The predictive value of radiomics-based machine learning for peritoneal metastasis in gastric cancer patients: a systematic review and meta-analysis.

Background: For patients with gastric cancer (GC), effective preoperative identification of peritoneal metastasis (PM) remains a severe challenge in clinical practice. Regrettably, effective early identification tools are still lacking up to now. With the popularization and application of radiomics method in tumor management, some researchers try to introduce it into the early identification of PM in patients with GC. However, due to the complexity of radiomics, the value of radiomics method in the early identification of PM in GC patients remains controversial. Therefore, this systematic review was conducted to explore the feasibility of radiomics in the early identification of PM in GC patients. Methods: PubMed, Cochrane, Embase and the Web of Science were comprehensively and systematically searched up to 25 July, 2022 (CRD42022350512). The quality of the included studies was assessed using the radiomics quality score (RQS). To discuss the superiority in diagnostic accuracy of radiomics-based machine learning, a subgroup analysis was performed by machine learning (ML) based on clinical features, radiomics features, and radiomics + clinical features. Results: Finally, 11 eligible original studies covering 78 models were included in this systematic review. According to the meta-analysis, the radiomics + clinical features model had a c-index of 0.919 (95% CI: 0.871-0.969), pooled sensitivity and specificity of 0.90 (0.83-0.94) and 0.87 (0.78-0.92), respectively, in the training set, and a c- index of 0.910 (95% CI: 0.886-0.934), pooled sensitivity and specificity of 0.78 (0.71-0.84) and 0.83 (0.74-0.89), respectively, in the validation set. Conclusions: The ML methods based on radiomics + clinical features had satisfactory accuracy for the early diagnosis of PM in GC patients, and can be used as an auxiliary diagnostic tool for clinicians. However, the lack of guidelines for the proper operation of radiomics has led to the diversification of radiomics methods, which seems to limit the development of radiomics. Even so, the clinical application value of radiomics cannot be ignored. The standardization of radiomics research is required in the future for the wider application of radiomics by developing intelligent tools of radiomics. Systematic review registration: https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=350512, identifier CRD42022350512.