Purpose or Case Report: Diffusion-weighted Magnetic Resonance Imaging (dMRI) is increasingly used to study the fetal brain in utero. dMRI enables streamlined tractography, a computation with unique applications such as white matter tract-specific analysis and structural connectivity assessment. However, due to low fetal dMRI data quality and the challenging nature of tractography, existing methods often produce highly inaccurate results. This study addresses these challenges by proposing an anatomically constrained tractography method that accurately segments fetal brain tissue directly within dMRI.
Methods & Materials: We develop a deep learning method to compute the segmentation within dMRI automatically. It involves acquiring fetal MRI scans and preprocessing with motion correction and resampling. Local fiber orientations are estimated using a diffusion tensor model, enhanced with a sharpening technique for improved accuracy. A deep learning model is developed for brain tissue segmentation in the dMRI, using diffusion tensor information modulated by fractional anisotropy. These segmentations are converted into five-tissue-type maps for anatomically constrained tractography. Streamline tracing is performed with the iFOD2 algorithm, using anatomical constraints to launch and terminate streamlines at appropriate boundaries. The method was validated against established tractography techniques and demonstrated improved accuracy, robustness, and success in reconstructing various tracts, as confirmed by expert radiologist ratings.
Results: In validation using 94 fetal brain MRIs, the method achieved an average Dice similarity coefficient (DSC) of 0.898 for white matter and 0.841 for cortical gray matter, outperforming Attention U-Net by 4.9% and 2.9%, respectively. The tractography method demonstrated a 99.9% success rate in reconstructing 11 major tracts, compared to 72.2% with iFOD2, 55.6% with FACT, and 52.2% with rk2. It consistently achieved high-quality ratings, averaging 4.27 out of 5, versus around 1.80 for the other methods, highlighting its potential to enhance fetal brain development and structural connectivity studies.
Conclusions: The study demonstrates that anatomically constrained tractography significantly improves the accuracy of fetal brain imaging. This method enhances dMRI-based assessments of brain connectivity, offering valuable insights for both scientific and clinical applications in early brain development.