Multi-Channel ECG and PCG Dataset for Heart Failure with Reduced Ejection Fraction (HFrEF) Detection

Contributors Peng Zhang, Fei Ma, and Qiang Li Overview This dataset provides 2,480 pairs of synchronized electrocardiogram (ECG) and phonocardiogram (PCG) recordings from 620 subjects for the detection of heart failure with reduced ejection fraction (HFrEF), serving as a valuable resource for the development of AI-based diagnostic models, multi-modality fusion methods, and clinical decision-support systems. The recordings were collected at Tongji Hospital, Tongji medical college, Huazhong University of Science and Technology (HUST) and Liyuan Hospital, HUST, using the widely available EKO DUO ECG + Digital Stethoscope. All subjects were diagnosed by three professional cardiologists based on the left ventricular ejection fraction (LVEF) measured by transthoracic echocardiography at the Cardiac Function Examination Center of Tongji Hospital, where they were diagnosed as HFrEF when LVEF ≤ 40%, and as non-HFrEF otherwise. Each subject was recorded sequentially at four standard auscultation sites (aortic valve, pulmonary valve, tricuspid valve, and mitral valve), producing four-channel ECG and PCG signals. Within each channel, ECG and PCG signals are synchronous, while signals across channels are asynchronous. Primary dataset composition Collected at Tongji Hospital (HUST) 500 subjects (295 males, 205 females; average age: 57.6 ± 12.9 years) 2000 pairs of 30-second synchronous ECG and PCG recordings 1760 pairs from 440 non-HFrEF subjects 240 pairs from 60 HFrEF subjects (diagnosed as LVEF ≤ 40% via echocardiography) Both HFrEF and non-HFrEF subjects may have comorbid conditions (e.g., valvular diseases, arrhythmia, cardiomyopathy) Data split into five subject-level folds for cross-validation Additional community-scenario test set Collected independently at Liyuan Hospital (HUST) from 120 subjects (5 HFrEF, 115 non-HFrEF), yielding 480 pairs of recordings. This test set simulates deployment in a community healthcare scenario. Technical details Recording duration: 30 seconds per site ECG sampling rate: 500 Hz PCG sampling rate: 4000 Hz Device: EKO DUO ECG + Digital Stethoscope Signals have been anonymized and de-identified in compliance with HIPAA Safe Harbor provisions. File structure HFrEF_5_folds.zip – contains the five-fold cross-validation dataset. HFrEF_community_scenario_test_set.zip – contains the community-scenario test set. Inside each archive: Signal files are stored in .wav format, named as: [ID]_[APEX/LLSB/LUSB/RUSB]_[ECG/PCG].wav where ID is the subject identifier, and APEX / LLSB / LUSB / RUSB indicate the auscultation site (mitral, tricuspid, pulmonary, and aortic valves respectively). Labels are stored in a corresponding *_label.csv, where: Column 1: Subject ID Column 2: Left ventricular ejection fraction (LVEF) Ethics This study was approved by the Ethical Committee of Tongji medical college, Huazhong University of Science and Technology, with Institutional Review Board Approval number of 2021-S271. Since ECG and PCG signals were appropriately anonymized, de-identified and retrospectively collected following the Health Insurance Portability and Accountability Act Safe Harbor provision, this study did not require written informed consent. Data Preprocessing This dataset was used in our research paper Hierarchical fusion of electrocardiogram and phonocardiogram data improves heart failure detection. Please refer to the official implementation of the paper for data preprocessing details. Citation If you find our dataset useful in your research, please consider citing: @article{MA2026101448, title = {Hierarchical fusion of electrocardiogram and phonocardiogram data improves heart failure detection}, journal = {Patterns}, volume = {7}, number = {3}, pages = {101448}, year = {2026}, issn = {2666-3899}, doi = {https://doi.org/10.1016/j.patter.2025.101448}, author = {Fei Ma and Haobo Zhang and Siyi Fang and Qimei Wang and Fan Lin and Lianying Chao and Zhiwei Wang and Qiang Li and Peng Zhang}, }