Keywords: Machine Learning/Artificial Intelligence, Cancer

Motivation: There is a lack of publicly available k-space data of breast dynamic contrast-enhanced (DCE) MRI that can be used for development of image reconstruction and machine learning methods for breast MRI.

Goal(s): We aim to make a publicly available radial k-space dataset of breast DCE-MRI which will promote development of fast and quantitative breast imaging methods.

Approach: Data of women undergoing routine diagnostic breast DCE-MRI exams have been acquired using a stack-of-stars radial imaging at 3T.

Results: Our fastMRI breast dataset includes radial k-space data and case-level labels for 275 cases (70 malignant, 158 benign and 47 no-lesion cases).

Impact: This work introduces the first large-scale dataset of radial k-space data for breast DCE-MRI acquired in diagnostic breast MRI exams. Having this dataset and accompanying reconstruction code publicly available, will support research and development of fast and quantitative breast DCE-MRI. 

Keywords: Quantitative Imaging, Relaxometry, QSM

Motivation: Current high-resolution T₂* and susceptibility mapping techniques remain time-consuming or suffer from geometric distortion.

Goal(s): Our goal was to achieve distortion-free and time-efficient quantification of whole-brain T₂* and susceptibility.

Approach: The multiple overlapping-echo detachment imaging (MOLED) method was extended to 3D acquisition for collecting more echoes for robust high-resolution parametric mapping. Single scan blip-up-down operation of two echo trains combined with deep learning reconstruction was used for distortion correction.

Results: 3D-MOLED enables high-quality T₂* and susceptibility mapping in 32 seconds, comparable to conventional 3D-GRE in 12 minutes, with Pearson’s correlation coefficient of 0.983 and 0.986, respectively.

Impact: Distortion-free whole-brain T₂* and susceptibility mapping at isotropic 1 mm³ resolution can now be achieved using our newly developed 3D-MOLED technique in only 32 seconds, which significantly improves the motion robustness of quantitative imaging in clinical examinations.

Keywords: Image Reconstruction, Machine Learning/Artificial Intelligence, Filter-exchange imaging, water exchange rate, parameter map reconstruction, denoise

Motivation: Water exchange measured by filter-exchange imaging (FEXI) is expected to serve as an important biomarker for several brain diseases. However, its estimation accuracy is easily affected by noise.

Goal(s): To develop an approach for reconstruction of FEXI parameters from noisy signals.

Approach: An end-to-end framework was constructed to achieve parameter reconstruction without corresponding labels. An adaptative deformable convolutional network was introduced to explore structural information. A loss function was designed to enhance network denoising performance.

Results: Simulation results under SNR=30~50 showed that the S2P achieved optimal results in the reconstruction of apparent water exchange rate, with PSNR of 27.44 and SSIM of 0.9050.

Impact: The S2P, an end-to-end framework, reconstructs high-quality FEXI parameter maps from only a single scan when it has been trained with noisy pairs, which can provide efficient and reliable medical images for clinical diagnosis.

Keywords: Machine Learning/Artificial Intelligence, Spectroscopy

Motivation: One of the major challenges for spectral fitting is the modeling of background signals.

Goal(s): Develop a deep learning model for quantitative detection of in vivo metabolites without relying on spectral fitting.

Approach: Spectral fingerprint representation is achieved by combining manifold learning and representation learning, with the tasks that include predicting metabolite concentrations, transverse relaxation times, and reconstructing individual metabolite signals.

Results: The t-SNE map illustrates that metabolites can be clustered based on the fingerprints generated by the model. The predicted metabolite concentrations and relaxation T₂s agree with those found in the literature. The spectral background or unregistered signals are effectively filtered out.

Impact: The deep learning model demonstrates high practical viability for the quantification of metabolite concentrations and relaxation T2s. It essentially searches for learned spectral fingerprints instead of relying on spectral fitting, the latter involves modeling all signals contained in the data.

Keywords: Quantitative Imaging, Image Reconstruction

Motivation: Joint MAPLE is an MR parameter mapping technique with improved results which suffers from long processing times.

Goal(s): We propose a fast version of Joint MAPLE as a self-supervised, model-based multi-parameter mapping technique capable of jointly mapping T₁, T₂^*, frequency and proton density in a whole brain volume ~50 times faster than the original version, while retaining its parameter mapping performance.

