Multiperspective tumor heterogeneity metrics for preoperative prediction of IASLC grading in clinical stage IA lung adenocarcinomas: A multicenter study.
Authors
Affiliations (6)
Affiliations (6)
- School of Mathematics and Computational Science, Xiangtan University, Xiangtan, 411105, Hunan, China; Hunan National Applied Mathematics Center, Xiangtan, 411105, Hunan, China; Hunan Key Laboratory for Computation and Simulation in Science and Engineering, Xiangtan, 411105, Hunan, China; Key Laboratory of Intelligent Computing & Information Processing of Ministry of Education, Xiangtan, 411105, Hunan, China. Electronic address: [email protected].
- College of Mathematical Medicine, Zhejiang Normal University, Jinhua, 321004, Zhejiang, China. Electronic address: [email protected].
- Department of Radiology, Xiangtan Central Hospital, Xiangtan, 411100, Hunan, China. Electronic address: [email protected].
- Department of Radiology, the Affiliated Hospital of Southwest Medical University, Luzhou, 646099, Sichuan, China. Electronic address: [email protected].
- Department of Radiology, The Third Xiangya Hospital of Central South University, Changsha, 410013, Hunan, China. Electronic address: [email protected].
- School of Mathematics and Computational Science, Xiangtan University, Xiangtan, 411105, Hunan, China; Hunan National Applied Mathematics Center, Xiangtan, 411105, Hunan, China; Hunan Key Laboratory for Computation and Simulation in Science and Engineering, Xiangtan, 411105, Hunan, China; Key Laboratory of Intelligent Computing & Information Processing of Ministry of Education, Xiangtan, 411105, Hunan, China. Electronic address: [email protected].
Abstract
Accurate preoperative prediction of the International Association for the Study of Lung Cancer (IASLC) grades is crucial for devising personalized treatment strategies for lung adenocarcinoma. The aim of this study was to develop and validate an integrative framework for preoperative prediction of pathological grades. A multicenter computed tomography (CT) imaging dataset, including 1226 patients with stage IA lung adenocarcinoma from Datasets 1-3, was analyzed. The cohort was randomly divided into a training set (n=858, 70%) and a test set (n=368, 30%). Additionally, an independent external validation set, Dataset 4 (n=266), was employed to evaluate the model's generalizability rigorously. We proposed a set of multiperspective tumor heterogeneity metrics by integrating local features and global pixel distribution patterns on CT images. Subsequently, a comprehensive assessment of 11 machine-learning models was performed, integrating clinicoradiological features with these novel tumor heterogeneity metrics to construct the integrative clinicoradiological multiperspective heterogeneity hybrid machine-learning framework (ICMH-HF). The light gradient boosting machine demonstrated superior performance and was subsequently selected as the core classifier for the ICMH-HF model. This choice underpinned remarkable predictive accuracy, with area under the receiver operating characteristic curve (AUC) values of 0.835 in the test set and 0.786 in the external validation set. SHAP-based interpretability analysis identified four key predictors that collectively enhanced the model's robustness. The ICMH-HF effectively combines quantitative clinicoradiological features with multiperspective tumor heterogeneity metrics. The proposed framework provides a potent and clinically applicable tool for the preoperative discrimination between Grade 1 and Grades 2-3 lung adenocarcinomas according to the IASLC grading system.