A CT-based Machine Learning Radiomics Model for the Prediction of Gastric Cancer Differentiation and Mechanism Exploration

Objectives To develop a CT-based radiomics model for predicting tumor differentiation in patients with gastric cancer. Exploring Rad-score correlation with gene expression and related mechanisms. Materials and Methods Clinical data and imaging of 162 gastric cancer patients were retrospectively analyzed. Patients were randomly allocated to training and validation cohorts. The least absolute shrinkage and selection operator (LASSO) methods were utilized to identify characteristics and develop the Rad-score. Clinical-radiomics models were developed and evaluated for predictive efficacy and clinical incremental value. Screening hub genes and exploring the pathways of hub genes through machine learning, bioinformatics analysis and experimental validation. Results Clinical-radiomics models based on N stage, M stage and Rad-score were developed. The receiver operating characteristic (ROC) curves indicated that the model had good predictive accuracy in the training (AUC = 0.872) and validation groups (AUC = 0.935). The calibration curves indicated a strong correlation between the observed values and the predicted outcomes. The decision curve analysis demonstrated a substantial net benefit associated with the clinical-radiomics model. The clinical impact curve (CIC) illustrated the effective clinical applicability of this model. Analysis of the sequencing data revealed that the key gene IGHG1 was significantly associated with Rad-score. The possible mechanisms are related to the TGF-β signaling, epithelial-mesenchymal transition and KRAS signaling pathway. Conclusions The predictive model based on N stage, M stage and Rad-score can effectively predict the differentiation in gastric cancer patients. Radiomics enables noninvasive prediction of tumor differentiation status while elucidating the expression levels of the IGHG1 and the underlying pathway.