EHR-EC, A De-identified Dataset of Endometrial Cancer from Electronic Health Records and Medical Images

Electronic health records are increasingly recognized as a valuable resource for improving computer-aided diagnosis, but the scarcity of large publicly available datasets hinders the adoption of such approaches. This paper presents EHR-EC, a multi-modal dataset of endometrial cancer derived from admissions between 2013 and 2025 in two hospitals in China, comprising 65,474 electronic health records and 49,851 medical images.

Information available includes patient measurements, diagnoses, procedures, treatments, cancer staging, pathology images, radiology images, and de-identified free-text clinical notes. All direct identifiers (e.g., patient names, national identification numbers, phone numbers, and detailed addresses) have been removed. Temporal information has also been de-identified, and internal hospital identifiers have been re-masked to further limit residual re-identification risks.