CT-Based Radiomics Showing Generalization to Predict Tumor Regression Grade for Advanced Gastric Cancer Treated With...

Area of Science:

• Oncology
• Medical Imaging
• Radiomics

Background:

• Advanced gastric cancer (AGC) requires effective neoadjuvant therapies.
• Predicting treatment response is crucial for optimizing patient outcomes.
• Radiomics offers potential for non-invasive assessment of treatment efficacy.

Purpose of the Study:

• To develop and validate a radiomics model for predicting treatment response in AGC patients.
• To assess the model's generalization across different neoadjuvant chemotherapy (NAC) and targeted therapy regimens.
• To identify radiomics features that correlate with tumor regression grade (TRG).

Main Methods:

• 373 AGC patients from five cohorts received neoadjuvant therapies (NAC or targeted therapy).
• CT scans were analyzed to extract 2,452 radiomics features.
• Mutual information and random forest algorithms were used for feature selection and model development to predict TRG.

Main Results:

• A radiomics model using 28 features demonstrated generalizability in predicting TRG for patients receiving NAC across four cohorts (AUCs ranging from 0.72 to 0.82).
• No significant difference in predictive performance was observed among NAC regimens (p > 0.05).
• The model showed poor predictive value for the targeted therapy cohort (SOXA), significantly worse than the training cohort (p = 0.010).

Conclusions:

• Radiomics is generalizable for predicting TRG in AGC patients undergoing NAC.
• This approach can assist in treatment selection, particularly for patients insensitive to NAC.
• Further research is needed to refine models for predicting response to targeted therapies.