Improving Xenon-129 lung ventilation image SNR with deep-learning based image reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Medicine

Background:

• Hyperpolarized 129Xe MRI is a valuable tool for assessing lung ventilation.
• Improving signal-to-noise ratio (SNR) is crucial for enhancing image quality and diagnostic accuracy.
• Deep learning (DL) offers potential for advanced image reconstruction techniques.

Purpose of the Study:

• To assess the feasibility and utility of DL-based reconstruction for improving SNR in 129Xe lung ventilation MRI.
• To evaluate the impact of DL reconstruction on quantitative ventilation metrics.
• To explore the potential for natural-abundance xenon MRI using DL.

Main Methods:

• Retrospective reconstruction of 129Xe lung MRI data from asthma/COPD patients using a DL pipeline at various denoising levels.
• Comparison of quantitative metrics (VDP, VH_I) between DL-reconstructed and conventionally denoised images.
• Prospective evaluation of DL reconstruction for natural-abundance xenon MRI in healthy volunteers.

Main Results:

• DL reconstruction significantly improved 129Xe ventilation image SNR compared to conventional methods.
• DL-reconstructed images showed minor positive bias in VDP and VH_I but preserved structural similarity better than alternative denoising.
• DL reconstruction achieved >3x SNR improvement, indicating feasibility of natural-abundance xenon MRI.

Conclusions:

• DL-based reconstruction substantially boosts 129Xe ventilation MRI SNR and maintains structural integrity.
• A minor bias in ventilation metrics due to image sharpness is observed but manageable.
• DL reconstruction facilitates cost-effective 129Xe ventilation imaging with natural-abundance xenon.