Keywords: Synthetic MR, Quantitative Imaging, Contrast-enhancement

Motivation: Gadolinium-based contrast agents (GBCAs) have become a cornerstone in clinical routine for lesions characterization and treatment monitoring. However, issues such as safety concerns related to deposition of GBCA in the body and brain, prolonged acquisitions, and cost increase advocate against its usage.

Goal(s): To replace the usage of GBCAs in post-contrast imaging with parametric maps and deep learning.

Approach: A cascade of two convolutional-neural-networks for pre- and post-contrast parametric mapping and the synthesis of post-contrast T1-weighted images from only two pre-contrast conventional weighted images.

Results: The proposed approach presents potential for predicting post-contrast T1w-enhancement without the usage of GBCAs.

Impact: The proposed deep learning approach provides both pre- and post-contrast parametric maps and, consequently, the capability of synthesizing any post-contrast image from only two pre-contrast conventional weighted images. Thus, it paves the way towards GBCAs-free acquisitions.

Keywords: Quantitative Imaging, Image Reconstruction

Motivation: Multi-contrast MR scans provide rich information for clinical diagnosis and research studies. However, long scan time is a limitation.

Goal(s): an implicit neural representation is proposed for accelerated multi-contrast parallel imaging reconstruction. The proposed scan-specific method obviates the need for fully sampled priors.

Approach: The spatial and temporal feature maps of an initial reconstruction are implicitly represented into the weights of a prior network. It exploits the physics-based parallel imaging forward model of sparsely sampled measurements.

Results: The proposed method outperforms the evaluated parallel imaging techniques at acceleration rates as high as R=16 in both reconstructed echo images and parameter mapping. 

Impact: The proposed scan-specific method reconstructs multi-contrast images by implicit representation of the feature maps learned from interim reconstructions and exploitation of parallel imaging forward model in the training stage. It outperforms evaluated parallel imaging techniques. 

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Denoising, Unsupervised Learning

Motivation: Traditional MRI scans, necessary for high SNR and clear images, are time-consuming and discomfort for patients. Shorter scans, meant to improve the patient experience, often compromise image quality and SNR. New deep learning techniques provide a solution to denoise MRI scans, even with limited data availability.

Goal(s): We aim to create an unsupervised MRI denoising method for real-world clinical settings, eliminating the need for clean or paired noisy images ensuring versatility and practicality.

Approach: We use an unsupervised diffusion-based denoising approach to denoise MRI scans.

Results: We achieve unsupervised denoising for MRI scans, outperforming previous methods and reducing time to 6 seconds.

Impact: Our approach denoises general MRI scans without extra clean or noisy data. It's suitable for real-world clinics, reducing patient MRI time. It enhances imaging quality, ensuring accurate diagnoses and faster clinical practices for patients and doctors.

Keywords: Motion Correction, Motion Correction, Neuroimaging, AI

Motivation: Data-driven retrospective motion correction methods currently face challenges with respect to robustness and long runtimes, which can be addressed by combining deep learning- and physics-based methods.

Goal(s): To validate a novel deep learning-assisted joint estimation algorithm on real motion-corrupted 3D MRI data.

Approach: A dataset of motion-corrupted data was acquired on 4 healthy volunteers. The performance of the proposed method was compared to a state-of-the-art deep learning method and a physics-based method.

Results: The proposed method outperformed the deep learning- and physics-based methods, yielding better image correction and converging faster.

Impact: The proposed retrospective motion correction method can be adopted into clinical practice as an alternative to rescanning, having demonstrated that it can salvage real motion-corrupted data without special hardware and requiring minimal sequence modifications.

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence

Motivation: Recent manifold-based models use unsupervised generative Variational smoothness regularization on manifold framework for improved recovery. This unsupervised method lacks a well-defined automated early stopping criterion and rely on subjective qualitative assessment only.

Goal(s): We aim to adapt a physics-guided early stopping criterion to V-SToRM framework leveraging non-cartesian multi-slice acquisition.

Approach: We developed a self-supervised variational manifold recovery method where we modified the original variational manifold scheme to integrate an early stopping criterion.

Results: With early-stopping criterion enforced, we observe a faithful reconstruction of spatiotemporal dynamics at epoch 45 and images without blocky/noise amplification artifacts at different temporal phases with suppressed temporal blurring artifacts.

Impact: While preserving the integrity of the ongoing joint learning of latent variables and generator weights, the adoption of early-stopping strategy in this context streamlines the computational complexity and consequently, rendering faithful and reproducible faster reconstructions.

Keywords: Machine Learning/Artificial Intelligence, Data Acquisition, MRI Acceleration, Fast Acquisition, New MRI Techniques

Motivation: Acquiring a high-quality 3D brain MRI in a short time is highly desired for infants/children studies.

Goal(s): To optimize and systemically evaluate an innovative technology, AI-assisted Compressed Sensing (ACS), used in Chinese Baby Connectome Project (CBCP) with faster 3D-T1w scans and preserved imaging fidelity.

