Validation of an Artificial Intelligence-Powered Smart Fetal Heart Rate Measurement Tool

First trimester complications leading to emergency department visits are common, affecting nearly one in three pregnant patients. Measurement of fetal heart rate (FHR) is important for risk stratifying these patients for complications, and supported by American College of Obstetrics and Gynecology (ACOG) guidelines. Among novice providers, obtaining this measurement can present technical challenges and be time consuming. In this prospective observational method comparison study, we validated the use of an artificial intelligence (AI) powered tool for measuring FHR, as well as traditional M-mode measurement at the bedside, compared to expert measurement.

After obtaining written informed consent, a crown rump length (CRL) and M-mode tracing was measured by an expert sonographer. Subsequently, a novice provider measured the FHR by using both a traditional M-mode method and a novel AI tool. The gold standard FHR was determined in a blinded fashion using the expert M-mode tracing. Measurements by the novice were compared to the expert measurement by Bland-Altman analysis. Further, deviation from the expert “gold standard” was assessed via regression, evaluating for effects of BMI and gestational age via CRL on novice accuracy for both methods.

An interim analysis was performed at 50 enrolled patients, demonstrating similar accuracy (by mean difference: MD) and a trend for greater precision (by 95% limit of agreement: LOA) of the AI tool (MD: -0.70 bpm, 95CI: +/- 4.1 bpm, LOA: -28 to +27 bpm) compared to traditional M-mode (MD: -0.2, 95CI: +/- 5.4 bpm, LOA -37 to +37 bpm). The AI tool failed to capture a heart rate in 6% of cases (95CI: 1.3% to 17%). Heteroscedasticity was present, whereby increasing CRL was associated with increased accuracy of both the AI tool (R^2: 0.16, p<0.05) and M-mode (R^2: 0.20, p<0.01). There was no effect of BMI.

This interim analysis demonstrated similar test characteristics between AI and M-mode measurement of FHR, with a trend for greater precision with the AI tool. Although both test methods suffered from similar degrees of inaccuracy for fetuses with small CRL, a trend for superior precision of the AI tool makes this a promising method of FHR measurement.