Jia Qing-Chun, Niu Ye, Xuan Qi-Fan, Miao Shi-di, Huang Wen-Juan, Liu Ping-Ping, Liu Le, Xie Han-Bing, Wang Qiu-Jun, Liu Zeng-Yao, Fu Shuang, Liu Yu-Xi, Zhao Lin, Li Yuan-Zhou, Wang Rui-Tao
Department of Internal Medicine, Harbin Medical University Cancer Hospital, Harbin Medical University, Harbin, Heilongjiang 150081, China.
School of Computer Science and Technology, Harbin University of Science and Technology, Harbin, Heilongjiang 150008, China.
iScience. 2024 Aug 19;27(10):110733. doi: 10.1016/j.isci.2024.110733. eCollection 2024 Oct 18.
We aimed to build and validate a computed tomography (CT)-based deep learning nomogram for discriminating granulomas from lung adenocarcinomas. A retrospective study of 1,159 patients with solitary lung nodules from three institutions in China who underwent pre-operative lung CT scans was performed. The patients were divided into one training, one validation, one test, and two external validation cohorts. Deep learning features were extracted from CT images. The least absolute shrinkage and selection operator (LASSO) regression model was used for dimension reduction and feature selection. Logistic regression analysis showed that age, gender, intranodular and perinodular (IPN) features, and adipose features were the significant predictors of malignancy presence (all < 0.05). The nomogram was built by incorporating these four factors and achieved better diagnostic accuracy than the single-factor model. The nomogram demonstrates satisfactory discrimination and calibration. In addition, decision curve analysis revealed the considerable clinical usefulness of the nomogram.
我们旨在构建并验证一种基于计算机断层扫描（CT）的深度学习列线图，用于鉴别肉芽肿与肺腺癌。对来自中国三个机构的1159例接受术前肺部CT扫描的孤立性肺结节患者进行了一项回顾性研究。将患者分为一个训练队列、一个验证队列、一个测试队列和两个外部验证队列。从CT图像中提取深度学习特征。采用最小绝对收缩和选择算子（LASSO）回归模型进行降维和特征选择。逻辑回归分析显示，年龄、性别、结节内及结节周围（IPN）特征和脂肪特征是恶性肿瘤存在的显著预测因素（均P<0.05）。通过纳入这四个因素构建列线图，其诊断准确性优于单因素模型。该列线图显示出令人满意的鉴别能力和校准效果。此外，决策曲线分析揭示了该列线图具有相当大的临床实用性。
