Automated fetal brain segmentation from 2D MRI slices for motion correction

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Fetal brain in-utero Magnetic Resonance Imaging (MRI) is crucial but challenged by maternal and fetal motion.
• Slice-to-volume reconstruction (SVR) methods are used, requiring accurate fetal brain isolation and motion correction.
• Current isolation and motion correction methods are often manual or semi-automatic, limiting efficiency.

Purpose of the Study:

• To develop an automated method for localizing and segmenting fetal brains from motion-affected MRI data.
• To integrate this automated segmentation with motion correction for improved Slice-to-Volume Reconstruction (SVR).
• To eliminate the need for manual delineation in fetal brain MRI preprocessing.

Main Methods:

• Utilized Maximally Stable Extremal Regions (MSER) for fetal brain localization.
• Employed a Bag-of-Words model with Scale-Invariant Feature Transform (SIFT) features for classification.
• Implemented a patch-based segmentation propagation with Conditional Random Fields (CRF), incorporating gestational age (GA) for prior knowledge.

Main Results:

• Achieved motion-corrected fetal brain volumes of clinically relevant quality in 85% of cases.
• Generated an automated segmentation of the reconstructed fetal brain with a mean Dice score of 93%.
• Demonstrated successful application across a gestational age range of 22 to 39 weeks.

Conclusions:

• The proposed automated method effectively addresses motion artifacts in fetal brain MRI.
• It significantly streamlines the preprocessing pipeline, enhancing diagnostic accuracy and efficiency.
• The high-quality segmentation output facilitates further downstream analysis in fetal neuroimaging.