Scholarlink

<p>Widely implemented lung cancer screening computed tomography (CT) provides a valuable opportunity for screening chronic obstructive pulmonary disease (COPD). This study aims to develop a deep learning model based on the Swin Transformer architecture to detect and stage COPD according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria using CT scans obtained from lung cancer screening. This study presents a new framework for COPD detection, staging, and severity classification. The model employs the Swin Transformer algorithm to integrate and analyze preprocessed CT images, emphysema features, and LAV-950% (percentage of lung volumes of less than or equal to –950 Hounsfield units). For model training, a dataset of 637 pulmonary nodule patients was prospectively and retrospectively collected from the Pulmonary Nodule Clinic of the First Affiliated Hospital of Chongqing Medical University between January 1, 2019, and September 30, 2023. The dataset was randomly divided into training and testing sets in a 4:1 ratio, maintaining balanced proportions across COPD severity levels. External validation was also performed using 1464 CT scans from the National Lung Screening Trial (NLST) cohort. For COPD detection, the model achieved an AUC of 0.829 (95%CI 0.764, 0.894), an F1 score of 0.801 (95%CI 0.732, 0.870), and an accuracy of 0.822 (95%CI 0.756, 0.888). In severity classification, the model obtained an F1 score of 0.763 (95%CI 0.690, 0.836) and an accuracy of 0.791 (95%CI 0.720, 0.861). For COPD staging, the F1 score and accuracy were 0.561 (95%CI 0.475, 0.646) and 0.789 (95%CI 0.718, 0.859), respectively. The proposed COPD-TransNet, which integrates Swin Transformer and LAV-950 features, outperformed mainstream methods in COPD detection (AUC 0.829 vs. 0.805–0.823). External validation on the NLST cohort (AUC 0.867) further demonstrates its robustness and clinical feasibility for COPD screening and staging based on chest CT.</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

COPD-TransNet: A Swin Transformer Network with Quantitative Emphysema Feature Fusion for COPD Detection and Staging from Opportunistic CT Scans

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Ao Liu,
  • Boyu Zhang,
  • Weiyi Li,
  • Min Ao,
  • Xinyu Xiong,
  • Junhao Mu,
  • Minghai Deng,
  • Qizhi Ye,
  • Xing Lu,
  • Li Yang

摘要

Widely implemented lung cancer screening computed tomography (CT) provides a valuable opportunity for screening chronic obstructive pulmonary disease (COPD). This study aims to develop a deep learning model based on the Swin Transformer architecture to detect and stage COPD according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria using CT scans obtained from lung cancer screening. This study presents a new framework for COPD detection, staging, and severity classification. The model employs the Swin Transformer algorithm to integrate and analyze preprocessed CT images, emphysema features, and LAV-950% (percentage of lung volumes of less than or equal to –950 Hounsfield units). For model training, a dataset of 637 pulmonary nodule patients was prospectively and retrospectively collected from the Pulmonary Nodule Clinic of the First Affiliated Hospital of Chongqing Medical University between January 1, 2019, and September 30, 2023. The dataset was randomly divided into training and testing sets in a 4:1 ratio, maintaining balanced proportions across COPD severity levels. External validation was also performed using 1464 CT scans from the National Lung Screening Trial (NLST) cohort. For COPD detection, the model achieved an AUC of 0.829 (95%CI 0.764, 0.894), an F1 score of 0.801 (95%CI 0.732, 0.870), and an accuracy of 0.822 (95%CI 0.756, 0.888). In severity classification, the model obtained an F1 score of 0.763 (95%CI 0.690, 0.836) and an accuracy of 0.791 (95%CI 0.720, 0.861). For COPD staging, the F1 score and accuracy were 0.561 (95%CI 0.475, 0.646) and 0.789 (95%CI 0.718, 0.859), respectively. The proposed COPD-TransNet, which integrates Swin Transformer and LAV-950 features, outperformed mainstream methods in COPD detection (AUC 0.829 vs. 0.805–0.823). External validation on the NLST cohort (AUC 0.867) further demonstrates its robustness and clinical feasibility for COPD screening and staging based on chest CT.