Imaging-based stratification of Claudin 18.2-positive gastric adenocarcinoma using <sup>18</sup>F-FDG PET/CT radiomics.
Authors
Affiliations (6)
Affiliations (6)
- Department of Nuclear Medicine, Zhejiang Cancer Hospital, Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences, Hangzhou, Zhejiang, 310022, China.
- Postgraduate Training Base Alliance of Wenzhou Medical University (Zhejiang Cancer Hospital), WenZhou, Zhejiang, 325000, China.
- Department of Nuclear Medicine, Aksu Prefecture First People's Hospital, Aksu, Xinjiang, 843000, China.
- Department of Nuclear Medicine, Aksu Prefecture First People's Hospital, Aksu, Xinjiang, 843000, China. [email protected].
- Department of Nuclear Medicine, Zhejiang Cancer Hospital, Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences, Hangzhou, Zhejiang, 310022, China. [email protected].
- Postgraduate Training Base Alliance of Wenzhou Medical University (Zhejiang Cancer Hospital), WenZhou, Zhejiang, 325000, China. [email protected].
Abstract
This study aimed to develop and validate a non-invasive, multimodal radiomics model based on preoperative <sup>1</sup>⁸F-FDG PET/CT to predict CLDN18.2 expression in gastric adenocarcinoma (GAC), addressing the limitations of intratumoral heterogeneity and invasiveness associated with endoscopic biopsies. This retrospective study enrolled 291 patients with pathologically confirmed GAC who underwent preoperative <sup>1</sup>⁸F-FDG PET/CT. The cohort was randomly divided into a training set (n = 204) and an independent validation set (n = 87). High-dimensional radiomic features were extracted from PET and CT images. Feature selection was performed using the minimum redundancy maximum relevance (mRMR) algorithm and LASSO regression. A radiomics signature (Rad-score) was constructed using XGBoost and integrated with clinical variables. Among five machine learning algorithms evaluated, AdaBoost was identified as the optimal model. Performance was assessed via receiver operating characteristic (ROC) analysis, and interpretability was visualized using SHapley Additive exPlanations (SHAP). CLDN18.2-positive tumors exhibited a distinct hypometabolic phenotype, characterized by significantly lower SUVmax (P = 0.012) and SUVmean (P < 0.001) compared to negative tumors. The combined multimodal model demonstrated superior discrimination, achieving an AUC of 0.926 (95% CI: 0.891-0.961) in the training cohort and 0.765 (95% CI: 0.665-0.865) in the validation cohort, consistently outperforming single-modality models. SHAP analysis highlighted PET-derived radiomic features and serum CEA as the most influential predictors, confirming the inverse correlation between metabolic activity and CLDN18.2 expression. The developed <sup>1</sup>⁸F-FDG PET/CT-based multimodal radiomics model serves as an effective, non-invasive tool for predicting CLDN18.2 expression in GAC. The study validates the hypometabolic nature of CLDN18.2-positive tumors, suggesting this radiomics approach can effectively complement traditional biopsy methods for patient stratification.