[2K5-OS-1a] 医療におけるAIの社会実装に向けて(1/2)

[2K5-OS-1a-03] Self-Supervised Contrastive Learning for Electrocardiograms to Detect Left Ventricular Systolic Dysfunction

〇Mitsuhiko Nakamoto¹, Satoshi Kodera¹, Hirotoshi Takeuchi¹, Shinnosuke Sawano¹, Susumu Katsushika¹, Kota Ninomiya¹, Hiroshi Akazawa¹, Issei Komuro¹ (1. Department of Cardiovascular Medicine, The University of Tokyo Hospital)

[[Online]]

Keywords:self-supervised learning, contrastive learning, electrocardiography

Self-supervised learning has been demonstrated to be a powerful way to use unlabeled data in computer vision tasks. In this study, we propose a self-supervised pretraining approach to improve the performance of deep learning models that detect left ventricular systolic dysfunction from 12-lead electrocardiography data. We first pretrain an encoder that can extract rich features from unlabeled electro- cardiography data using self-supervised contrastive learning, and then fine-tune the model on the downstream dataset using the pretrained encoder. In experiments, our proposed approach achieved higher performance than the supervised baseline method, using only 28% of the labels used by the baseline method.