Artificial Intelligence-enhanced Prenatal Genomic Screening and Diagnosis: A Systematic Review of Cell-free DNA- and Sequencing-based Methods, Clinical Utility, Ethical Challenges, and Future Directions

Artificial intelligence (AI) is increasingly being explored as a means to extend the analytical and clinical scope of prenatal genomic screening, yet reported performance and implementation remain heterogeneous across applications. This systematic review synthesizes current evidence on AI-assisted methods in cell-free DNA (cfDNA) screening, noninvasive detection of monogenic and copy-number variants, and AI-supported interpretation of fetal exome and genome sequencing. In accordance with PRISMA 2020 guidelines, structured searches of major biomedical databases and registries were conducted through January 2025. Of 1,284 records identified, 38 studies met the inclusion criteria and were included in a qualitative synthesis. The reviewed literature encompassed machine learning models leveraging cfDNA fragmentomic features, deep-learning architectures for fetal variant inference, automated noninvasive prenatal testing (NIPT) pipelines, and AI-enabled prioritization of variants in fetal exome and genome sequencing. Across studies, AI-based approaches were reported to improve defined analytic tasks compared with conventional workflows. For trisomy risk prediction, several models demonstrated high discriminative performance, with reported area under the receiver operating characteristic curve values commonly ranging from approximately 0.90 to 0.97. In genome-wide copy-number variation and monogenic variant detection, selected AI frameworks reported sensitivities and specificities frequently exceeding 90% under controlled study conditions, alongside high overall accuracy. AI-assisted fetal-fraction estimation models showed improved robustness in low fetal fraction samples, and multiple studies described recovery of clinically relevant information from borderline or noninformative NIPT results. Integration of AI-derived genomic outputs with prenatal ultrasound phenotyping was also associated with higher diagnostic yields in cohorts with fetal structural anomalies. Nevertheless, the evidence base is constrained by retrospective study designs, population bias, limited external validation, and inconsistent reporting of model development. Ethical and implementation challenges, including explainability, equity, and communication of probabilistic results, remain prominent. Overall, current evidence suggests that AI-enhanced prenatal genomics is progressing toward early clinical integration, while underscoring the need for standardized validation frameworks, diverse population datasets, and robust governance to support responsible use in perinatal care.