Artifact elimination for magnetic resonance imaging

Abstract

Magnetic resonance imaging (MRI) is a versatile imaging modality that provides non-invasive, non-ionizing and quantitative characterization of tissues for clinical diagnoses and preclinical studies. However, MR images inherently suffer from various artifacts (including noise) due to practical constraints, such as limited data acquisition time, hardware imperfections and reconstruction errors. Such artifacts not only compromise MR image quality, but may also hamper the diagnostic accuracy. The main scope of this dissertation is to develop advanced artifact elimination approaches for MRI in presence of limited k-space data sampling and hardware limitations. First, a novel noise reduction method was proposed for diffusion MRI. Diffusion MRI provides a powerful approach to map tissue microstructures. However, it intrinsically suffers from low signal-to-noise ratio (SNR), especially when spatial resolution or diffusion weighting value is high. The proposed method exploits non-local self-similarity as well as local anatomical similarity within multiple diffusion-weighted images (DWIs) acquired with the same anatomical geometry but different diffusion directions. Specifically, for each small 3D reference patch sliding within 2D DWI, non-local but similar patches are searched by matching image contents within entire DWI dataset, and then structured into a patch matrix. The resulting patch matrices are denoised by enforcing low-rankness via weighted nuclear norm minimization, and finally back-distributed to DWI space. The proposed method achieved significant noise reduction while preserving structural details, and consistently outperformed the existing Marchenko-Pastur principal component analysis (MPPCA) denoising method. Further, the denoised DWIs led to substantially improved DTI parametric maps, exhibiting significantly less noise and revealing more microstructural details. Second, a multi-slice acquisition and joint calibrationless reconstruction strategy was developed for aliasing artifact elimination in accelerated multi-slice 2D Cartesian MRI. Here, multi-slice multi-channel data are first acquired with random or uniform phase encoding (PE) undersampling while orthogonally alternating PE direction among adjacent slices. They are then jointly reconstructed through a low-rank multi-slice Hankel tensor completion (MS-HTC) approach. The proposed multi-slice acquisition and joint reconstruction strategy augments the overall incoherency and enables more effective utilization of the coil sensitivity and image content similarities among adjacent slices, leading to substantial suppression of aliasing artifacts. This new approach is applicable to random as well as uniform PE undersampling, thus can be easily implemented in practice for 2D multi-slice Cartesian parallel imaging without any coil sensitivity calibration. Third, a deep learning electromagnetic interference (EMI) cancellation technique was presented for MRI with no or incomplete radiofrequency (RF) shielding. Clinical MRI scanners all require bulky and enclosed RF shielding cage in practice to prevent external EMI signals during MRI scanning. They also require all scanner electronics within shielding cage to be of high quality (i.e., minimal EMI emission). The proposed approach utilizes separate EMI sensing coils to simultaneously acquire EMI signals and subsequently predicts and removes EMI signals from signals acquired by MRI receive coil. It effectively and robustly removed various EMI from both external and internal sources, producing final image SNRs highly comparable to those obtained using a fully enclosed RF shielding cage in 0.055T human brain imaging.