A Multi-Modal Electrocardiogram and Electronic Health Record Dataset for Cardiac Diagnosis
Description
This dataset is a multi-modal Electrocardiogram (ECG) and Electronic Health Record (EHR) dataset developed for cardiac diagnosis research and artificial intelligence applications. It comprises 585 12-lead ECG images collected from 200 patients, including 100 patients from the Coronary Care Unit (CCU) and 100 patients from the Emergency Room (ER). A detailed clinical health record in Electronic Health Record (EHR) format is available for each patient, providing clinical and demographic information associated with the corresponding ECG data. The dataset therefore enables the integration of ECG image information with patient-level clinical information for multimodal analysis and cardiac diagnosis research. For each ECG image, three physician-answered question–answer pairs were developed covering three clinical aspects: cardiac rhythm, ECG findings, and diagnosis. This results in a total of 1,755 physician-answered question–answer pairs (585 ECG images × 3 questions per image). The question–answer pairs are designed to support research in multimodal learning, visual question answering, ECG interpretation, clinical decision-support systems, and large language model applications in cardiology. The dataset is intended for research and development of artificial intelligence and machine learning methods for ECG analysis, multimodal clinical reasoning, and cardiac diagnosis. Patient data are represented using study-specific identifiers and are provided in a research-oriented structure.
Files
Steps to reproduce
To develop a comparable ECG–EHR dataset, the following steps are recommended: 1. Obtain Institutional Ethical Approval: Obtain approval from the relevant Institutional Review Board (IRB) or Institutional Ethics Committee of the researcher's educational or research institution before commencing data collection. 2. Obtain Clinical-Site Approval: Obtain formal approval from the participating hospital or clinical site through its applicable Ethical Review Board (ERB) or equivalent institutional authority. 3. Obtain Informed Patient Consent: Provide participants with comprehensive information about the study, including its purpose, data to be collected, intended research use, data protection measures, potential risks, and their rights. Obtain informed consent before collecting research data. 4. Collect ECG and Clinical Data: Collect 12-lead ECG data and the corresponding clinical/EHR information according to the approved study protocol and inclusion/exclusion criteria. 5. Ensure Secure Data Handling: Store collected data in a secure, access-controlled research environment and implement appropriate safeguards throughout data collection, processing, analysis, and storage. 6. De-identify the Data: Remove or appropriately transform direct identifiers and assess potential indirect identifiers to minimize the risk of participant re-identification before data are used for research sharing or analysis. 7. Perform Quality Control and Analysis: Conduct appropriate data-quality checks, organize and link the ECG and EHR modalities, and perform the intended statistical, machine-learning, deep-learning, or multimodal analyses. 8. Maintain Clinical-Site and Participant Safeguards: Data collection and research activities must not disrupt routine clinical operations. Researchers must follow the approved clinical-site procedures and must not override, interfere with, or supersede any clinical decision, workflow, or safety procedure. Note: All data collection, processing, storage, analysis, and sharing activities should comply with applicable institutional ethical approvals, informed-consent requirements, data-protection policies, and the regulations governing the participating clinical site.
Institutions
- National University of Sciences and TechnologyIslamabad, Islamabad
- Wah Medical CollegePunjab, Wah Cantt