Automatic Fetal Middle Sagittal Plane Detection in Ultrasound Using Generative Adversarial Network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Obstetrics and Gynecology

Background:

• Accurate fetal growth and abnormality assessment in early pregnancy relies on the fetal middle sagittal plane (MSP).
• Ultrasound (US) image quality and operator expertise can impact the precision of manual MSP detection.
• Existing methods for fetal MSP identification face challenges due to image variability and operator dependence.

Purpose of the Study:

• To develop and evaluate an automated system for precise fetal middle sagittal plane (MSP) detection from three-dimensional (3D) ultrasound (US) volumes.
• To assess the performance of a generative adversarial network (GAN) framework for fetal MSP identification.
• To provide a reliable and efficient tool for early pregnancy fetal assessments.

Main Methods:

• A deep learning-based neural network was designed as a filter to generate masks for MSP extraction from 3D US data.
• The system utilized a seed point obtained via deep learning from 218 first-trimester fetal 3D US volumes.
• The middle sagittal plane (MSP) was automatically extracted using the developed image analysis system.

Main Results:

• The proposed automated system demonstrated feasible and excellent performance in fetal MSP detection compared to manual methods.
• No significant difference was observed between the semi-automatic and fully automatic system's MSP detection accuracy.
• The automatic system achieved inference times up to two times faster than the semi-automatic approach.

Conclusions:

• The developed system provides precise fetal MSP measurements, enhancing early pregnancy diagnostics.
• This automated approach for fetal MSP detection and measurement is expected to have significant clinical utility.
• The system's methodology holds potential for application in other clinical domains requiring precise anatomical plane identification.