(320) Machine Learning for the Prediction of Placenta Accreta Spectrum- Systematic Review and Diagnostic Meta Analysis

Antenatal detected placenta accreta spectrum disorder (PAS) is associated with improved outcomes compared with first recognized PAS at delivery. We conducted a systematic review and diagnostic meta-analysis assessing the ability of machine learning (ML) for the prediction of placenta accreta spectrum disorders (PAS) from MRI-based texture data.

Study Design:

Systematic review was conducted using PRISMA guidelines. MEDLINE, Scopus, the Cochrane Library, and ClinicalTrials.gov were searched from inception to June 30, 2023, for studies that described the use of ML for the prediction of PAS. We then conducted a diagnostic meta-analysis to assess the validity of ML algorithms for prediction PAS and reported their respective sensitivity and specificity. Finally, a pooled sensitivity and specificity and area under the curve (AUC) were computed as well as diagnostic odds ratio (OR) with 95% confidence intervals (CI) using a random effects model.

Results:

Among 15 studies initially retrieved, three were included with a total of 201 patients. We were able to identify five different ML algorithms used to predict PAS from MRI-based texture data points. The five algorithms and their respective sensitivities and specificities are reported in Figure 1. In the reported trials, the ML algorithm successfully predicted 88 patients and successfully excluded 87 patients. The ML algorithm falsely predicted 15 patients and falsely excluded 11 patients. The AUC of the pooled trial was (diagnostic OR: 47; 95% CI 20.42,107.87).

Conclusion:

ML-base MRI texture algorithms were predictive of PAS with high specificity. Our findings suggest that ML-based MRI texture algorithms have the potential to improve antenatal detection and ultimately pregnancy outcomes.