Computer Vision for Identification of Increased Fetal Heart Variability in Cardiotocogram

Area of Science:

• Obstetrics and Gynecology
• Medical Artificial Intelligence
• Fetal Monitoring

Background:

• Increased fetal heart rate variability (IFHRV) is a critical early indicator of intrapartum fetal hypoxia.
• Accurate identification of IFHRV is crucial for timely intervention during childbirth.

Purpose of the Study:

• To evaluate the agreement between a machine learning (ML) algorithm (SALKA) and expert obstetricians in recognizing IFHRV patterns.
• To assess the efficacy of computer vision analysis in identifying early signs of fetal distress.

Main Methods:

• Cardiotocograms from 4,988 singleton term births were analyzed.
• The SALKA computer vision model was compared against the visual analysis of two expert obstetricians.
• Agreement was quantified using Cohen's kappa coefficient and other statistical measures.

Main Results:

• IFHRV patterns were identified in 11.7% of cardiotocograms by visual analysis.
• SALKA demonstrated an almost perfect agreement (Cohen's kappa = 0.981) with expert visual analysis.
• SALKA achieved high sensitivity (0.981) and a low false-negative rate (0.01), with excellent agreement in cases of neonatal acidemia (0.993).

Conclusions:

• The SALKA ML algorithm, utilizing computer vision, effectively identifies IFHRV with high accuracy.
• SALKA's performance closely matches expert obstetricians in detecting early fetal distress signs.
• This technology shows particular promise for identifying fetal distress in cases of neonatal acidemia.