Contrast-enhanced CT Radiomics for High-Grade Pattern Identification and Prognostic Stratification in Lung Adenocarcinoma with Consolidation-to-Tumor Ratio of 25% or More.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, Tianjin Medical University Cancer Institute and Hospital, National Clinical Research Center for Cancer/Tianjin's Clinical Research Center for Cancer/Key Laboratory of Cancer Prevention and Therapy, Huanhuxi Rd, Tiyuanbei, Hexi District, Tianjin 300060, China.
- Department of Radiology, The Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital, HNHC Key Laboratory of Oncology Medical Imaging Response Assessment, Zhengzhou, China.
- Department of Radiology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China.
Abstract
Background Radiomics may preoperatively identify high-grade patterns (HGPs) in lung adenocarcinoma (ADC) and assist in clinical decision-making. Purpose To develop and evaluate a machine learning model based on preoperative contrast-enhanced CT images to predict HGPs and explore the model's prognostic value. Materials and Methods Patients with clinical stage I invasive ADC who underwent surgery (January 2017 to May 2025) were retrospectively enrolled from three centers. Binary (low risk: HGPs < 20%; high risk: HGPs ≥ 20%) and ternary (HGP0: HGPs = 0; HGP1: 0 < HGPs < 20%; HGP2: HGPs ≥ 20%) classification analyses were performed based on the proportion of HGPs. Multivariable logistic regression analysis was used to determine independent predictors of HGPs. XGBoost classifier-based radiomic models and combined models (radiomics-predicted probabilities plus clinical variables plus CT semantic features) were constructed and evaluated for discriminability, calibration ability, and clinical utility. Kaplan-Meier and Cox regression analyses were conducted to identify prognostic factors for overall survival (OS) and recurrence-free survival (RFS). Results A total of 1181 patients (median age, 61 years [IQR, 54-66 years]; 694 female) were allocated to the training (<i>n</i> = 667), internal test (<i>n</i> = 279), and external test (<i>n</i> = 235) sets. The combined model achieved the best discrimination in both binary (training: area under the receiver operating characteristic curve [AUC], 0.87 [95% CI: 0.84, 0.90]; internal test: AUC, 0.80 [95% CI: 0.74, 0.85]; external test: AUC, 0.84 [95% CI: 0.78, 0.90]) and ternary (training: microaverage AUC, 0.80 [95% CI: 0.78, 0.82]; internal test: microaverage AUC, 0.74 [95% CI: 0.70, 0.77]; external test: microaverage AUC, 0.72 [95% CI: 0.68, 0.76]) classification analyses. Model-predicted high-risk group was an independent prognostic factor for both OS (binary: hazard ratio [HR] = 1.98, <i>P</i> =.04; ternary: HR = 2.93, <i>P</i> =.03) and RFS (binary: HR = 3.33, <i>P</i> < .001; ternary: HR = 5.10, <i>P</i> < .001) and was consistently confirmed across subgroup analyses. Conclusion The combined model, integrating clinical variables, CT semantic features, and radiomics-predicted probabilities, effectively predicted high-grade patterns in lung ADC and showed strong potential for prognostic risk stratification. © RSNA, 2026 <i>Supplemental material is available for this article.</i> See also the editorial by Arita and Kocak in this issue.