Scholarlink

In this paper, a new hybrid segmentation method based on SSL (semi-supervised learning) was developed for samples with image sequences, not all of which were labeled. Thus, this method can find application in areas where labeling is expensive or requires a certain specialist, such as in medicine. The developed method was evaluated on a sample of echocardiography images of patients with infective endocarditis in the context of a real-world task of segmenting heart valve anomalies. As a result, the accuracy gain compared to supervised learning is 5% in the IOU metric, while with other SSL methods it is on average 3%.

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

Semi-supervised Learning Based Image Sequence Segmentation Using Recurrent Autoencoder

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Victor Sineglazov,
  • Andrew Sheruda

摘要

In this paper, a new hybrid segmentation method based on SSL (semi-supervised learning) was developed for samples with image sequences, not all of which were labeled. Thus, this method can find application in areas where labeling is expensive or requires a certain specialist, such as in medicine. The developed method was evaluated on a sample of echocardiography images of patients with infective endocarditis in the context of a real-world task of segmenting heart valve anomalies. As a result, the accuracy gain compared to supervised learning is 5% in the IOU metric, while with other SSL methods it is on average 3%.