Abstract
Accurate cardiac output (CO) estimation from impedance cardiography (ICG) signals is vital for non-invasive monitoring of cardiovascular disorders, including heart failure, arrhythmias, and hemodynamic instability. However, the reliability of ICG-based assessment is often limited by noise artifacts that obscure clinically relevant features. This study introduces a novel three-stage denoising framework integrating Variational Mode Decomposition (VMD), Non-Local Means (NLM), and Discrete Wavelet Transform (DWT) to enhance ICG signal quality for robust CO estimation. The method was validated on the publicly available ReBeatICG dataset, which includes annotated signals from 24 subjects and reflects real-world noise sources such as motion artifacts and baseline drift. Experimental results demonstrate that the proposed VMD-NLM-DWT approach achieves a maximum of 1.2 dB improvement in signal-to-noise ratio (SNR), an average 13% reduction in mean squared error (MSE), and 9% lower percent root mean square difference (PRD) compared to leading two-stage denoising methods. The framework also enhances fiducial point detection (F1-score increase up to 4.4%) and preserves high heart rate variability (HRV) fidelity (correlation coefficient 0.91), with the highest denoising robustness index (DRI) observed across a wide range of noise conditions. These findings confirm that the proposed method outperforms state-of-the-art alternatives in preserving both signal fidelity and clinically significant features under both stationary and non-stationary noise. Furthermore, all performance improvements are statistically validated using paired t-tests and effect size analysis ([Formula: see text], Cohen's [Formula: see text]) and achieves top scores in PSNR and SSIM compared to all baselines. Computational profiling demonstrates feasibility for real-time, continuous cardiac output monitoring in clinical and ambulatory care, supporting its broader application in cardiovascular diagnostics.