CT radiomics for noninvasive prediction of histologic differentiation in gastric cancer.
Authors
Affiliations (4)
Affiliations (4)
- Department of Chinese Integrative Medicine Oncology, The First Affiliated Hospital of Anhui Medical University, Hefei, Anhui, China.
- Department of Respiratory, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, Anhui, China.
- Department of Radiology, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, Anhui, China.
- Department of Geriatrics, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, Anhui, China.
Abstract
The histological differentiation grade of gastric cancer critically influences treatment and prognosis. While CT radiomics shows promise for noninvasive prediction, its relationship with gene expression remains unclear. This study aimed to develop a clinical-radiomics model for predicting tumor differentiation in gastric cancer patients and to explore the underlying mechanisms. We retrospectively analyzed clinical data and CT images from 162 gastric cancer patients, who were randomly assigned to training and validation cohorts. The least absolute shrinkage and selection operator (LASSO) method was used to select features and construct the Rad-score. Subsequently, clinical-radiomics models were built and evaluated for their predictive efficacy and clinical incremental value. Furthermore, hub genes were screened, and their associated pathways were investigated using machine learning, bioinformatics analysis, and experimental validation. A clinical-radiomics model based on N stage, M stage and Rad-score was developed. The receiver operating characteristic (ROC) curves indicated that the model had preliminary evidence of predictive potential within this single-center cohort (training AUC = 0.872, validation AUC = 0.935). The calibration curves indicated a reasonable concordance between the observed values and the predicted outcomes in this retrospective sample. The decision curve analysis demonstrated a net benefit that requires further confirmation in external cohorts. The clinical impact curve (CIC) demonstrated the model's potential clinical applicability, which warrants validation in prospective settings. Sequencing data further revealed that the key gene IGHG1 was significantly associated with the Rad-score, with potential mechanisms involving the TGF-beta signaling pathway. The clinical-radiomics model, incorporating N stage, M stage, and Rad-score, serves as a preliminary research tool for assessing tumor differentiation in gastric cancer patients within a single-center setting. External validation is required before any consideration of clinical generalization. Radiomics enables noninvasive evaluation of differentiation status while generating hypotheses regarding its underlying mechanisms.