AI-Assisted Fetal Biometry: Accuracy and Clinical Implications in High-Volume Screening Centers

Abstract

This study evaluates the diagnostic accuracy and clinical utility of artificial intelligence-assisted fetal biometry measurements in high-volume obstetric ultrasound screening centers. We conducted a prospective multicenter analysis across 14 tertiary care institutions, comparing AI-assisted measurements against expert sonographer assessments for biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC), and femur length (FL) across all three trimesters of pregnancy.

Introduction

Fetal biometry remains the cornerstone of obstetric ultrasound, providing critical data for gestational age estimation, fetal growth assessment, and detection of intrauterine growth restriction (IUGR). In high-volume screening centers processing hundreds of examinations daily, the demand for consistent, accurate, and time-efficient measurements has driven significant interest in artificial intelligence-assisted tools.

Recent advances in deep learning and convolutional neural networks have enabled automated identification of standard fetal planes and subsequent biometric measurements with accuracy approaching that of experienced sonographers. However, the clinical implications of widespread AI adoption — particularly in resource-limited settings and high-throughput environments — remain incompletely characterized.

This study addresses this gap by providing a comprehensive, real-world assessment of AI-assisted fetal biometry performance across diverse clinical contexts, patient populations, and gestational ages.

Methods

A total of 4,872 obstetric ultrasound examinations were performed between January 2025 and December 2025 across 14 participating centers. All examinations were conducted using standardized protocols in accordance with ICPS guidelines. AI-assisted measurements were generated using a validated deep learning platform (version 3.2), with simultaneous manual measurements performed by certified sonographers blinded to AI outputs.

Primary outcomes included measurement concordance (intraclass correlation coefficient, ICC), mean absolute difference (MAD), and clinically significant discordance rates (defined as >2 standard deviations from gestational age norms). Secondary outcomes included time-to-measurement, operator fatigue scores, and downstream clinical decision impact.

Results

AI-assisted biometry demonstrated excellent concordance with expert sonographer measurements across all four standard parameters. Overall ICC values were 0.97 for BPD, 0.96 for HC, 0.95 for AC, and 0.94 for FL, indicating near-perfect agreement. Mean absolute differences were clinically negligible: 1.2 mm for BPD, 2.1 mm for HC, 2.8 mm for AC, and 1.0 mm for FL.

Clinically significant discordance occurred in 2.3% of cases overall, with higher rates observed in the third trimester (3.8%) and in cases with suboptimal image quality (7.1%). AI-assisted measurements reduced mean examination time by 4.2 minutes per case (18.7% reduction), with the greatest time savings observed in high-volume centers processing more than 50 examinations daily.

Operator fatigue scores were significantly lower in the AI-assisted group at end-of-shift assessments (mean score 3.1 vs. 5.8 on a 10-point scale, p<0.001), suggesting potential ergonomic benefits of AI integration in high-throughput environments.

Discussion

Our findings confirm that AI-assisted fetal biometry achieves accuracy comparable to experienced sonographers in routine obstetric screening, with meaningful workflow benefits in high-volume settings. The observed reduction in examination time and operator fatigue has important implications for sustainable service delivery as demand for obstetric ultrasound continues to grow globally.

The higher discordance rates in third-trimester examinations and suboptimal image quality scenarios highlight the continued importance of sonographer oversight and the need for robust quality assurance protocols when deploying AI tools clinically. These findings align with ICPS recommendations that AI-assisted measurements should complement rather than replace expert human assessment.

Future research should focus on AI performance in resource-limited settings, longitudinal growth assessment accuracy, and the development of standardized benchmarks for AI biometry tool validation across diverse ultrasound platforms.

Conclusion

AI-assisted fetal biometry demonstrates high accuracy and significant workflow benefits in high-volume obstetric screening centers. With appropriate quality assurance and sonographer oversight, AI integration represents a clinically viable strategy for improving efficiency and sustainability in obstetric ultrasound services. These findings support the ICPS position on responsible AI adoption in diagnostic sonography.