Intratumoral and peritumoral radiomics for forecasting microsatellite status in gastric cancer: a multicenter study.

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Accurate preoperative prediction of microsatellite instability (MSI) status in gastric cancer (GC) is crucial for personalized treatment.
• CT-derived radiomics offers a non-invasive method to analyze tumor characteristics.

Purpose of the Study:

• To evaluate the effectiveness of CT-derived peritumoral and intratumoral radiomics in predicting MSI status preoperatively in GC patients.
• To develop and validate predictive models integrating radiomics and clinical data.

Main Methods:

• Retrospective analysis of 364 GC patients with contrast-enhanced CT scans.
• Extraction of radiomics features from intratumoral (IR) and intratumoral plus peritumoral (IPR) regions.
• Development of six radiomic models using Support Vector Machine (SVM), Linear Support Vector Classification (LinearSVC), and Logistic Regression (LR).
• Creation of a combined model integrating radiomics score (Radscore) with clinical and CT semantic features.

Main Results:

• The LinearSVC model using IPR achieved an Area Under the Curve (AUC) of 0.802 in external validation.
• The combined model demonstrated superior AUCs (0.891 internal, 0.856 external) compared to radiomics and clinical models alone.
• Calibration plots and Decision Curve Analysis (DCA) confirmed the combined model's clinical utility and relevance.

Conclusions:

• A combined model integrating IPR radiomics with clinical characteristics accurately predicts MSI status in GC.
• This approach supports the development of personalized treatment strategies for GC patients based on their MSI status.