ObjectiveTo systematically review the diagnostic accuracy of MRI in preschool children with autism spectrum disorder. MethodsDatabases including Web of Science, PubMed, The Cochrane Library, EMbase, CBM, CNKI, WanFang Data, and VIP were electronically searched for studies on MRI in diagnosis of preschool children with autism spectrum disorder from inception to January 2022. Two reviewers independently screened the literature, extracted data, and assessed the risk of bias the included studies. Then, meta-analysis was performed using Meta-Disc, RevMan 5.4 and Stata 16.0 software. ResultsA total of 17 studies were finally included. The pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnosis odds ratio and area under the curve of SROC were 0.78 (95%CI 0.76 to 0.81), 0.77 (95%CI 0.75 to 0.79), 3.92 (95%CI 2.83 to 5.41), 0.28 (95%CI 0.23 to 0.35), 14.8 (95%CI 9.31 to 23.52) and 0.86, respectively. Subgroup analysis revealed that the source of patients, age, prediction methods and analysis features were potential sources of heterogeneity. Heterogeneity in the subgroup of 1-3 years old was small, and the pooled sensitivity, specificity and area under the curve of SROC were 0.81 (95%CI 0.76 to 0.85), 0.82 (95%CI 0.78 to 0.86) and 0.87, respectively. The pooled sensitivity, specificity and area under the curve of SROC for non-site studies were 0.80 (95%CI 0.75 to 0.84), 0.80 (95%CI 0.76 to 0.85) and 0.86, respectively. ConclusionMRI has a high accuracy in the diagnosis of preschool children with autism spectrum disorder; however, the possibility of misdiagnosis and missed diagnosis should be considered. Due to the limited quantity and quality of the included studies, more high-quality studies are needed to verify the above conclusion.