Abstract
BACKGROUND: To assess the precision of artificial intelligence (AI) in aiding the diagnostic process of congenital heart disease (CHD). METHODS: PubMed, Embase, Cochrane, and Web of Science databases were searched for clinical studies published in English up to March 2024. Studies using AI-assisted ultrasound for diagnosing CHD were included. To evaluate the quality of the studies included in the analysis, the Quality Assessment Tool for Diagnostic Accuracy Studies-2 scale was employed. The overall accuracy of AI-assisted imaging in the diagnosis of CHD was determined using Stata15.0 software. Subgroup analyses were conducted based on region and model architecture. RESULTS: The analysis encompassed a total of 7 studies, yielding 19 datasets. The combined sensitivity was 0.93 (95% confidence interval (CI): 0.88-0.96), and the specificity was 0.93 (95% CI: 0.88-0.96). The positive likelihood ratio was calculated as 13.0 (95% CI: 7.7-21.9), and the negative likelihood ratio was 0.08 (95% CI: 0.04-0.13). The diagnostic odds ratio was 171 (95% CI: 62-472). The summary receiver operating characteristic (SROC) curve analysis revealed an area under the curve of 0.98 (95% CI: 0.96-0.99). Subgroup analysis found that the ResNet and DenNet architecture models had better diagnostic performance than other models. CONCLUSIONS: AI demonstrates considerable value in aiding the diagnostic process of CHD. However, further prospective studies are required to establish its utility in real-world clinical practice. THE PROSPERO REGISTRATION: CRD42024540525, https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=540525.