Robust Fiber Orientation Distribution Function Estimation Using Deep Constrained Spherical Deconvolution for Diffusion MRI

TL;DR

This paper tackles the reproducibility challenges of fiber orientation distribution function (fODF) estimation in diffusion MRI across multi-site and longitudinal studies. It introduces a 3D patch-based deep constrained spherical deconvolution (D-CSD) framework that maps spherical-harmonic representations of DW-MRI signals to 45 SH coefficients (even orders up to $L=8$) while explicitly enforcing scan–rescan invariance through a dedicated loss and intra-subject data augmentation. The approach outperforms traditional model-based CSD and voxel-wise DL baselines in intra-subject consistency and downstream connectome biomarkers, and generalizes to unseen data (e.g., BLSA) with finetuning. This work advances diffusion MRI harmonization by directly regularizing microstructure estimates, improving reliability for tractography and biomarker discovery across sites and sessions.

Abstract

Diffusion-weighted magnetic resonance imaging (DW-MRI) is a critical imaging method for capturing and modeling tissue microarchitecture at a millimeter scale. A common practice to model the measured DW-MRI signal is via fiber orientation distribution function (fODF). This function is the essential first step for the downstream tractography and connectivity analyses. With recent advantages in data sharing, large-scale multi-site DW-MRI datasets are being made available for multi-site studies. However, measurement variabilities (e.g., inter- and intra-site variability, hardware performance, and sequence design) are inevitable during the acquisition of DW-MRI. Most existing model-based methods (e.g., constrained spherical deconvolution (CSD)) and learning based methods (e.g., deep learning (DL)) do not explicitly consider such variabilities in fODF modeling, which consequently leads to inferior performance on multi-site and/or longitudinal diffusion studies. In this paper, we propose a novel data-driven deep constrained spherical deconvolution method to explicitly constrain the scan-rescan variabilities for a more reproducible and robust estimation of brain microstructure from repeated DW-MRI scans. Specifically, the proposed method introduces a new 3D volumetric scanner-invariant regularization scheme during the fODF estimation. We study the Human Connectome Project (HCP) young adults test-retest group as well as the MASiVar dataset (with inter- and intra-site scan/rescan data). The Baltimore Longitudinal Study of Aging (BLSA) dataset is employed for external validation. From the experimental results, the proposed data-driven framework outperforms the existing benchmarks in repeated fODF estimation. The proposed method is assessing the downstream connectivity analysis and shows increased performance in distinguishing subjects with different biomarkers.

Figures (7)

• Figure 1: Overview of the proposed method. The constrained spherical deconvolution (CSD) is affected by measurement factors of DW-MRI signals (e.g., hardware, reconstruction algorithms, acquisition parameters). The right panel shows the inter-site variability during modeling, even for the data that are collected from the same patient.
• Figure 2: Deep learning architecture. A 3D patch-wise convolutional neural network is proposed to fit the fiber orientation distribution function (fODF) from the spherical harmonics (SH) using $3\times3\times3$ cubic DW-MRI signals. A contrastive loss is introduced to reduce the intra-subject variability.
• Figure 3: Qualitative results of fODF modeling. Visualizations of fODF of the proposed deep learning (DL) method and the results from CSD modeling on two testing subjects in MASiVar. We took the same patch(matched with same color of the border) from the results in the crossing fiber area. The same voxel is matched with the same color of the circle.
• Figure 4: ACC histogram and ACC spatial map. This figure depicts the histogram of ACC between full diffusion directions' reconstruction and fewer directions' reconstruction while using the proposed deep learning (DL) method and the results from CSD modeling on the 2 testing subjects in MASiVar. The ACC spatial maps are the comparison between (1) the fODFs of reconstruction from CSD with full diffusion directions and (2) fewer diffusion directions' CSD and DL estimator on two testing subjects.
• Figure 5: Quantitative result of performances of methods/modeling with diffusion direction dropout. ACC is calculated at specific intervals - every 5 diffusion directions - such as at 45, 50, 55, and so on, enabling us to evaluate how the consistency of the model's/method's output was preserved despite the reduction in diffusion gradient directions. The dropout (drops from 96 to the subset of 45 directions) was performed randomly 10 times. The mean ACC and the std are calculated and shown in the line chart. The left panel shows that the deep learning-based method maintains high consistency than the CSD reconstruction when both are compared with the silver standard (full-direction CSD) during the diffusion direction dropout. The right panel presents a similar assessment, except comparing itself using the full-direction modeling.