Detailed Delineation of the Fetal Brain in Diffusion MRI via Multi-Task Learning.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Developmental Neuroscience

Background:

• Diffusion-weighted MRI (dMRI) is crucial for studying fetal brain development in utero.
• Current computational methods for fetal dMRI analysis are limited, hindering research potential.
• Automated, accurate, and reproducible analysis tools are needed for fetal neuroimaging.

Purpose of the Study:

• To develop and validate a unified computational framework for fetal brain dMRI analysis.
• To enable automated segmentation of brain tissue, white matter tracts, and anatomical regions.
• To advance fetal neuroimaging by improving tractography and connectivity assessments.

Main Methods:

• Developed a multi-task deep learning method trained on 97 annotated fetal brains.
• Implemented automated segmentation of white matter, gray matter, and cerebrospinal fluid.
• Enabled segmentation of 31 white matter tracts and parcellation into 96 anatomical regions.

Main Results:

• Achieved high accuracy with mean Dice similarity coefficients: 0.865 (tissue), 0.825 (tracts), and 0.819 (parcellation).
• Validated the method on independent external data, demonstrating generalizability.
• The framework successfully performs tissue segmentation, tract segmentation, and regional parcellation.

Conclusions:

• The developed computational framework provides accurate and reproducible analysis of fetal brain dMRI.
• This method significantly enhances capabilities for fetal brain tractography and structural connectivity analysis.
• The framework is poised to advance the field of fetal neuroimaging research and clinical applications.