NEAR-SSB: Noisy ECG Artifact Removal via Spectral Schrodinger Bridge

We propose NEAR-SSB, a novel ECG denoising framework that extends Image-to-Image Schrödinger Bridge (I2SB) by replacing isotropic Brownian diffusion with a spectrally structured stochastic process. Unlike conventional diffusion models that assume white Gaussian corruption, NEAR-SSB incorporates a frequency-aware Power Spectral Density (PSD) prior that constrains restoration trajectories to biologically plausible ECG frequency bands. This spectral prior improves preservation of clinically relevant waveform morphology while suppressing structured physiological artifacts such as baseline wander, electrode motion, and muscle noise. Across multiple benchmark noise conditions, NEAR-SSB consistently outperforms traditional filtering, diffusion baselines, and isotropic Schrödinger Bridge variants, establishing a robust framework for physiologically grounded generative restoration.