Bridging Prenatal Diagnostics and AI: A Systematic Review and Meta-Analysis of the Efficacy of Advanced Algorithms in Identifying Congenital Fetal Abnormalities

Congenital fetal abnormalities have emerged as a major cause of infant mortality globally. About 3% of pregnancies are reported as fetal structural anomaly through ultrasound, ranging from minor defect to severe organ system anomalies. Our study aimed to evaluate effectiveness of Artificial intelligence (AI) algorithms in prediction of fetal heart and brain abnormalities by using meta-analysis approach. In this study, the “Reporting Items for Systematic Review and Meta-Analysis (PRISMA)" guidelines were applied for screening and selection of research articles. We searched the research articles according to research aims from Google scholar, PubMed, and Ovid MEDLINE. The data search was limited to January 2015 to May 2024. Two researchers independently screened the studies to detect research aim oriented studies. After screening of titles, those researchers assessed all abstract for eligibility. The risk bias of included studies was assessed by two researchers who used Cochrane library tool (version 5.4.0). Using the DerSimonian-Laird technique, a bivariate random effect model that generated SROC plots, we pooled test estimates. This method is particularly helpful in visualizing variability in the sensitivity and specificity across studies, accounting for the heterogeneity. This improves the reliability of the SROC plot, offering a clear summary of diagnostic performance Using the software RevMan 5.4, we carried out every statistical analysis, particularly heterogeneity through the Q test. About 243,456 screened fetus or pregnant women of 11-32 gestational weeks through 10 studies based on AI algorithms (machine learning, deep learning and AI diagnostic tool). The imaging protocols including ultrasound and MRI were used to take visuals of fetal brain and heart abnormalities in all of including studies. DL and ML algorithms provided high-performance diagnostic predictions, mostly include CNN models that assist in providing accurate diagnostic results with high sensitivity and specificity. The AI algorithm model showed a generally good accuracy scores for detection of brain and heart abnormalities and reference machine learning findings, with sensitivity ranging from 82 to 99% and specificity from 78 to 99%. Fig. 5's area under the SROC curve was 0.960. In contrast to the general detection, ML and DL seems to be more sensitive in diagnosing fetal brain and heart abnormalities. Our study provided the scientific evidence that AI models including deep learning and Machine learning are effective in generating highly reliable detection or classification results for diagnosis of fetal brain and heart anomalies. These strategies can help in reducing infant mortalities, improving treatment outcomes, and postpartum outcomes among women.