Evaluation of automated tool for two-dimensional fetal biometry

Area of Science:

• Medical imaging
• Obstetrics
• Artificial intelligence in healthcare

Background:

• Fetal biometry is crucial for assessing gestational age and fetal growth.
• Manual measurement of fetal biometry from ultrasound images can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To assess the accuracy and reliability of an automated tool for measuring fetal head circumference (HC), abdominal circumference (AC), and femur length (FL) on frozen 2D ultrasound images.
• To evaluate the performance of an algorithm in automatically identifying structures and placing calipers for biometric measurements.

Main Methods:

• 100 singleton pregnancies (20-40 weeks gestation) underwent ultrasound examinations.
• An automated algorithm was used to measure HC, AC, and FL on three images per variable.
• Caliper placement was subjectively assessed and measurements were compared to real-time manual measurements.

Main Results:

• The automated tool correctly identified structures in 99% of images, with no failures for HC and FL.
• Subjective assessment found clinically acceptable caliper placement in 53.3% of images, with minor or major adjustments needed in 35.6% and 10.1%, respectively.
• Mean percentage errors for HC, AC, and FL were -0.81%, 2.40%, and 3.76%, respectively, with good agreement to manual measurements.

Conclusions:

• The automated tool demonstrates high accuracy in measuring fetal biometry on frozen ultrasound images.
• While the tool shows promise, manual correction of caliper placement is still required for clinical application.
• Further algorithmic improvements can enhance the automated tool's performance and reduce measurement bias.