Key result
Patient Contrastive Learning of Representations (PCLR) outperformed models trained from scratch and standard ECG features across multiple tasks when using fewer than 5,000 labeled examples.
Why the study?
Supervised machine learning applications in health care are often limited by a scarcity of labeled training data.
Does Patient Contrastive Learning of Representations (PCLR) pre-training improve the performance of ECG modeling on clinical tasks when labeled training data is scarce?
Does Patient Contrastive Learning of Representations (PCLR) pre-training improve the performance of ECG modeling on clinical tasks when labeled training data is scarce?
PCLR pre-training on large unlabeled ECG datasets generates expressive representations that significantly improve model performance on downstream clinical tasks when labeled data is scarce.
PCLR may aid ECG modeling with scarce labels; leaves open clinical utility pending prospective validation.
Supervised machine learning applications in health care are often limited due to a scarcity of labeled training data. To mitigate this effect of small sample size, we introduce a pre-training approach, Patient Contrastive Learning of Representations (PCLR), which creates latent representations of ECGs from a large number of unlabeled examples. The resulting representations are expressive, performant, and practical across a wide spectrum of clinical tasks. We develop PCLR using a large health care system with over 3.2 million 12-lead ECGs, and demonstrate substantial improvements across multiple new tasks when there are fewer than 5,000 labels. We release our model to extract ECG representations at https://github.com/broadinstitute/ml4h/tree/master/model_zoo/PCLR.
No takes yet. Share an insight, caveat, or question.
Diamant et al. (2021) studied ECG modeling (n=404,929). Patient Contrastive Learning of Representations (PCLR) vs. Models trained from scratch, standard ECG features, and rhythm classification trained encoder was evaluated on Performance on downstream clinical tasks (age regression, sex, LVH, and AF classification). Patient Contrastive Learning of Representations (PCLR) outperformed models trained from scratch and standard ECG features across multiple tasks when using fewer than 5,000 labeled examples.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: