Back

HuBERT-ECG: a self-supervised foundation model for broad and scalable cardiac applications

Coppola, E.; Savardi, M.; Massussi, M.; Adamo, M.; Metra, M.; Signoroni, A.

2024-11-18 health informatics
10.1101/2024.11.14.24317328 medRxiv
Show abstract

Deep learning models have shown remarkable performance in electrocardiogram (ECG) analysis, but the limited availability and size of ECG datasets have constrained their success, resulting in systems that are more task specialists than versatile generalists. To counter this, we introduce HuBERT-ECG, a novel self-supervised foundation ECG model pre-trained on a large and diverse dataset of 9.1 million 12-lead ECGs encompassing 164 cardiovascular conditions. By simply adding a proper output layer, HuBERT-ECG can be fine-tuned for a wide array of downstream tasks, from diagnosing diseases to predicting future cardiovascular events. Across 17 diverse real-world scenarios, HuBERT-ECG achieves AUROCs from 0.843 on small datasets to 0.99 on larger sources. When fine-tuned to detect 164 overlapping conditions simultaneously, our model delivers AUROCs above 0.9 and 0.95 for up to 140 and 97 diseases, respectively. HuBERT-ECG can also predict death events within a 2-year follow-up with AUROCs up to 0.91. We release pre-trained models and code as building baselines.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.