Key result
Hybrid CNN-Transformer deep reconstruction framework outperforms classical algorithms, achieving ~35 dB PSNR with rapid inference.
A novel hybrid CNN-Transformer architecture for compressed sensing reconstruction improves signal quality and inference speed for ECG data compared to classical algorithms.
May enable efficient ECG reconstruction for remote monitoring; leaves open clinical validation of diagnostic impact in cardiovascular cohorts.
This paper proposes a structure-constrained deep reconstruction framework for compressed sensing in shift-invariant spaces (SISs). The reconstruction is formulated as an inverse operator estimation problem derived from the matrix factorization H(ω)=W(ω)A and approximated using a hybrid CNN–Transformer architecture. Residual dilated convolutions capture localized signal structures, while the Transformer module models global frequency-domain dependencies. A variational inference-inspired regularization mechanism is incorporated to implicitly learn sparsity-aware priors. Experiments on both synthetic SIS signals and real-world ECG data demonstrate consistent improvements over classical optimization-based algorithms (ISTA, OMP) and a deep unfolding baseline (ISTA-Net+). At a 30% sampling rate, the proposed method achieves a PSNR of 35.46 dB. The feed-forward design eliminates iterative reconstruction, achieving a GPU inference time of 0.85 ms per signal.
No takes yet. Share an insight, caveat, or question.
Ling et al. (2026) studied ECG signal reconstruction. Structure-constrained deep reconstruction framework (hybrid CNN-Transformer) vs. Classical optimization-based algorithms (ISTA, OMP) and ISTA-Net+ was evaluated on PSNR at 30% sampling rate. A hybrid CNN-Transformer deep reconstruction framework achieved a PSNR of 35.46 dB at a 30% sampling rate, outperforming classical algorithms with a GPU inference time of 0.85 ms per signal.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: