Real Clinical Data

This dataset contains de-identified, single-lead electrocardiogram (ECG) signals collected in real-world clinical settings using HeartVoice devices. Signals are sampled at 125 Hz and stored as JSON files that group segments by patient. Two top-level categories are provided: VT/, which contains segments recorded during ventricular tachycardia (VT) episodes, and N_VT/, which contains segments from non-VT (normal/other symptom) periods. Each JSON entry stores the floating-point ECG waveform together with lightweight metadata—including a stable patient identifier—so that researchers can pair each patient’s VT segments with their corresponding non-VT controls for within-subject analyses. The data have been organized to facilitate reproducible benchmarking: file-level labels reflect the VT vs. non-VT status, while metadata fields inside each record (e.g., sample_rate, user_id, labels, ecg_data) preserve original acquisition context without exposing protected health information. Overall, this resource aims to bridge controlled algorithm development and bedside deployment by providing clean, machine-readable ECG segments with explicit patient-level correspondence between arrhythmic and normal rhythms.