Abstract
Artificial Intelligence (AI) is a dynamic area of computer science that is constantly expanding its practical benefits in various fields. The aim of this study was to analyze AI-guided radiological assessment of femoral neck fractures by performing a systematic review and multilevel meta-analysis of primary studies. The study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) on May 21, 2024 [CRD42024541055]. The updated Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines were strictly followed. A systematic literature search of PubMed, Web of Science, Ovid (Med), and Epistemonikos databases was conducted until May 31, 2024. Critical appraisal using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool showed that the overall quality of the included studies was moderate. In addition, publication bias was presented in funnel plots. A frequentist multilevel meta-analysis was performed using a random effects model with inverse variance and restricted maximum likelihood heterogeneity estimator with Hartung-Knapp adjustment. The accuracy between AI-based and human assessment of femoral neck fractures, sensitivity and specificity with 95% confidence intervals (CIs) were calculated. Study heterogeneity was assessed using the Higgins test I (2) (low heterogeneity <25%, moderate heterogeneity: 25%-75%, and high heterogeneity >75%). Finally, 11 studies with a total of 21,163 radiographs were included for meta-analysis. The results of the study quality assessment using the QUADAS-2 tool are presented in Table 2. The funnel plots indicated a moderate publication bias. The AI showed excellent accuracy in assessment of femoral neck fractures (Accuracy = 0.91, 95% CI 0.83 to 0.96; I (2) = 99%; p < 0.01). The AI showed good sensitivity in assessment of femoral neck fractures (Sensitivity = 0.87, 95% CI 0.77 to 0.93; I (2) = 98%; p < 0.01). The AI showed excellent specificity in assessment of femoral neck fractures (Specificity = 0.91, 95% CI 0.77 to 0.97; I (2) = 97%; p < 0.01). AI-guided radiological assessment of femoral neck fractures showed excellent accuracy and specificity as well as good sensitivity. The use of AI as a faster and more reliable assessment tool and as an aid in radiological routine seems justified.