Keywords: Artifacts, Artifacts, Nyquist Ghost Correction

Motivation: 3D-EPI suffers from inherent sampling errors, similar to conventional 2D-EPI, which result in Nyquist ghosting and shading artifacts that arise from non-linear phase errors in the sampled data.

Goal(s): To demonstrate that the Dual-Polarity GRAPPA (DPG) reconstruction method can be extended and applied effectively to 3D-EPI data.

Approach: We modified the DPG kernel to extend to all three sampling directions, to better accomodate 3D-EPI data.

Results: Phantom and in-vivo data demonstrate that DPG provides reconstructed images with lower levels of ghost artifacts and reduced artifacts from non-linear phase errors.  This extension to DPG will enable high-fidelity reconstructions in future 3D-EPI applications.

Impact: Our extension of Dual-Polarity GRAPPA to a 3D reconstruction kernel can improve image quality for 3D-EPI applications, providing higher fidelity and fewer artifacts than current conventional methods for improved imaging performance in emerging high-resolution fMRI and ultra-high-field imaging applications. 

Keywords: Signal Modeling, Quantitative Imaging, Simulation, B1 mapping

Motivation: Generation of multiple MR quantitative contrasts from an efficient multi-pathway multi-echo sequence would be highly useful for non-invasive MRgFUS breast cancer therapy assessment.

Goal(s): Develop physics models that enable neural networks to accurately estimate tissue properties from multi-pathway multi-echo imaging.

Approach: A Bloch solver was implemented that directly models spectroscopic and position information. Simulated signal magnitudes for a multi-pathway multi-echo sequence were used to train neural networks to estimate flip angle, T1, T2, and T2*.

Results: RMS error of parameter estimates for noisy/noiseless evaluation data were 0.4/0.3° for flip angle, 40/9 ms for T1, 10/2 ms for T2, and 7/1.7 ms for T2*.

Impact: Multi-pathway multi-echo imaging with machine learning-based MR parameter estimation shows promise in rapidly collecting quantitative data for evaluation of breast cancer treatment. The implemented Bloch solver enables versatile simulation of biological tissues through direct modeling of spectroscopic and position information.

Keywords: Artifacts, Quantitative Susceptibility mapping

Motivation: We address the issue of phase singularities and artefacts in phase-critical imaging applications, such as QSM, especially in more complicated scenarios: 7T, under-sampled, or single-echo scans.

Goal(s): To develop an image reconstruction method, “MORSE-PI”, that provides high SNR, artefact-free and singularity-free phase for structural and functional brain imaging.

Approach: MORSE-PI extends our previous approach, MORSE-CODE, by adding a Virtual Reference Coil computation that is used to correct phase offsets in the MORSE-CODE sensitivity estimates.

Results: MORSE-PI reconstructs structural and functional brain images at 3T and 7T free from artefacts and phase singularities yielding high SNR QSM.

Impact: MORSE-PI flexibly provides high SNR, fold-over-free and singularity-free phase images for single-echo and multi-echo structural GRE and functional EPI scans with real-time reconstruction. MORSE-PI naturally lends itself to phase-based imaging techniques such as structural and functional QSM.

Keywords: Quantitative Imaging, Image Reconstruction, Pixel-wise fitting network; Model-based deep learning

Motivation: Pixel-wise fitting networks are robust to error and enhance quantitative MRI (qMRI) over classical fitting. Combining them with qMRI reconstruction can achieve high performance in accelerated qMRI. 

Goal(s): We propose qDC-CNN, combining a pixel-wise fitting network with an unrolled reconstruction network, improving qMRI reconstruction performance.

Approach: We simulated multi-slice multi-echo data using the Brainweb database, comparing five models with different reconstruction and parameter fitting networks.

Results: qDC-CNN provided the highest-quality image reconstructions among all tested models.

Impact: The exceptional reconstruction performance of qDC-CNN, which combines a pixel-wise fitting network with an unrolled reconstruction network, has broad applications in accelerating various quantitative MRI tasks, offering superior results and potential advancements in medical imaging and beyond.

Keywords: Image Reconstruction, Image Reconstruction, Non-linear Encoding Field, ART, Kaczmarz, ART, SART, O-space Imaging, OSI

Motivation: The extremely time-consuming iterative algebraic reconstruction techniques(ART), e.g., Kaczmarz, have been used for image reconstruction in MRI with nonlinear encoding fields, e.g., O-space imaging (OSI).

Goal(s): We evaluate the effectiveness and performance of the linear image reconstruction technique (LIRT) for MRI with nonlinear encoding fields with an application of image reconstruction for O-space imaging (OSI).

Approach: The rotation theorem and linearity of 2D Fourier Transform lay the theoretical basis for LIRT image reconstruction for OSI.

Results: We compared the computation time of LIRT with SART for OSI image reconstruction. The final reconstructed images show no visually perceived non-linear distortion for OSI.

