Offline Reconstruction of Diffusion MRI Acquisitions for Comparison Between Complex PCA-Based and AI-Based Denoising

Optimal diffusion MRI (dMRI) data for image denoising is often unavailable from scanner reconstruction. In this work, we make available an offline reconstruction pipeline for GE dMRI acquisitions, giving access to complex dMRI data. Furthermore, we compare the efficacy of GE HealthCare's AIR-Recon DL (ARDL), a proprietary convolutional neural network-based reconstruction and denoising approach, to patch-based MPPCA_SVS and NORDIC denoising methods on high-resolution dMRI data.

We developed an end-to-end offline dMRI reconstruction pipeline for GE HealthCare acquisitions, augmenting the Orchestra software development kit, and validated its output against scanner reconstruction. We used it to compare MPPCA_SVS, NORDIC, and ARDL denoising approaches, considering underlying metrics reflecting noise variance and bias, such as the signal profiles in highly anisotropic areas, and secondary downstream measurements, such as fiber orientation estimation and white matter tractography.

Our validated offline reconstruction supports various in-plane/out-of-plane accelerations and partial Fourier reconstruction methods. Unlike scanner reconstruction, it provides access to complex dMRI data, enabling denoising in the complex domain, which demonstrated superior noise floor suppression compared with magnitude-constrained denoising. PCA-based denoising methods had improved spatial resolution, contrast-to-noise and more robust fiber orientation estimation compared with ARDL.

We found significant gains in dMRI data quality when using the proposed offline reconstruction pipeline, allowing complex-domain denoising to obtain high-quality data at high spatial resolution and b-value, using a wide-bore scanner and a standard PGSE EPI sequence. MPPCA_SVS and NORDIC (4D PCA-based) outperformed ARDL (2D) in terms of spatial resolution and reduction of noise variance.