Approach: We compared ACS with conventional techniques along the entire processing pipeline. CBCP-ACS was also compared with the existing infant cohort (BCP) with P2 acceleration regarding the derived development trajectories.

Results: Results suggested that CBCP data, with halved acquisition time, had comparable quality and derived neuroscience findings to BCP, indicating that a large cohort study with ACS is mature.

Impact: Equip high-resolution 3D-T1w MRI acquisition with ACS effectively shortens the acquisition time by 44%, providing a stable and robust solution for building large-scale infant/children brain imaging cohorts. The established technique could also facilitate clinical scans and patient studies.

Keywords: Machine Learning/Artificial Intelligence, Elastography, Stiffness Maps

Motivation: AI has proven itself in improving MRI reconstructions, yet its potential in estimating MRE stiffness maps with rapidly acquired data remains unexplored.

Goal(s):  Investigating the untapped potential of AI in MR Elastography, promising advancements in diagnostic accuracy and efficiency of this modality.

Approach: 3D FEM was used to create the dataset, and Deep Learning was used to reconstruct the stiffness maps from sparse wavefield data

Results: The Deep Learning model was able to effectively reconstruct the MRE-Stiffness maps at high acceleration rates. The model's performance was reported in terms of SSIM.

Impact: This innovative study leverages deep learning and finite element modeling to reconstruct liver stiffness maps from under-sampled MR Elastography data. The proposed AI approach demonstrates robustness and potential for accelerating stiffness estimation, paving the way for improved tissue stiffness estimation.

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, Respiratory-resolved 4D MRI, Deep Learning reconstruction, radiation therapy planning

Motivation: To contribute to the clinical evidence generation for 4D MRI in radiation therapy planning.

Goal(s): Emphasize the impact of a DL reconstruction with amplitude and phase binning on respiratory motion characterization.

Approach: 4D MRI data of 10 healthy volunteers and 8 patients were acquired using a free-breathing T1-weighted stack-of-stars sequence at 1.5T or 3T.

Results: Independent of the binning strategy, DL reconstruction consistently improves image quality and conspicuity of small anatomical details with the potential to shorten scan times. Differences of binning strategies become prominent for irregular breathers, where amplitude binning reveals larger motion ranges than phase binning.

Impact: To foster the ultimate goal of clinical adoption of 4D MRI for radiotherapy planning, we present an enhanced 4D MRI application supporting multiple binning strategies and an embedded DL reconstruction. 

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Reconstruction, cardiac cine imaging

Motivation: Cardiac cine MRI is the gold standard for cardiac functional assessment but requires acquiring several slices under multiple breath-holds, leading to limited number of cardiac phases, patient fatigue and misregistration between slices.

Goal(s): To develop a novel undersampled reconstruction based on Implicit Neural Representations (INR) to enable continuous cardiac cine MRI in a single heartbeat.

Approach: INRs allow implicitly regularized reconstruction of radial cardiac cine MRI without ECG gating. The proposed method is compared to a fully sampled acquisition and iterative SENSE in a healthy subject.

Results: The proposed approach shows comparable results to the fully sampled images but offering higher temporal resolution.

Impact: The proposed method allows implicitly regularized single heartbeat reconstruction of radial cardiac cine MRI without ECG gating, offerings potential improvements in cardiac cine acquisition efficiency and patient comfort.

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Motivation: Monte Carlo (MC) Dropout, a powerful uncertainty quantification (UQ) method for deep learning-based reconstruction, can impact reconstruction performance. Finding ways to enhance reliability assessment without compromising performance is essential.

Goal(s): This study aims to provide advisory information on how to incorporate MC Dropout into a model-based unrolled neural network, and to evaluate the reliability of UQ.

Approach: Different architectures with varying dropout rates are used to assess image quality. Images with visible structural aberrations and artificial perturbation are tested.

Results: Findings indicate that appropriate MC Dropout configurations improve reconstruction quality, and UQ maps effectively identify structural anomalies in images.

Impact: This research enhances the reliability of DL reconstructions by systematically investigating MC Dropout’s impact on reconstruction performance, particularly in scenarios lacking ground-truth references. The findings guide the incorporation of uncertainty quantification techniques, improving the overall quality of medical imaging applications.

Keywords: Machine Learning/Artificial Intelligence, Brain, ASL, 3D Cartesian TSE, CS-AI, Deep Learning

Motivation: Brain ASL images are frequently obtained at relatively low spatial resolutions, necessitating a desire for higher-resolution ASL MRI without extended scan times. 

Goal(s): To evaluate the potential of employing CS-AI for accelerating higher-resolution brain ASL MRI

Approach: Acceleration of higher-resolution 3D Cartesian TSE ASL MRI was achieved using CS-AI at 2, 3, and 4-fold rates, and its performance was compared with SENSE.

Results: CS-AI-accelerated 3D brain ASL images exhibited good SNR and quality, surpassing those acquired with SENSE, without affecting CBF quantification.

