3KG: Contrastive Learning of 12-Lead Electrocardiograms using Physiologically-Inspired Augmentations

We propose 3KG, a physiologically-inspired contrastive learning approach that generates views using 3D augmentations of the 12-lead electrocardiogram. We evaluate representation quality by fine-tuning a linear layer for the downstream task of 23-class diagnosis on the PhysioNet 2020 challenge training data and find that 3KG achieves a 9.1% increase in mean AUC over the best self-supervised baseline when trained on 1% of labeled data. Our empirical analysis shows that combining spatial and temporal augmentations produces the strongest representations. In addition, we investigate the effect of this physiologically-inspired pretraining on downstream performance on different disease subgroups and find that 3KG makes the greatest gains for conduction and rhythm abnormalities. Our method allows for flexibility in incorporating other self-supervised strategies and highlights the potential for similar modality-specific augmentations for other

[1]  Alexei Baevski,et al.  wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations , 2020, NeurIPS.

[2]  Erik Reinertsen,et al.  Patient Contrastive Learning: a Performant, Expressive, and Practical Approach to ECG Modeling , 2021, ArXiv.

[3]  Radek Martinek,et al.  Comparison of Different Electrocardiography with Vectorcardiography Transformations , 2019, Sensors.

[4]  Kaiming He,et al.  Momentum Contrast for Unsupervised Visual Representation Learning , 2019, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[5]  Dani Kiyasseh,et al.  CLOCS: Contrastive Learning of Cardiac Signals , 2020, ICML.

[6]  W. Einthoven,et al.  On the direction and manifest size of the variations of potential in the human heart and on the influence of the position of the heart on the form of the electrocardiogram. , 1950, American heart journal.

[7]  Rickey E Carter,et al.  An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction , 2019, The Lancet.

[8]  Leslie N. Smith,et al.  Cyclical Learning Rates for Training Neural Networks , 2015, 2017 IEEE Winter Conference on Applications of Computer Vision (WACV).

[9]  Geoffrey H. Tison,et al.  Automated and Interpretable Patient ECG Profiles for Disease Detection, Tracking, and Discovery , 2018, Circulation. Cardiovascular quality and outcomes.

[10]  Kaiming He,et al.  Improved Baselines with Momentum Contrastive Learning , 2020, ArXiv.

[11]  Oriol Vinyals,et al.  Representation Learning with Contrastive Predictive Coding , 2018, ArXiv.

[12]  Competency in interpretation of 12-lead electrocardiograms: a summary and appraisal of published evidence , 2003 .

[13]  Geoffrey E. Hinton,et al.  Big Self-Supervised Models are Strong Semi-Supervised Learners , 2020, NeurIPS.

[14]  Geoffrey E. Hinton,et al.  A Simple Framework for Contrastive Learning of Visual Representations , 2020, ICML.

[15]  Andrew Y. Ng,et al.  MoCo Pretraining Improves Representation and Transferability of Chest X-ray Models , 2020, MIDL.

[16]  A. Etemad,et al.  Self-Supervised ECG Representation Learning for Emotion Recognition , 2020, IEEE Transactions on Affective Computing.

[17]  S. Salerno,et al.  Competency in Interpretation of 12-Lead Electrocardiograms: A Summary and Appraisal of Published Evidence , 2003, Annals of Internal Medicine.

[18]  P. Anversa,et al.  Gender differences and aging: effects on the human heart. , 1995, Journal of the American College of Cardiology.

[19]  Ali Etemad,et al.  Self-Supervised Learning for ECG-Based Emotion Recognition , 2020, ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[20]  Dahua Lin,et al.  Unsupervised Feature Learning via Non-Parametric Instance-level Discrimination , 2018, ArXiv.

[21]  Ali Bahrami Rad,et al.  Classification of 12-lead ECGs: the PhysioNet/Computing in Cardiology Challenge 2020 , 2020, 2020 Computing in Cardiology.

[22]  Yoshua Bengio,et al.  Learning deep representations by mutual information estimation and maximization , 2018, ICLR.

[23]  J. Hurst,et al.  Methods used to interpret the 12‐lead electrocardiogram: Pattern memorization versus the use of vector concepts , 2000, Clinical cardiology.

[24]  H. Kennedy,et al.  Ambulatory (Holter) electrocardiography technology. , 1992, Cardiology clinics.

[25]  Cordelia Schmid,et al.  What makes for good views for contrastive learning , 2020, NeurIPS.

[26]  F. Fesmire,et al.  Usefulness of automated serial 12-lead ECG monitoring during the initial emergency department evaluation of patients with chest pain. , 1998, Annals of emergency medicine.

[27]  Michal Valko,et al.  Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , 2020, NeurIPS.

[28]  Phillip Isola,et al.  Contrastive Multiview Coding , 2019, ECCV.

[29]  R R Bond,et al.  Assessing computerized eye tracking technology for gaining insight into expert interpretation of the 12-lead electrocardiogram: an objective quantitative approach. , 2014, Journal of electrocardiology.

[30]  Rickey E. Carter,et al.  Screening for cardiac contractile dysfunction using an artificial intelligence–enabled electrocardiogram , 2019, Nature Medicine.

[31]  Oncel Tuzel,et al.  Subject-Aware Contrastive Learning for Biosignals , 2020, ArXiv.

[32]  Andrew Y. Ng,et al.  MedAug: Contrastive learning leveraging patient metadata improves representations for chest X-ray interpretation , 2021, MLHC.

[33]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[34]  Frank Bogun,et al.  Accuracy of electrocardiogram interpretation by cardiologists in the setting of incorrect computer analysis. , 2006, Journal of electrocardiology.

[35]  Nils Strodthoff,et al.  Self-supervised representation learning from 12-lead ECG data , 2021, Comput. Biol. Medicine.

[36]  Dani Kiyasseh,et al.  PCPs: Patient Cardiac Prototypes , 2020, ArXiv.

[37]  Dani Kiyasseh,et al.  DROPS: Deep Retrieval of Physiological Signals via Attribute-specific Clinical Prototypes , 2020, ArXiv.