Approach: A fast whole brain reconstruction, transfer learning and a rapid initialization in optimization is incorporated.

Results: Results show that fast Joint MAPLE retains the mapping performance of the original version and outperforms existing methods.

Impact: Fast Joint MAPLE estimates T₁, T₂^*, frequency and proton density of a volume ~50 times faster than the original version with the same performance. A fast volume reconstruction, transfer learning and a rapid initialization is incorporated for faster mapping.

Keywords: Machine Learning/Artificial Intelligence, fMRI

Motivation: Multi-echo fMRI holds the promise of more potential applications, however it suffers from long readout lengths.

Goal(s): Explore the possibility of using deep learning(DL) reconstruction for highly under-sampled spiral multi-echo fMRI acquisition.

Approach: Multi-echo data from four subjects were collected for DL training. Multi-echo fMRI data from another subject was used for testing the DL model. The DL model has been designed and modified to enable the reconstruction of ten-fold under-sampled fMRI images for BOLD analysis.

Results: The DL model has not only reconstructed the multi-echo spiral fMRI with good image quality, but also preserved its BOLD sensitivity with the highly under-sampled data.

Impact: With the help of DL, multi-echo fMRI may become more versatile for clinical use and future studies.

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

Motivation: Echo Modulation Curve T2 Mapping (EMC-T2) mapping can generate highly accurate, precise, and reproducible T2 quantification. However, the standard EMC-T2 framework requires ~10 echoes and a cumbersome post-processing step for pixel-wise dictionary matching.

Goal(s): This work proposes a deep learning version of EMC-T2 mapping, called DeepEMC-T2, to enable efficient and accurate estimation of T2 maps from fewer echoes without requiring a dictionary.

Approach: DeepEMC-T2 was developed using a spatiotemporal convolutional neural network, which estimates both T2 and PD maps directly from multi-echo spin-echo images.

Results: DeepEMC-T2 enables efficient and accurate T2 mapping and requires only smaller number of echoes compared to standard EMC-T2.

Impact: Standard EMC-T2 enables accurate T2 quantification but previously required a complicated post-processing step that made clinical translation challenging. DeepEMC-T2 enables efficient and accurate T2 quantification with fewer echoes. This could facilitate more widespread translation of this technique into clinical practice.

Keywords: Image Reconstruction, Image Reconstruction, Multi-Contrast, Joint Optimization, Optimized Sampling

Motivation: The current multi-contrast MRI sampling and reconstruction methods cannot efficiently collect complementary information to achieve better reconstruction performance.

Goal(s): A method was designed to generate corresponding sampling masks for each contrast image in a multi-sequence clinical scanning scenario, collect the optimal complementary information for better application in multi-contrast joint reconstruction.

 

Approach: We jointly optimized the sampling and reconstruction of multi-contrast images, and designed learnable acceleration ratio and decoder feature fusion for images with different contrasts.

Results: The PSNR and SSIM metrics of reconstructed images have significantly improved, and different sampling masks can be generated for different contrasts and sampling order.

Impact: The method of jointly optimizing the sampling and reconstruction of multi-contrast images in a single scan may provide a powerful tool for accelerating and optimizing the MRI scanning process and improving the reconstructed quality of the multi-contrast images.

Keywords: Machine Learning/Artificial Intelligence, Data Analysis

Motivation: While noise propagation in linear imaging methods like SENSE is well-studied, similar analysis is not common in deep learning-based MRI reconstructions.

Goal(s): Evaluate, characterize, and compare noise propagation in deep learning-based and linear MRI reconstruction under varying conditions.

Approach: We uses Monte Carlo simulations to empirically analyze mean and variance of knee MRI images reconstructed by SENSE and deep learning methods. 

Results: SENSE yields unstructured, relatively-uniform noise distribution, while deep learning methods produce anatomically structured noise with substantial variability across tissues, acceleration factors, and noise levels. Noise-aware deep learning reconstruction shows more uniform noise propagation and reduced tissue-specific variability.

Impact: Elucidating noise propagation in deep MRI reconstructions could direct algorithm refinement, optimizing image quality and reliability for clinical application. Simply reconstructing noise-propagation maps in routine protocols may help in image interpretation.

Keywords: Image Reconstruction, Cardiovascular, Deep learning, