Related Experiment Video
Updated: Aug 6, 2026
Predictive Immune Modeling of Solid Tumors
Published on: February 25, 2020
CT-based Intratumoral Heterogeneity Score for Preoperative Prediction of Lymphovascular Invasion and Prognosis in
Zhichao Zuo1, Yunhua Li2, Xuehong Long3
1Department of Radiology, The Central Hospital of Xiangtan (The Affiliated Hospital of Hunan University), Xiangtan, 411100, Hunan, China. zuo_z@smail.xtu.edu.cn.
Background:
Lymphovascular invasion (LVI) is a key determinant of recurrence and survival in lung adenocarcinoma (LUAD), but preoperative identification remains challenging. We developed and validated a CT-based intratumoral heterogeneity (ITH) score for LVI prediction, prognostic stratification, and exploratory radiogenomic interpretation.
Methods:
We retrospectively included 1322 patients from three institutions and one public dataset. The discovery cohort was divided into training (n = 590) and internal validation (n = 253) sets; an independent cohort (n = 362) was used for external validation and prognostic analysis. A Stanford Radiogenomics set (n = 117) was used for exploratory transcriptomic analysis. Multiple machine learning algorithms were trained using the same feature matrix and cohort partitions.
Results:
Random forest was selected as the final classifier, with AUCs of 0.886 (95% confidence interval [CI] 0.842-0.923) in internal validation and 0.850 (95% CI 0.806-0.892) in external validation. The model showed acceptable calibration and positive net benefit across low-to-intermediate threshold probabilities. High ITH remained associated with worse recurrence-free survival (hazard ratio [HR] 12.92; 95% CI 3.90-42.78; P < 0.001) and overall survival (HR 26.47; 95% CI 3.52-198.76; P = 0.001). Exploratory radiogenomic analysis showed enrichment of hypoxia-, angiogenesis-, epithelial-mesenchymal transition-, and glycolysis-related pathways.
Conclusions:
The CT-based ITH score may support preoperative risk stratification for LVI and prognosis in LUAD. Radiogenomic findings suggest potential biological correlates and should be considered hypothesis-generating.