DR-net-Mamba: Selective State-Space Modeling for Long-Range ECG Time-Series Denoising

2026-09-28

AI Summary

This study addresses the degradation of diagnostic reliability in long-range electrocardiogram (ECG) signals caused by non-stationary noise, alongside the efficiency and accuracy bottlenecks of conventional models in long-sequence modeling. To overcome these challenges, this work proposes a Mamba-enhanced denoising model that innovatively embeds a selective state space mechanism within a convolutional bottleneck architecture. By integrating local feature extraction with linear-complexity long-range temporal modeling, the proposed method effectively circumvents the quadratic complexity limitations of Transformers and the high inference costs of diffusion models. Experimental results demonstrate that the model achieves superior signal-to-noise ratios and minimal root mean square errors on both synthetic and real-world datasets. Furthermore, it significantly enhances downstream diagnostic performance for critical pathological features, such as ST/T changes, as evidenced by improved macro-AUROC scores.

Abstract

Electrocardiogram (ECG) recordings are corrupted by non-stationary noise sources that degrade diagnostic reliability, particularly in ambulatory and long-duration recordings. Deep learning denoisers exist, but convolutional architectures are limited by their receptive field, transformer-based models scale quadratically with sequence length, and diffusion-based approaches incur prohibitive inference cost. We propose a Mamba-augmented model that inserts selective state-space blocks at the convolutional bottleneck, combining local feature extraction with long-range temporal modeling at linear complexity. We comprehensively evaluate the proposed model with respect to reconstruction fidelity, noise robustness, recording-length scaling, and downstream diagnostic classification across over 40 pathology classes. On synthetic and real datasets, our model achieves the highest SNR and lowest RMSE, with the Mamba advantage increasing with sequence length and in low-SNR regimes. On classification with two independent classifiers, the proposed Mamba-based models achieve the best macro AUROC among all denoisers and improve over their convolutional base models. Calibration is more nuanced and classifier-dependent: denoising improves Binary Cross-Entropy and Brier score on Inception1D but often fails to beat the noisy input on ResNet1D-Wang, and the lead-specific Mamba variant is the only denoiser to improve both calibration metrics over the noisy baseline on both classifiers. Per-class analysis reveals a morphology-dependent benefit: Mamba substantially improves ST/T-change diagnoses, which depend on broad, context-sensitive waveforms.

Problem

• ECG denoising
• long-range time-series
• non-stationary noise
• deep learning
• diagnostic reliability

Innovation

• Selective State-Space Model
• Mamba
• ECG Denoising
• Long-Range Temporal Modeling
• Linear Complexity