Abstract
BACKGROUND AND PURPOSE: Peritoneal metastasis significantly impacts prognosis and treatment strategies in ovarian cancer. Traditional imaging techniques have limited sensitivity in preoperative detection. Radiomics-based machine learning models offer a promising non-invasive approach to improve diagnostic accuracy. This study systematically reviews and meta-analyzes their predictive performance. MATERIALS AND METHODS: A systematic search was conducted up to June 2025 for studies that developed and validated machine learning models based on radiomic features for the prediction of peritoneal metastasis in ovarian cancer. Quality of included studies was evaluated using the METRICS and QUADAS-2 tool. Pooled sensitivity, specificity, and area under the curve (AUC) were calculated via bivariate random-effects meta-analysis. Heterogeneity, sensitivity, and publication bias analyses were also conducted. RESULTS: Six studies were included in the systematic review and qualitative synthesis, with five studies comprising 448 individual participants derived exclusively from validation cohorts meeting criteria for meta-analysis. Radiomics models yielded a pooled AUC of 0.81 (95% CI: 0.71–0.88), accompanied by a sensitivity of 73% and specificity of 77%. Clinical models demonstrated a pooled AUC of 0.82 (95% CI: 0.79–0.85). The pooled AUC for the combined clinical–radiomics models was 0.87 (95% CI: 0.83–0.89). Methodological quality assessed via METRICS ranged from 64.5% to 86.3%, with a mean score of 77.6%. The QUADAS-2 assessment indicated a low overall risk of bias, with minor concerns noted in the patient selection and index test domains. No significant heterogeneity, publication bias, or outliers were detected. CONCLUSION: Radiomics-based machine learning models demonstrate potential as tools for the prediction of peritoneal metastasis in ovarian cancer and may assist in preoperative risk stratification. Further large-scale, multicenter prospective studies with standardized methodologies and external validation are necessary to confirm clinical applicability. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12880-025-02068-3.