Impact: LIRT-OSI further demonstrate how LIRT can be accommodated to a wide range of nonlinear MRI field systems for quick image reconstruction and assessment, which is the first time such a technique has been proposed to the best of our knowledge.

Keywords: Quantitative Imaging, Electromagnetic Tissue Properties, Biomarkers

Motivation: Conventional two sequences acquisition in conductivity tensor imaging (CTI) can cause geometric mismatch between two acquisition data which may influence subsequent reconstruction.

Goal(s): We aim to simultaneously reconstruct the high-frequency conductivity and to fit microstructure parameters with data from single sequence.

Approach: We propose a novel data acquisition strategy which reconstructs the high-frequency conductivity from the phase of non-diffusion weighted data of RESOLVE. 

Results: The reconstructed high-frequency conductivity and fitted microstructure parameters has matched geometric images, and the low-frequency conductivity can be successfully reconstructed.

Impact: We show the conventional two sequence acquisition is reduced to the single sequence acquisition which avoids the geometric mismatch and obtains higher precision of low-frequency conductivity.

Keywords: Image Reconstruction, High-Field MRI

Motivation: Long acquisition times facilitate the acquisition of high resolution multiparametric maps at 7T, however long scan times can lead to motion artefacts even with accelerated k-space sampling.

Goal(s): Our goal is to reduce acquisition time and reduce motion artefacts using variable-density sampling.

Approach: We use the variable-density-cartesian-trajectory (VD-CASPR) to retrospectively undersample the k-space of fully-sampled multi-echo GRE data in spiral-like interleaves for acceleration factors 6, 8 and 10. HDPROST regularisation framework is used to reduce motion artefacts taking advantage of the multiple echoes by patch-based denoising.

Results: HDPROST enables greater acceleration potential taking avantage of the information redundancy and incoherent aliasing across echoes.

Impact: This work demonstrates a reconstruction technique that may allow for faster high-resolution quantitative mapping which may be beneficial for a range of neurological applications, especially in the identification and characterization of small scale brain architecture and its alteration in pathology.

Keywords: Image Reconstruction, Image Reconstruction

Motivation: Deep Learning MRI reconstruction requires access to high-performance compute hardware (GPUs) but reconstruction times can still remain slow. 

Goal(s): Use domain specific hardware accelerators to speed up deep learning MRI reconstruction compared to classic CPU/GPU implementations.

Approach: We use an open source SoC design and simulation framework, Chipyard, and a neural network accelerator, Gemmini, to run inference on a pretrained MRI reconstruction network. The results are measured on FireSim, an FPGA simulation framework. 

Results: Simulated inference times of pretrained models on Gemmini are much faster compared to CPU and show similar image quality metrics compared to the inference run on a GPU. 

Impact: Shows the potential of developing custom compute hardware designed to accelerate deep learning MRI reconstruction.

Keywords: Image Reconstruction, Brain

Motivation: While deep networks have shown great effectiveness for post-acquisition MRI resolution enhancement, their training requires an enormous of datasets. Deep Image Prior (DIP) is a novel approach that leverages the inductive bias of deep convolutional architecture, allowing for MRI super-resolution without the need for training. 

Goal(s): We aim to improve the capabilities of DIP and thus achieve resolution enhancement.

Approach: We introduced a hybrid regularizer that integrates total variation with a neural network denoiser into the DIP framework. 

Results: Validated on 5T MR datasets, our method further improved on DIP and generated high-resolution MRI with realistic details, outshining several competing methods.

Impact: The proposed unsupervised method offered a robust framework for MRI super-resolution reconstruction that leverages intrinsic image structure to ensure resolution enhancement without the need for training data, thus boosting the efficiency of medical imaging and potentially benefiting clinical diagnostics.

Keywords: Artifacts, fMRI (resting state), Distortion correction; Imaging translation

Motivation: State-of-the-art correction of field inhomogeneity induced image distortion in EPI based fMRI images usually depends on additional scans, which might be compromised by subject motion.

Goal(s): To develop correction distortion method for BOLD image without field map or reversed phase encoding direction acquisition. 

Approach: We introduce a novel approach leveraging using a 3D self-attention conditional generative adversarial network (SC-GAN) and iterative registration to correct fMRI distortion with distorted fMRI and structural MRI. 

Results: Visual and quantitative results indicate that our method effectively correct fMRI distortion without any additional scan. The correction improves after iteration.

Impact: Our approach leverage image synthesis for EPI based fMRI distortion correction, which bypass the need for additional scans, and prevent potential inaccuracy due to motion. This approach may help improve fMRI preprocessing.

Keywords: Image Reconstruction, Data Processing

Motivation: Quantification of the effect of sub-sampled k-space data in magnetic resonance image reconstruction by providing joint image reconstruction and uncertainty quantification.

