Gemini Flash 3.6 Versus Human Observers for FIGO-Based Interpretation of Intrapartum Cardiotocography

Cardiotocography (CTG) is the standard method for monitoring fetal heart rate and uterine contractions during labor to detect signs of fetal hypoxia. However, human CTG interpretation is often subjective and subject to high interobserver variability, even among experienced clinicians. Multimodal artificial intelligence (AI) models have emerged as potential diagnostic aids capable of evaluating CTG tracings. This prospective diagnostic accuracy study aims to evaluate the diagnostic performance of Gemini Flash 3.6 in interpreting intrapartum CTG recordings according to the 2015 International Federation of Gynecology and Obstetrics (FIGO) guidelines, and to compare its accuracy directly against independent human expert obstetricians.

Participants include pregnant women admitted in active labor at term with a singleton cephalic pregnancy undergoing continuous CTG monitoring. De-identified digital images of the final 60 minutes of CTG tracing prior to delivery will be independently evaluated by Gemini Flash 3.6 and three expert obstetricians. Each evaluator will classify the tracings into one of three FIGO categories: Normal, Suspicious, or Pathological. All evaluators will be blinded to clinical details, delivery outcomes, and other observers' interpretations.

The primary objective is to assess the diagnostic sensitivity of Gemini Flash 3.6 in identifying neonates with a 5-minute Apgar score less than 7. Secondary objectives include comparing the diagnostic performance of the AI model with that of individual expert obstetricians, measuring interobserver agreement among all raters, and assessing associations with 1-minute Apgar scores and umbilical artery blood gas parameters. The AI model will be used solely as an investigational research tool, and its evaluations will not be available to the clinical team or used to guide patient care.