AID-DTI: Accelerating High-fidelity Diffusion Tensor Imaging with Detail-Preserving Model-based Deep Learning

TL;DR

High-fidelity diffusion tensor imaging (DTI) from sparsely sampled q-space is challenging due to noise. The paper introduces AID-DTI, a model-based deep learning framework that uses a singular-value–based regularizer to align the dominant spectra between prediction and ground truth, enabling accurate DTI metrics from $6$ diffusion-weighted volumes plus a $b=0$ image. AID-DTI employs an adaptive, deeply supervised loss combining $L_{Data}$ and $L_{Reg}$ via $\Sigma_{GT}$ and $\Sigma_{Pred}$ and demonstrates superior performance on Human Connectome Project data compared with $q$-DL, CNN, and MESC-SD in both quantitative metrics ($MSE$, $PSNR$, $SSIM$) and qualitative detail. The approach offers a practical route to fast, high-fidelity DTI in clinical and neuroscientific settings, with flexibility to extend to other diffusion models and multi-parametric MR imaging.

Abstract

Deep learning has shown great potential in accelerating diffusion tensor imaging (DTI). Nevertheless, existing methods tend to suffer from Rician noise and detail loss in reconstructing the DTI-derived parametric maps especially when sparsely sampled q-space data are used. This paper proposes a novel method, AID-DTI (Accelerating hIgh fiDelity Diffusion Tensor Imaging), to facilitate fast and accurate DTI with only six measurements. AID-DTI is equipped with a newly designed Singular Value Decomposition (SVD)-based regularizer, which can effectively capture fine details while suppressing noise during network training. Experimental results on Human Connectome Project (HCP) data consistently demonstrate that the proposed method estimates DTI parameter maps with fine-grained details and outperforms three state-of-the-art methods both quantitatively and qualitatively.

Figures (3)

• Figure 1: Diffusion physics-informed AID-DTI pipeline. The input is a single $b=0$ image and six DWI volumes sampled along uniform diffusion-encoding directions. The output is the high-quality volume of DTI metrics estimation.
• Figure 2: The upper row displays the plot of singular values along with the trends of PSNR and SSIM, showcasing how they vary with the number of selected singular values. The lower row exhibits the original FA image, the noisy FA image, as well as four reconstructed images and corresponding error maps obtained from the noisy FA using SVD, where the top 5, 20, 40, and 140 singular values are employed, respectively.
• Figure 3: The ground truth, estimated DTI parameters FA, AD, and MD, and corresponding residual maps based on q-DL, CNN, MESC-SD (baseline), and AID-DTI (ours) in a test subject with 6 diffusion directions at b-values of 1000$s/{mm}^2$.