University of Oxford (Gu et al.)
Model weights not public. Contact creators for more information.
Multimodal cardiac-sensing foundation model pretrained with generative masked pretraining on ECG, PPG, and paired clinical/machine-generated text reports from roughly 1.7 million individuals across three large-scale critical-care and outpatient ECG datasets. A channel-embedding scheme lets the same model accept any combination of 12-lead ECG, single-lead/wearable ECG, and PPG. The resulting embeddings transfer to diagnostic classification, demographic recognition, vital-sign measurement, clinical-outcome prediction, and ECG question answering. Pretrained weights require a signed academic-access agreement rather than an open download.
Architecture
Transformer
Transformer encoder with generative masked pretraining; variable-channel embedding scheme supports 12-lead ECG, single-lead/wearable ECG, and PPG in any combination
Framework
PyTorch
Added to catalog
2026-08-10
Publicly available 15% subset of the CODE (Clinical Outcomes in Digital Electrocardiology) dataset from the Telehealth Network of Minas Gerais; full CODE dataset (>2 million exams) is request-only.
Multi-parameter ICU bedside-monitor waveform recordings (ECG, PPG/pulse oximetry, arterial blood pressure) from critical-care patients at Beth Israel Deaconess Medical Center; one of three corpora (with MIMIC-IV-ECG and CODE-15%) used to pretrain CSFM on data from ~1.7 million individuals combined.
800,035 12-lead ECG-report pairs from 161,352 subjects at Beth Israel Deaconess Medical Center.
Multimodal cardiac biosignal embedding (ECG/PPG, any channel combination)
Demographic and vital-sign regression (e.g. age, BMI) from cardiac biosignals