FetDTIAlign: A Deep Learning Framework for Affine and Deformable Registration of Fetal Brain dMRI

TL;DR

FetDTIAlign addresses the challenging problem of registering fetal brain diffusion MRI across rapid developmental stages by combining a dual-encoder affine stage with modality-specific encoders for deformable registration, and by integrating tensor reorientation directly into the optimization. The framework leverages a multi-task loss on FA and diffusion tensors, tract segmentation guidance, and a multi-scale recurrent decoder with correlation-based blocks to produce plausible, high-accuracy alignments. Across gestational ages 23–36 weeks and an external dHCP cohort, FetDTIAlign outperforms classical methods (TBSS, DTI-TK) and recent DL baselines (SynthMorph, VoxelMorph) in tract-wise Dice, FA similarity, and tensor correspondence, while maintaining deformation plausibility (low NJD%). This method enables precise cross-subject and tract-specific analyses of early brain development and shows promise for studying deviations from typical trajectories with potential clinical impact.

Abstract

Diffusion MRI (dMRI) provides unique insights into fetal brain microstructure in utero. Longitudinal and cross-sectional fetal dMRI studies can reveal crucial neurodevelopmental changes but require precise spatial alignment across scans and subjects. This is challenging due to low data quality, rapid brain development, and limited anatomical landmarks. Existing registration methods, designed for high-quality adult data, struggle with these complexities. To address this, we introduce FetDTIAlign, a deep learning approach for fetal brain dMRI registration, enabling accurate affine and deformable alignment. FetDTIAlign features a dual-encoder architecture and iterative feature-based inference, reducing the impact of noise and low resolution. It optimizes network configurations and domain-specific features at each registration stage, enhancing both robustness and accuracy. We validated FetDTIAlign on data from 23 to 36 weeks gestation, covering 60 white matter tracts. It consistently outperformed two classical optimization-based methods and a deep learning pipeline, achieving superior anatomical correspondence. Further validation on external data from the Developing Human Connectome Project confirmed its generalizability across acquisition protocols. Our results demonstrate the feasibility of deep learning for fetal brain dMRI registration, providing a more accurate and reliable alternative to classical techniques. By enabling precise cross-subject and tract-specific analyses, FetDTIAlign supports new discoveries in early brain development.

Figures (21)

• Figure 1: Illustrated summary of the study.
• Figure 2: Affine registration approach in FetDTIAlign. The fetus shown in this figure has a gestational age of 25 weeks.
• Figure 3: Deformable registration approach in FetDTIAlign using an FA encoder (in cyan), a Tensor encoder (in purple), and a multi-scale recurrent registration decoder (in grey). The fetus shown in this figure has a gestational age of 25 weeks.
• Figure 4: The number of subjects in each Gestational Age (GA) group by Training and Test split for the BCH dataset
• Figure 5: Spatial normalized FA images of the GA23 group using FSL TBSS-(1) pipeline. The final results were automatically transformed to MNI152_1mm space, and visualized in the axial, coronal, and sagittal planes.