Ultrasound Plane Pose Regression: Assessing Generalized Pose Coordinates in the Fetal Brain

Area of Science:

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

Background:

• Acquiring 3D fetal models from 2D ultrasound images is challenging for trainees.
• Existing methods often require additional sensors, complicating the process.

Purpose of the Study:

• To develop an ultrasound plane localization system for 3D fetal visualization, training, and guidance.
• To improve the accuracy of 3D fetal brain reconstruction from 2D ultrasound data without extra sensors.

Main Methods:

• Utilized a convolutional neural network (CNN) regression model to predict the 6D pose of ultrasound planes.
• Analyzed the accuracy of a normalized fetal brain reference frame for transventricular (TV) standard plane (SP) acquisition.
• Investigated the impact of data registration quality on model performance.
• Implemented data augmentation and larger training datasets.

Main Results:

• Achieved median errors of 2.97 mm for translation and 6.63° for rotation in ultrasound plane localization.
• Demonstrated improved accuracy compared to previous work through enhanced training strategies.
• Quantified the accuracy of the normalized fetal brain reference frame.

Conclusions:

• The developed ultrasound plane localization system enhances 3D fetal visualization and training capabilities.
• The system offers a sensor-free approach to improving obstetric ultrasound skill acquisition.
• Data augmentation and larger datasets significantly improve the accuracy of fetal brain ultrasound plane prediction.