Organizations: Leiden Institute of Advanced Computer Science (LIACS), Leiden University, Leiden, The Netherlands. · the School of Computer Science and Technology, University of Science and Technology of China, Anhui, China.
Twelve-lead electrocardiography (ECG) is essential for cardiovascular diagnosis, but its long-term acquisition in daily life is constrained by complex and costly hardware. Recent efforts have explored reconstructing ECG from low-cost cardiac vibrational signals such as seismocardiography (SCG), however, due to the lack of a dataset, current methods are limited to limb leads, while clinical diagnosis requires multi-lead ECG, including chest leads. In this work, we propose Vib2ECG, the first paired, multi-channel electro-mechanical cardiac signal dataset, which includes complete twelve-lead ECGs and vibrational signals acquired by inertial measurement units (IMUs) at six chest-lead positions from 17 subjects. Based on this dataset, we also provide a benchmark. Experimental results demonstrate the feasibility of reconstructing electrical cardiac signals at variable locations from vibrational signals using a lightweight 364 K-parameter U-Net. Furthermore, we observe a hallucination phenomenon in the model, where ECG waveforms are generated in regions where no corresponding electrical activity is present. We analyze the causes of this phenomenon and propose potential directions for mitigation. This study demonstrates the feasibility of mobile-device-friendly ECG monitoring through chest-lead ECG prediction from low-cost vibrational signals acquired using IMU sensors. It expands the application of cardiac vibrational signals and provides new insights into the spatial relationship between cardiac electrical and mechanical activities with spatial location variation.
Figures & tables
Fig. 1: Conceptual overview of SCG-to-ECG reconstruction for mobile and wearable cardiac monitoring. While SCG offers a cost-effective sensing modality, its limited interpretability motivates the reconstruction of ECG signals. Existing approaches are constrained by the lack of paired chest-lead SCG-ECG datasets. This work addresses these gaps by providing a paired chest-lead dataset based on twelve-lead system and establishing a SCG-to-ECG reconstruction benchmark.
Fig. 2: The overview of the model and data collection approach. Left: In the hardware setup, an IMU is attached to an electrode buckle to form an electromechanical sensing pair that records colocated ECG and vibrational signals at six positions (V1–V6). All pairs are controlled by a single FPGA with a shared global clock, enabling timestamp-based interpolation for cross-modal alignment. Middle: The acquired vibration signals are decomposed into low-frequency signals (SCG, 2–20 Hz) and high-frequency phonocardiogram-like components (PCGL, > 20 Hz), which are normalized separately and then together used as model inputs. Right: The baseline U-Net model takes SCG and PCGL as inputs to reconstruct the corresponding ECG signals, supervised by the aligned ground-truth ECG through a reconstruction loss.
Fig. 3: SCG (0-20 Hz) contains over 96% of the total energy of the original signal. If the PCG-like (PCGL) component (above 20 Hz) is not separately extracted and normalized, it will be invisible within the original waveform, and the information carried in the PCGL band will be ignored.
Fig. 4: A typical comparative example with different inputs. (a) Reconstruction with only SCG as input. The outputs are less precise in amplitude and timing but exhibit fewer hallucinations, i.e., generated ECG waveforms that do not actually exist. (b) Reconstruction with SCG and PCGL as inputs. The outputs are precise and exhibit fewer hallucinations. (c) Reconstruction with only PCGL as input. The outputs are more precise but exhibit more hallucinations.
Fig. 5: (a) Hallucination percentage with different inputs. “PCGL” and “SCG” denote models using a single normalized PCGL or SCG channel as input, respectively. “Raw” indicates the use of a single channel of the original acceleration signal without decomposition. “Both” represents a two-channel input consisting of normalized PCGL and SCG signals. (b) L1 distance with different inputs. (c) Hallucination percentage with different time interval between training and testing data. (d) L1 distance with different time interval between training and testing data.
Fig. 6: Factors contributing to the occurrence of hallucinations: (a) noise interference, (b) misclassification of the second heart sound (S2) as the first heart sound (S1), and (c) semantic loss at segment boundaries.
Fig. 7: Results from Subject 10 demonstrate that as long as the input signal quality is sufficient for the model to obtain enough information to determine whether and when a heartbeat occurs, no hallucinations are produced in the output, even when the ECG signal used as ground truth contains substantial noise.
Estimating electrocardiography (ECG) from a chest-worn inertial measurement unit (IMU) enables continuous heart rate (HR) monitoring without the discomfort of electrodes. We propose TinyCardioUNet, a lightweight UNet that uses all six IMU axes without prior channel selection, refines its bottleneck with a graph neural network that encodes inter-axis dependencies, and employs tensor decomposition with automatic variational Bayesian rank selection for parameter reduction. On a public dataset, TinyCardioUNet achieves an RMSE of 0.098 and a Pearson correlation coefficient of 0.677 with only 36.0k parameters and remains comparatively robust to additive noise, demonstrating accurate ECG reconstruction with a compact model.
Seungwoo Han, Ingon Chanpornpakdi, Motoi Noda +3
Department of Electrical Engineering and Computer Science, Tokyo University of Agriculture and Technology, Japan
Reconstructing a 12-lead electrocardiogram (ECG) from a reduced lead set is an ill-posed inverse problem due to anatomical variability. Standard deep learning methods often ignore underlying cardiac pathology losing vital morphology in precordial leads. We propose Pathology-Aware Multi-View Contrastive Learning, a framework that regularizes the latent space through a pathological manifold. Our architecture integrates high-fidelity time-domain waveforms with pathology-aware embeddings learned via supervised contrastive alignment. By maximizing mutual information between latent representations and clinical labels, the framework learns to filter anatomical "nuisance" variables. On the PTB-XL dataset, our method achieves approx. 76% reduction in RMSE compared to state-of-the-art model in patient-independent setting. Cross-dataset evaluation on the PTB Diagnostic Database confirms superior generalization, bridging the gap between hardware portability and diagnostic-grade reconstruction.
Youssef Youssef, Jitin Singla
Department of Computer Science and Engineering, Indian Institute of Technology Roorkee, Roorkee 247667, India · Department of Biosciences and Bioengineering, Indian Institute of Technology Roorkee, Roorkee 247667, India
Electrocardiograms (ECGs) are among the most widely used diagnostic tools for cardiovascular diseases, and a large amount of ECG data worldwide appears only in image form. However, most existing automated ECG analysis methods rely on access to raw signal recordings, limiting their applicability in real-world and resource-constrained settings. In this paper, we present ECG-Scan, a self-supervised framework for learning clinically generalized representations from ECG images through dual physiological-aware alignments: 1) Our approach optimizes image representation learning using multimodal contrastive alignment between image and gold-standard signal-text modalities. 2) We further integrate domain knowledge via soft-lead constraints, regularizing the reconstruction process and improving signal lead inter-consistency. Extensive benchmarking across multiple datasets and downstream tasks demonstrates that our image-based model achieves superior performance compared to existing image baselines and notably narrows the gap between ECG image and signal analysis. These results highlight the potential of self-supervised image modeling to unlock large-scale legacy ECG data and broaden access to automated cardiovascular diagnostics.
Hung Manh Pham, Jialu Tang, Aaqib Saeed +3
University of Cambridge · Singapore Management University · Eindhoven University of Technology, Netherlands