Goal(s): Image reconstruction and uncertainty quantification from sub-sampled k-space measurements.

Approach: The problem is formulated within a Bayesian framework as an inverse problem, and prior distributions are assigned to the unknown model parameters. A Markov chain Monte Carlo (MCMC) method, based on a split-and-augmented Gibbs sampler, is then used to sample the resulting posterior distribution.

Results: The model is demonstrated using a real brain image from the human connectome project (HCP) to reconstruct images and provide uncertainty measures from sub-sampled k-space data.

Impact: We introduced an image reconstruction and uncertainty quantification algorithm from under-sampled k-space data. The results showed that the algorithm can quantify the effect of reduced samples, enabling fast imaging. Future work can investigate this approach using low-field MRI.

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Patient auto-positioning in low-field MRI

Motivation: Automated patient positioning and Specific Absorption Rate (SAR) estimation in MRI is crucial for optimized image quality. Achieving these objectives necessitates precise patient parameter estimation. Typically, manual estimation of patient parameters, such as height and weight, is error-prone and time-intensive.

Goal(s): To assess the 3D camera's potential for acquiring depth images suitable for deep learning (DL)-based estimation of patient height and weight.

Approach: We employed 3D camera technology to capture depth images of patients on MRI tables, enabling DL-based height and weight estimation.

Results: Our evaluation study demonstrated the 3D camera's effectiveness in acquiring depth images for accurate patient height and weight estimation.

Impact: Current deep learning-driven 3D camera methods enhance MR imaging workflows with the goal of achieving standardized and higher-quality image acquisition by accurately predicting patient height and weight.

Keywords: Image Reconstruction, Image Reconstruction

Motivation: Dictionary matching, used extensively in quantitative MRI, is resistant to standard approaches to spatial regularisation due to its discrete nature.

Goal(s): Demonstrate the feasibilty of efficient spatial regularisation of parameter maps produced via dictionary matching, in this work using total variation (TV) regularisation.

Approach: Spatially regularised dictionary matching was formulated as an optimisation on a discrete Markov random field, and the result optimisation problem solved using a primal-dual strategy, with the efficient iterative solver FastPD.

Results: TV regularisation improved apparent quality of parameter maps in phantoms and in vivo.

Impact: The proposed technique offers a means to improve the quality of parameter maps from any quantitative framework employing dictionary matching, covering a wide range of possible anatomies and clinical applications.

Keywords: Artifacts, High-Field MRI, Quality Assurance, 7T fMRI, phantom

Motivation: The quality of the high-resolution fMRI relies on the stability and high-performance of the ultra-high field scanners. A standardized quality assurance (QA) pipeline for 7T scanners is urgently needed.

Goal(s): We aimed to establish a comprehensive QA pipeline for 7T fMRI to monitor the stability and performance of scanners.

Approach: First, we designed an agar phantom for 7T fMRI. Second, we optimized the scanning parameters for high-resolution fMRI. Third, we developed an analysis program for QA report addressing Nyquist ghosting.

Results: The Nyquist ghosting rate reflected the phase error during acquisition. The QA metrics described the stability and performance of the scanner.

Impact: We firstly provide the QA pipeline for 7T high-resolution fMRI. The daily QA scanning routine serves as a valuable tool to monitor the stability and high-performance of the scanners, thereby contributing to the overall quality control of fMRI data.

Keywords: Data Acquisition, Data Acquisition, Noise, SNR, detectability, estimation

Motivation: Pixel SNR is a common metric for sequence design and scan implementation, however it is not ideal for detectability and estimation.

Goal(s): To take and modify an existing noise measure from the literature and test it against pixel SNR for both white and red (high spatial frequency) noise.

Approach: Ten individuals performed a detection task on 100 synthetic low SNR images with white or red noise.

Results: The new metric was more predictive of detection than pixel SNR or circle radius.  Images with red noise had worse pixel SNR but better detection rates.

Impact: A proposed SNR metric is more relevant than pixel SNR for both detection and estimation, and consistent with our findings of better detectability in red noise over white noise.  This is important for the design of clinical pulse sequences.

Keywords: Image Reconstruction, Sparse & Low-Rank Models

Motivation: The reconstruction quality of CS-MRI is significantly affected by the selection of shrinkage threshold.

Goal(s): Find a self-adaptive threshold for every iteration, every slice and every wavelet sub-band in compressed sensing reconstruction.

Approach: We propose an adaptive threshold selection method by combining an bayes-based adaptive wavelet shrinkage denoising method with compressed sensing reconstruction.

Results: Our threshold based on the coefficients in sparse transform domain has a better reconstruction performance compared with an optimal fixed threshold.

Impact: We propose an adaptive threshold selection method for compressed sensing reconstruction, which promote the reconstruction quality and avoid the manual selection of parameter.