BEAN: Brain Extraction and Alignment Network for 3D Fetal Neurosonography.

Area of Science:

• Neuroimaging
• Medical AI
• Fetal Development

Background:

• Automated brain extraction and alignment are crucial for neuroimage analysis.
• Lack of automated solutions for 3D ultrasound (US) has hindered its use in fetal neuroimaging.
• Routine 3D US scans offer potential for studying fetal brain development.

Purpose of the Study:

• To develop an automated method for fetal brain extraction and alignment from 3D US scans.
• To improve the utility of 3D US as a neuroimaging modality for fetal development research.
• To introduce the Brain Extraction and Alignment Network (BEAN) for these tasks.

Main Methods:

• A multi-task convolutional neural network (CNN) named BEAN was designed.
• BEAN features two branches: one for brain extraction and another for similarity alignment.
• The network was trained and tested on 356 fetal head 3D scans (14-30 weeks gestation).

Main Results:

• BEAN significantly outperformed existing methods for fetal brain extraction and alignment.
• Achieved state-of-the-art performance with a mean Dice Similarity Coefficient (DSC) of 0.94 for extraction and 0.93 for alignment.
• Demonstrated consistent alignment of key brain structures and improved visibility in averaged scans.

Conclusions:

• BEAN provides an accurate and consistent automated solution for fetal brain extraction and alignment from 3D US.
• This advancement facilitates the use of routinely acquired 3D US for studying fetal brain development.
• The developed tool enables reliable analysis of neuroanatomical structures throughout gestation.