[Diffusion tensor field estimation based on 3D U-Net and diffusion tensor imaging model constraint].

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Context:

• Diffusion Tensor Imaging (DTI) is crucial for mapping white matter tracts.
• Acquiring high-quality DTI data typically requires numerous diffusion-weighted (DW) images.
• Low signal-to-noise ratio (SNR) and limited DW images pose challenges for accurate DTI parameter estimation.

Purpose:

• To develop and validate a 3D U-Net based network (3D DTI-Unet) for precise DTI quantification.
• To enable accurate estimation of DTI parameters from a reduced set of DW images (1 non-DW, 6 DW).
• To address the challenge of low SNR in diffusion magnetic resonance imaging (dMRI) data.

Summary:

• The 3D DTI-Unet network processes noisy dMRI data, predicting a noise-reduced non-DW image and an accurate diffusion tensor field.
• The network incorporates a diffusion tensor imaging (DTI) model constraint, ensuring reconstructed dMRI data aligns with physical principles.
• Performance evaluation demonstrated superior quantitative and visual results compared to existing denoising algorithms (MP-PCA, GL-HOSVD).

Impact:

• Enables faster DTI acquisition by significantly reducing the number of required DW images.
• Improves the reliability of DTI-based quantitative diagnosis, particularly in challenging low-SNR scenarios.
• Offers a robust solution for accurate diffusion tensor field estimation in neuroimaging research and clinical applications.