A Dual-Domain Denoising Method Based on Physical Intelligent Neural Network for Diffusion Tensor Imaging

Motivation: Diffusion Tensor Imaging (DTI) is limited by the long acquisition time and low signal-to-noise ratio. Noisy images hinder the observation of the anisotropy of water molecule diffusion, which further impacts clinical assessment. Goal(s): To design a deep learning-based method for DTI signal-to-noise ratio improvement. Approach: The proposed Dual-domain Tensor Denoising (DuTD) network leverages the structural, diffusion, and tensor information of DTI for denoising and fractional anisotropy (FA) generation. Results: Extensive results show that DuTD can improve the signal-to-noise ratio and remain more diffusion information efficiently compared with other advanced methods. Impact: The proposed DuTD network effectively enhances the SNR of images and describes diffusion information more accurately in both public datasets and private Parkinson's datasets.