¹⁸F-FDG PET/CT Radiomics-Based Multimodality Fusion Model for Preoperative Individualized Noninvasive Prediction of...

Area of Science:

• Oncology
• Radiology
• Machine Learning

Background:

• Advanced gastric cancer (AGC) poses significant challenges in predicting peritoneal metastasis (PM).
• Accurate prediction of PM is crucial for treatment planning and patient outcomes in AGC.
• Current diagnostic methods may have limitations in detecting PM effectively.

Purpose of the Study:

• To develop and validate a multimodality fusion (MMF) model for predicting PM in AGC.
• To integrate 18F-fluorodeoxyglucose (FDG) PET/CT radiomics, clinical factors, and expert diagnoses using machine learning.
• To enhance the individual prediction accuracy of PM in advanced gastric cancer patients.

Main Methods:

• A cohort of 167 patients with AGC undergoing preoperative PET/CT and surgery was analyzed.
• PET/CT radiomic signatures were extracted using classic and kernelled support tensor machine (KSTM) models.
• A multimodality fusion (MMF) model was established by integrating PET/CT signatures, a clinical nomogram, and expert diagnoses via evidential reasoning.

Main Results:

• The MMF model demonstrated high predictive performance with an AUC of 94.16% (training) and 90.84% (testing).
• The MMF model significantly outperformed clinical nomograms and expert diagnoses in prediction accuracy.
• The model showed robust generalization, even for challenging subtypes like mucinous adenocarcinoma and signet ring cell carcinoma.

Conclusions:

• The 18F-FDG PET/CT radiomics-based MMF model shows significant clinical utility for predicting PM in AGC.
• Combining information from multiple modalities is essential for comprehensive and accurate PM prediction.
• This approach offers a promising tool for improving the management of advanced gastric cancer.