Identifying sinonasal inverted papilloma by machine learning: a systematic review and meta-analysis

Background Sinonasal inverted papilloma (IP) is a benign tumor of the sinonasal mucosa, which may become malignant. Machine learning (ML) has been applied to improve the accuracy in the diagnosis of various diseases, but no studies have evaluated the performance of ML for IP diagnosis. This systematic review and meta-analysis aimed to explore the diagnostic performance of ML for IP. Methods We systematically searched articles from PubMed, Cochrane, Embase, and Web of Science up to July 22, 2025. The quality assessment of diagnostic accuracy studies tool (QUADAS-2) was used to assess the risk of bias, and the bivariate mixed-effect model was used for meta-analysis. Results 17 studies involving 3321 participants were included. In the validation set, the sensitivity and specificity of ML constructed based on radiomics for identifying IP and malignant tumors were 0.84 (95%CI: 0.77-0.89) and 0.82 (95% CI: 0.74 ~ 0.88), respectively. The sensitivity and specificity of ML constructed based on radiomics and clinical features for identifying IP and malignant tumors were 0.85 (95%CI: 0.78-0.90) and 0.87 (95% CI: 0.80 ~ 0.92), respectively. Conclusion Our study shows that ML has a favorable performance in the differential diagnosis of IP. More prospective studies are needed to validate and develop universal tools. Systemic Review Registration https://www.crd.york.ac.uk/PROSPERO/view/CRD42023430417 , identifier CRD42023430417.