In-utero Three Dimension High Resolution Fetal Brain Diffusion Tensor Imaging

Shuzhou Jiang, Hui Xue, Serena J. Counsell, Mustafa Anjari, Joanna M. Allsop, Mary A. Rutherford, Daniel Rueckert, Joseph V. HajnalMICCAI 2007 · DOI

Abstract

Mapping the microstructural development of the fetal brain in-utero is a critical yet challenging task due to unpredictable fetal motion. This paper presents a novel pipeline for high-resolution, 3D Diffusion Tensor Imaging (DTI) of the fetal brain. By integrating snapshot imaging with a hierarchical slice-to-volume registration and scattered data reconstruction, the method successfully overcomes motion artifacts to produce reliable Apparent Diffusion Coefficient (ADC) and Fractional Anisotropy (FA) maps. Validation in both adult and fetal cohorts demonstrates the pipeline's accuracy and its potential for clinical diagnosis of neurodevelopmental pathologies.

Introduction

Diffusion Tensor Imaging (DTI) is a powerful MRI technique that measures the diffusion of water molecules to characterize tissue microstructure and map white matter tracts. In the context of fetal development, DTI offers a window into brain maturity and connectivity. However, DTI is exceptionally sensitive to motion, requiring multiple diffusion-sensitized images (at least six directions plus a baseline $b_0$ image) for every slice. In an in-utero environment, fetal movement and maternal respiration frequently corrupt these measurements, making traditional 3D reconstruction nearly impossible without significant data loss.

Methodology

The proposed pipeline adapts the Snapshot imaging with Volume Reconstruction (SVR) technique to the specific requirements of DTI. The process is divided into three fundamental stages designed to transform a collection of motion-corrupted 2D slices into a coherent 3D volume.

Data Acquisition and Redundancy

To mitigate the impact of motion, researchers utilized a "looping" strategy. The fetal brain was repeatedly scanned using 15 different diffusion directions. This redundant sampling ensures that even if specific slices are discarded due to extreme motion, sufficient data remains to reconstruct the full 3D tensor field.

Hierarchical Slice-to-Volume Registration

The core of the motion correction process is a hierarchical registration algorithm. First, a self-consistent 3D baseline ($b_0$) volume is reconstructed. Subsequently, every diffusion-weighted slice is registered to this volume. This is performed in stages: registering large blocks of slices first, then smaller sub-packages, and finally individual slices. A critical component of this stage is Directional Correction: as the fetus rotates, the orientation of the diffusion gradient relative to the brain changes. The pipeline mathematically rotates the diffusion direction vector for every voxel to maintain the physical accuracy of the DTI calculations.

Scattered Data Reconstruction

Once aligned, the data points are no longer situated on a uniform grid. The researchers addressed this as a scattered data interpolation problem. Using Sparse Equations and Least Squares Regression (LSQR), the diffusion tensor is estimated at every point on a regular 3D grid. A positive-definite constraint is applied to ensure the resulting tensors are physically plausible.

Results and Validation

The pipeline was validated using an adult "gold standard" model and a fetal cohort.

Adult Validation

To establish a baseline for accuracy, an adult subject was imaged while stationary and again while performing deliberate head rotations (up to 30 degrees). The reconstruction system achieved a standard error of only 5.6% for ADC and 6.7% for FA compared to the stationary gold standard, with registration accuracy within 1/4 of a voxel.

Validation of the reconstruction pipeline showing the transition from motion-corrupted data to a high-quality reconstructed Fractional Anisotropy (FA) volume.

Fetal Cohort Findings

The study successfully produced 3D high-resolution (2mm isotropic) ADC and FA maps for a cohort of eight fetuses (24–34 weeks gestational age). The quantitative measurements for key brain regions, such as the thalamus and internal capsule, were consistent with ex-utero data from premature infants, confirming the method's reliability for in-utero assessment.

Clinical Application: Pathology Visualization

The pipeline's ability to visualize structural abnormalities was demonstrated in subjects with brain pathologies. In a case of agenesis of the corpus callosum, the 3D DTI maps clearly identified the absence of normal white matter connectivity, providing a level of detail not possible with standard 2D snapshots.

In-utero 3D DTI reconstruction of a fetal brain with pathology, illustrating the clarity of the reconstructed ADC and FA maps compared to motion-corrupted raw data.

Conclusion and Future Work

The development of this 3D DTI pipeline represents a significant advancement in fetal neuroimaging. While FA maps remain more sensitive to noise than ADC maps, the ability to reconstruct a consistent 3D tensor field despite significant motion opens the door to advanced applications like fetal tractography. Future work will focus on "tracing" the wiring of the developing brain, potentially allowing for the early detection of connectivity-based neurodevelopmental disorders.

Summary updated

BibTeX
@inproceedings{jiang2007utero,
  author = {Shuzhou Jiang and Hui Xue and Serena J. Counsell and Mustafa Anjari and Joanna M. Allsop and Mary A. Rutherford and Daniel Rueckert and Joseph V. Hajnal},
  title = {{In-utero Three Dimension High Resolution Fetal Brain Diffusion Tensor Imaging}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2007},
  doi = {10.1007/978-3-540-75757-3\_3},
}
Figure