Deep learning reconstruction for 129Xe diffusion-weighted MRI enables use of natural abundant Xenon and improved image acceleration

Purpose (i) To assess whether 129Xe apparent diffusion coefficient (ADC) and diffusive length scale (LmD) metrics are quantitatively preserved with deep learning (DL) accelerated acquisition and reconstruction and (ii) to evaluate the feasibility of 129Xe diffusion weighted imaging with natural-abundance xenon at increased acceleration factors. Methods Twenty three-dimensional compressed sensing (CS) accelerated 129Xe DW MRI datasets were gathered from a cohort of patients with asthma, chronic obstructive pulmonary disease (COPD) and idiopathic pulmonary fibrosis (IPF). Images were retrospectively reconstructed with DL based CS, denoising and de-ringing reconstruction, and compared to conventional CS. ADC and diffusive length scales (LmD) were assessed and compared between conventional CS and DL reconstructions. Prospectively acquired DL reconstruction was then assessed in three healthy volunteers who underwent 129Xe DW MRI with both natural-abundance and enriched xenon mixes. Results DL reconstruction qualitatively improved the sharpness, SNR and image quality of 129Xe DW images. In the retrospective study, DL reconstruction produced a slight bias in ADC (5.4%) and LmD (0.8%) values when compared with conventional CS reconstruction. In the prospective study, DL reconstruction significantly improved the SNR of natural-abundance xenon images and produced ADC and LmD values comparable to those achieved with 129-enriched xenon. Conclusion DL-based CS, denoising and de-ringing significantly improves SNR and image sharpness in 3D 129Xe diffusion-weighted MRI while exhibiting a slight bias in ADC and LmD. This approach enables the use of natural-abundance xenon and higher acceleration factors, offering substantial cost reduction and improved clinical feasibility for hyperpolarized 129Xe lung morphometry.