ASO Author Reflections: Preoperative Noninvasive Prediction of Peritoneal Metastasis in Advanced Gastric Cancer: A ¹⁸F-FDG PET/CT Radiomics-Based Multimodality Fusion Model

Accurate noninvasive prediction of peritoneal metastasis (PM) in advanced gastric cancer (AGC) is clinically valuable for avoiding unnecessary surgical exploration. While ¹⁸F-fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) is useful for cancer staging, it has challenges detecting PM in gastric mucinous adenocarcinoma and signet ring cell carcinoma (MC and SRCC) due to poor uptake. Radiomics, which could extract high-throughput hand-crafted features from PET/CT images, had shown value in PM prediction^1,2. Additionally, a model that automatically learns image features, such as our previously proposed kernelled support tensor machine (KSTM)³, is capable of capturing unique characteristics for specific prediction tasks. The construction of PET/CT image signatures using both hand-crafted and self-learning features, as well as integrating them with clinical factors and experts’ diagnoses on the basis of PET/CT images, shows promise in accurately predicting PM in AGC. …