Abstract
In Obstetrics and Gynecology (OB/GYN) discipline journals, mostly preliminary work (e.g., proof-of-concept algorithms or methods) in AI applied to this discipline is reported, and clinical validation remains an unmet prerequisite. Improvement driven by new AI research guidelines is expected. However, these guidelines currently cover only a part of the AI approaches (nonsymbolic) reported in this review; hence, updates need to be considered to include symbolic AI methodologies.
Introduction
While artificial intelligence (AI) is often touted as a revolutionary force in medicine, its actual footprint in specialized clinical literature can be difficult to discern. There is a significant "translation gap" between the development of novel algorithms in computer science and their adoption by clinical practitioners. This systematic review analyzes two decades of AI research within core OB/GYN journals, revealing a field dominated by preliminary "proof-of-concept" studies. While machine learning and knowledge-based methods are both gaining ground, the lack of external clinical validation remains a critical barrier to real-world deployment.
The authors sought to establish the actual contributions of AI as reported specifically within core OB/GYN journals. They aimed to identify which subdomains (e.g., fetal medicine, oncology) are leading the charge, which AI methodologies are most prevalent, and—most importantly—whether these tools are being validated rigorously enough for clinical use.
Methodology
The researchers conducted a systematic review following PRISMA guidelines, searching PubMed for articles published between January 2000 and April 2020. They filtered specifically for "core" OB/GYN journals to see what information is actually reaching the specialty's primary readership.
The study categorized AI into two main branches:
- Nonsymbolic AI (Machine Learning): Includes "black box" approaches like Artificial Neural Networks (ANNs) and Deep Learning.
- Symbolic AI (Knowledge-Based): Includes explainable, human-readable systems like ontologies and logic-based reasoning.
Key Results and Trends
The review analyzed the data sources and methodologies used across 66 included studies. A staggering 82% of AI research related to OB/GYN is published outside of core OB/GYN journals (e.g., in computer science or general imaging journals). This suggests that clinical journals are waiting for high-level validation before publishing AI research.
Methodological Split and Subdomains
The field is not just about Deep Learning. While Machine Learning (ML) accounted for 59% of the papers, symbolic AI (Knowledge Bases) represented a significant 38%. Obstetrics (41%) and Assisted Reproductive Medicine (33%) were the primary adopters. Within ML, ANNs were the most popular tool, used in 64% of ML-focused papers.
The Validation Bottleneck
A major finding was that 86% (57/66) of the studies performed validation on only a single dataset. No instances of external clinical validation were reported in the reviewed core journal articles. Most reported work remains at the "proof-of-concept" or "proof-of-feasibility" stage. The lack of external validation means these tools are not yet ready for the "prime time" of daily clinical practice.
Growth of the Field
The general trend of AI citations in PubMed has grown exponentially since 2000, reflecting a massive surge in interest across the broader medical community, even as the proportion in core OB/GYN journals remains relatively small.
Discussion and Future Directions
Current reporting guidelines like SPIRIT-AI focus heavily on nonsymbolic (ML) approaches. The authors argue that these guidelines need to be updated to include symbolic AI to ensure the entire field moves toward better reproducibility.
For AI to truly contribute to the OB/GYN field, researchers must move beyond preliminary algorithm reports and prioritize external clinical validation. Furthermore, editorial boards should consider including more AI experts to help bridge the gap between technical innovation and clinical application. The goal is to move from "proof-of-concept" to tools that can safely and effectively assist in clinical decision-making.
Summary updated