uniform weights to the beats detected from all multimodal physiological signals, thus enabling effective participation of beats from NC signals when both cardiovascular signals are corrupted. Fusion of the NC signals with noisy cardiovascular signals using the MMVF method improves heart-rate estimation accuracy over that of single ECG beat detectors like gqrs, epltd, and slope sum function and Teager-Kaiser energy (SSF-TKE) up to 98.81%, 97.95%, and 87.98%, respectively. This method has yielded robust heart-rate estimation within clinically acceptable error limits in concurrently highly noisy cardiovascular signals (ECG: up to a signal-to-noise ratio (SNR) of -70 dB and ABP: up to 100% noise duration in noisy segments) by their fusion with NC signals. This study serves as empirical evidence for the robustness of the MMVF method in scenarios where there are extremely noisy cardiovascular signals and NC signals with ECG R-peak artifacts.