Artificial Intelligence (AI) in the diagnosis and prediction of adverse pregnancy outcomes for Placenta Accreta Spectrum Disorders (PAS): a systematic review and meta-analysis of diagnostic accuracy.

BACKGROUND

Precise prenatal diagnosis of Placenta accreta spectrum disorders (PAS) is challenging, and the diagnostic performance of conventional imaging modalities remains suboptimal. Artificial intelligence (AI) technologies have emerged as promising tools in assisting image analysis and improving diagnostic accuracy of PAS. Therefore, this study aims to systematically evaluate the diagnostic performance of AI models in diagnosing PAS and predicting adverse pregnancy outcomes (APO) associated with PAS.

METHODS

A systematic search was conducted across multiple databases, including PubMed, Embase, and Cochrane Library, to identify studies assessing the diagnostic performance of AI-based models in PAS or their ability to predict APO. Diagnostic metrics such as sensitivity, specificity, area under the curve (AUC), positive likelihood ratio, negative likelihood ratio, and summary receiver operating characteristic (SROC) curves were pooled to evaluate diagnostic accuracy. Heterogeneity was assessed using Cochran Q and I2 statistics, and meta-regression and subgroup analysis were conducted to examine potential sources of heterogeneity. The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool was utilized to assess the study quality.

RESULTS

A total of 16 studies involving 4,457 participants were included. The pooled results showed that AI models exhibit high sensitivity (88%, 95% CI: 81%-93%) and specificity (88%, 95% CI: 76%-94%) for diagnosing PAS, with an excellent AUC of 0.94 (95% CI: 0.91-0.96). Moreover, AI models also indicated promising performance in predicting clinically significant APO such as massive hemorrhage and hysterectomy, yielding a pooled sensitivity of 80% (95% CI: 73%-85%), specificity of 86% (95% CI: 78%-92%), and AUC of 0.87 (95% CI: 0.84-0.90). Meta-regression and subgroup analysis identified study design as a primary source of heterogeneity.

CONCLUSIONS

AI algorithms exhibited favorable performance for diagnosing PAS and predicting APO associated with PAS, suggesting the clinical translation potential of AI in enhancing the efficiency of diagnostic workflows and potentially reducing maternal morbidity and mortality.