Joint Denoising and Reconstruction of T2-Weighted PROPELLER MRI of the Lung at 0.55T Using Self-Supervised Deep Learning

Motivation: Lung MRI at 0.55T is highly susceptible to noise due to low field strength and low proton density, complicating accurate pathology identification. Goal(s): To improve 0.55T T2-weighted PROPELLER lung MRI quality through a self-supervised joint reconstruction and denoising model. Approach: Self-supervised training split each blade along the readout direction, with one part as input and the other for loss evaluation. MPPCA, used for comparison, was applied along the coil dimension, followed by GRAPPA and NUFFT. Lung MRI was validated against CT. Results: The self-supervised model enhances clarity and structural detail in 0.55T T2-weighted PROPELLER lung MRI, aligning well with CT. Impact: We developed a self-supervised learning-based joint reconstruction and denoising scheme for lung MRI at 0.55T. The proposed self-supervised model enhances image quality by reducing noise and improving structural clarity.