Scholarlink

Brugada syndrome, a rare hereditary cardiac condition, poses significant challenges in diagnosis due to the scarcity of available data. This study explores the application of a self-supervised contrastive learning framework (SSL) to enhance the detection of Brugada syndrome using ECG data. We hypothesize that SSL can effectively leverage large, open-access datasets to improve diagnostic accuracy for this rare condition. Our approach employs the CinC2020 dataset to pretrain the SSL EfficientNet backbone architecture. We compare the performance of this SSL-pretrained model against its supervised counterpart trained directly on Brugada syndrome data. The results demonstrate the superiority of SSL-based approach. EfficientNet with SSL pretraining achieved an AUROC of 84.2% compared to 70.2% for its supervised counterpart. These findings highlight the potential of SSL in overcoming data scarcity challenges for rare medical conditions. By leveraging large, diverse ECG datasets, SSL enables more robust feature learning, improving overall generalization capacity. Our work holds promise for enhancing diagnostic tools for Brugada syndrome and potentially other rare cardiac disorders, ultimately leading to improved patient care and prognosis.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Brugada Syndrome Diagnosis Through Self-supervised Contrastive Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Beatrice Zanchi,
  • Giuliana Monachino,
  • Luigi Fiorillo,
  • Giulio Conte,
  • Francesca Dalia Faraci

摘要

Brugada syndrome, a rare hereditary cardiac condition, poses significant challenges in diagnosis due to the scarcity of available data. This study explores the application of a self-supervised contrastive learning framework (SSL) to enhance the detection of Brugada syndrome using ECG data. We hypothesize that SSL can effectively leverage large, open-access datasets to improve diagnostic accuracy for this rare condition. Our approach employs the CinC2020 dataset to pretrain the SSL EfficientNet backbone architecture. We compare the performance of this SSL-pretrained model against its supervised counterpart trained directly on Brugada syndrome data. The results demonstrate the superiority of SSL-based approach. EfficientNet with SSL pretraining achieved an AUROC of 84.2% compared to 70.2% for its supervised counterpart. These findings highlight the potential of SSL in overcoming data scarcity challenges for rare medical conditions. By leveraging large, diverse ECG datasets, SSL enables more robust feature learning, improving overall generalization capacity. Our work holds promise for enhancing diagnostic tools for Brugada syndrome and potentially other rare cardiac disorders, ultimately leading to improved patient care and prognosis.