Detailed delineation of the fetal brain in diffusion MRI via multi-task learning

TL;DR

The paper addresses automated delineation of the fetal brain in diffusion MRI, a challenging task due to low SNR, fetal motion, and rapid developmental changes. It introduces a unified multi-task deep learning framework that jointly performs tissue segmentation (WM, CGM, SGM, CSF), 31 white matter tract segmentation, and cortex/deep gray nuclei parcellation (96 regions) using a DTI-based input and a cascade of task-specific FCNs with a patch-wise attention module and homoscedastic task uncertainty loss. The study annotates 97 fetal brains and demonstrates that the MTL model achieves mean Dice scores of 0.865 for tissue, 0.825 for tracts, and 0.819 for parcellation, outperforming single-task and baseline models, with code and a Docker release provided to enable adoption. Overall, the method enables fast, accurate, and reproducible fetal dMRI analysis, with potential to significantly improve fetal tractography, tract-specific analyses, and connectome-based assessments of prenatal brain development.

Abstract

Diffusion-weighted MRI is increasingly used to study the normal and abnormal development of fetal brain in-utero. Recent studies have shown that dMRI can offer invaluable insights into the neurodevelopmental processes in the fetal stage. However, because of the low data quality and rapid brain development, reliable analysis of fetal dMRI data requires dedicated computational methods that are currently unavailable. The lack of automated methods for fast, accurate, and reproducible data analysis has seriously limited our ability to tap the potential of fetal brain dMRI for medical and scientific applications. In this work, we developed and validated a unified computational framework to (1) segment the brain tissue into white matter, cortical/subcortical gray matter, and cerebrospinal fluid, (2) segment 31 distinct white matter tracts, and (3) parcellate the brain's cortex and delineate the deep gray nuclei and white matter structures into 96 anatomically meaningful regions. We utilized a set of manual, semi-automatic, and automatic approaches to annotate 97 fetal brains. Using these labels, we developed and validated a multi-task deep learning method to perform the three computations. Our evaluations show that the new method can accurately carry out all three tasks, achieving a mean Dice similarity coefficient of 0.865 on tissue segmentation, 0.825 on white matter tract segmentation, and 0.819 on parcellation. The proposed method can greatly advance the field of fetal neuroimaging as it can lead to substantial improvements in fetal brain tractography, tract-specific analysis, and structural connectivity assessment.

Figures (9)

• Figure 1: Schematic representation of the multi-task deep learning framework. Given the DTI map as input ($x$), the model computes tissue segmentation ($y_{\text{sg}}$), white matter tract segmentations ($y_{\text{tr}}$), and parcellation ($y_{\text{pc}}$).
• Figure 2: Example tissue segmentation maps predicted by the proposed method for three test fetuses at 25, 30, and 33 gestational weeks.
• Figure 3: Example commissural tract segmentation masks predicted by the proposed method for three test fetuses at 25, 30, and 33 weeks of gestation. Green shows the reference tract, and red shows the segmentation computed by the proposed method. In this example, for the fetus at 25 gestational weeks, the tractography-based method used to generate the "ground truth" failed.
• Figure 4: Example association tract segmentation masks predicted by the proposed method for three test fetuses at 25, 30, and 33 weeks of gestation. Green shows the reference tract, and red shows the segmentation computed by the proposed method.
• Figure 5: Example projection tract segmentation masks predicted by the proposed method for three test fetuses at 25, 30, and 33 weeks of gestation. Green shows the reference tract, and red shows the segmentation computed by the proposed method.