Impact: This investigation may improve the clinical utility of brain ASL, particularly in quantifying perfusion alterations in small-sized lesions.

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Neuro, Multiple Sclerosis, Reconstruction

Motivation: The infeasibility to collect large datasets of high-resolution, fully-sampled 7T data hinders the training of deep learning reconstructions for accelerated high-resolution 7T scans.

Goal(s): Demonstrate that combining pretraining on fully-sampled scans from 1.5/3T and self-supervised finetuning on a few undersampled 7T scans enables robust reconstruction.

Approach: High-resolution(0.5mm isotropic), 3D, 7T MP2RAGE scans of multiple sclerosis patients are used for finetuning/testing.

Results: Deep learning reconstructions from one 7T scan have higher apparent SNR than a median of GRAPPA reconstructions from three scans, currently used for assessment, with similar tissue contrast and lesion conspicuity, potentially reducing scan time by a factor of three.  

Impact: Very high-resolution 3D scans can require infeasible acquisition time to generate images suitable for clinical use. Our work shows the potential for deep learning reconstructions to reduce scan time by a factor of three, without fully-sampled 7T data.  

Keywords: Machine Learning/Artificial Intelligence, Quantitative Susceptibility mapping

Motivation: Current deep learning Quantitative Susceptibility Mapping (QSM) methods often rely on rigorous supervised training with paired data of the input and susceptibility maps and are only capable of specific one-to-one reconstructions.

Goal(s): In this study, we introduce the Diffusion Model QSM (DM-QSM), a controllable generative model capable of synthesizing high-quality susceptibility maps without the need for supervised training.

Approach: The DM-QSM method can produce controllable susceptibility maps with different measurements as the guidance.

Results: DM-QSM is versatile and suitable for many-to-one task including QSM super resolution and dipole inversion for both simulated and in-vivo tests.

Impact: This manuscript investigates the application of 3D generative models on QSM. It demonstrates robustness against acqusition artifacts for in-vivo test, and shows the potential beyond current tasks and is able to solve inverse problems like single-step QSM reconstruction.

Keywords: Machine Learning/Artificial Intelligence, Susceptibility, 3DEPI, SWI, QSM, Deep-Learning

Motivation: There is a strong clinical desire to accelerate the established SWI protocols. The clinical adoption of quantitative susceptibility maps (QSM) is hindered in part by long scan time and cumbersome offline QSM processing.

Goal(s): This study aims to substantially accelerate the SWI/QSM acquisitions while providing high-quality inline SWI/QSM images.

Approach: A flow-compensated, CAIPIRINHA-accelerated 3D echo-planar imaging (3DEPI) sequence was used to create 1-minute protocols with matching resolution to conventional sequences. Deep-learning reconstruction and super-resolution were used to enhance image quality.

Results: Compared with the established approach, the 1-minute 3D-EPI protocols provided 4× to 6× speed-up, while the DL reconstruction provided superior image quality.

Impact: Besides the general improvement in throughput, the 1-minute SWI protocols with improved image quality may enhance the role of MRI-SWI in acute care. The high-quality in-line QSM and susceptibility map weighted images (SWMI) will facilitate their clinical evaluation and adoption.

Keywords: Machine Learning/Artificial Intelligence, Vessels

Motivation: Altered angiogenesis is reported to be a key feature of vascular perturbation in AD brains.

Goal(s): There is missing global mapping tool to investigate the brain-wide vascular distribution pattern changes due to degenerative pathogenesis in AD brains. 

Approach: Here, we applied the pre-trained 3D U-net to super-resolve iron-particle (MION)-based CBV images from 75µm to 37.5 µm.

Results: By performing voxel-wise differential vascular density mapping analysis, we have revealed AD-specific vascular distribution changes from in vivo MION-based MRI images.

Impact: We developed an ultra-high resolution MION-based vascular density mapping method to verify the altered brain-wide vascular distribution pattern in AD mice. Using a pre-trained 3D U-net neural network, we can produce vascular maps with super-resolved 37.5µm isotropic resolution.

Keywords: Artifacts, Motion Correction

Motivation: Motion artifacts in mGRE MRI scans reduce image quality, increasing the risk of misdiagnosis and often necessitating repeat scans, negatively impacting patient care and diagnostic accuracy.

Goal(s): Our goal is to develop a novel deep learning-based framework for reducing motion artifacts in mGRE MRI k-space data, ensuring the generation of high-quality images.

Approach: The methodology proceeds by detecting and correcting motion-corrupted phase encoding lines within the k-space domain, employing a two-stage DeepFillv2 algorithm. It also integrates motion parameter estimation to enhance the framework's robustness.

Results: The model’s effectiveness in identifying and rectifying motion artifacts in MRI was confirmed through quantitative and qualitative evaluation.

Impact: The proposed k-space domain framework progresses by identifying phase encoding lines affected by motion and repairing them using deep learning techniques, thereby proving improved image quality and demonstrating potential as a diagnostic aid.