Scholarlink

<p>The diagnosis of thyroid nodules usually relies on the professional experience of physicians, which can easily lead to misdiagnosis and missed diagnoses. This paper introduces an intelligent diagnosis model for thyroid nodules, building upon a hybrid framework of U-Net and residual networks. Experimental results show that the average diagnostic accuracy of this model for the thyroid gland in the TN-SCUI 2020 dataset (4554 images, with a ratio of 8:1:1 for training set, validation set and test set) is greater than 92%, and the highest single-class accuracy rate reaches 95%. At the same time, the Dice index and Jaccard index of the model in the TN3K dataset (3493 images, with a ratio of 8:1:1 for training set, validation set and test set) can reach up to 86.16% and 85.28%, and the diagnostic accuracy is improved by 6% compared to the baseline model. The novelty of this article lies in the deep integration of U-shaped network++, residual network, and Transformer, which fills the gap in global semantic modeling and dynamic weight optimization. However, the proposed model still needs prospective clinical validation, and its adaptability to complex scenarios needs to be improved. In the future, more external validations and lightweight experiments will be conducted.</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

An intelligent diagnosis method for thyroid nodules using UNet++ integrated with ResNet and transformer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月18日

查看全文
  • Minghui Guo

摘要

The diagnosis of thyroid nodules usually relies on the professional experience of physicians, which can easily lead to misdiagnosis and missed diagnoses. This paper introduces an intelligent diagnosis model for thyroid nodules, building upon a hybrid framework of U-Net and residual networks. Experimental results show that the average diagnostic accuracy of this model for the thyroid gland in the TN-SCUI 2020 dataset (4554 images, with a ratio of 8:1:1 for training set, validation set and test set) is greater than 92%, and the highest single-class accuracy rate reaches 95%. At the same time, the Dice index and Jaccard index of the model in the TN3K dataset (3493 images, with a ratio of 8:1:1 for training set, validation set and test set) can reach up to 86.16% and 85.28%, and the diagnostic accuracy is improved by 6% compared to the baseline model. The novelty of this article lies in the deep integration of U-shaped network++, residual network, and Transformer, which fills the gap in global semantic modeling and dynamic weight optimization. However, the proposed model still needs prospective clinical validation, and its adaptability to complex scenarios needs to be improved. In the future, more external validations and lightweight experiments will be conducted.