Scholarlink

<p>Thyroid-associated ophthalmopathy (TAO) is an organ-specific autoimmune disease that severely affects sufferers’ health&#xa0;and&#xa0;life. Clinical activity score (CAS) is one of the significant methods for the early diagnosis of TAO. However, the acquisition of CAS scores relies heavily on the clinician’s subjective experience. Accurate identification of TAO regions segmented by scientific techniques is one of the essential prerequisites for the objective acquisition of the CAS scores. But the currently proposed models have shortcomings of high label annotation costs, etc. Therefore, a cross-scale Gaussian wavelet model with decomposed cross-consistency (CGWM) is proposed for TAO multifocal region semi-supervised segmentation. First, the encoder-decoder dual-tree complex wavelet model is employed for the interpretable extraction of the robust multi-directional features of the diseased region and the resolution recovery. Subsequently, a cross-scale Gaussian hybrid attention mechanism is developed for flexible fine multi-scale contextual feature extraction of unlabeled images by introducing the Gaussian probability model. Finally, a novel and simple decomposition cross-consistency with a decomposition penalty is proposed for further strengthening the consistency between the two segmentation networks and ensuring that the prediction decision boundary is more accurately located in the low-density regions. Compared to other selected benchmark models and fully supervised segmentation techniques, extensive experiments showed that the CGWM achieved state-of-the-art segmentation performance with an accuracy of 84.21% using only 100 tagged images for training, with good prospects for clinical application due to its low annotation cost and strong interpretability.</p>

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

A cross-scale Gaussian wavelet model with decomposition cross-consistency for TAO multifocal region semi-supervised segmentation

  • 期刊论文

  • 发表日期：

    2025年4月1日

查看全文
  • Haipeng Zhu,
  • Hong He,
  • Xuefei Song

摘要

Thyroid-associated ophthalmopathy (TAO) is an organ-specific autoimmune disease that severely affects sufferers’ health and life. Clinical activity score (CAS) is one of the significant methods for the early diagnosis of TAO. However, the acquisition of CAS scores relies heavily on the clinician’s subjective experience. Accurate identification of TAO regions segmented by scientific techniques is one of the essential prerequisites for the objective acquisition of the CAS scores. But the currently proposed models have shortcomings of high label annotation costs, etc. Therefore, a cross-scale Gaussian wavelet model with decomposed cross-consistency (CGWM) is proposed for TAO multifocal region semi-supervised segmentation. First, the encoder-decoder dual-tree complex wavelet model is employed for the interpretable extraction of the robust multi-directional features of the diseased region and the resolution recovery. Subsequently, a cross-scale Gaussian hybrid attention mechanism is developed for flexible fine multi-scale contextual feature extraction of unlabeled images by introducing the Gaussian probability model. Finally, a novel and simple decomposition cross-consistency with a decomposition penalty is proposed for further strengthening the consistency between the two segmentation networks and ensuring that the prediction decision boundary is more accurately located in the low-density regions. Compared to other selected benchmark models and fully supervised segmentation techniques, extensive experiments showed that the CGWM achieved state-of-the-art segmentation performance with an accuracy of 84.21% using only 100 tagged images for training, with good prospects for clinical application due to its low annotation cost and strong interpretability.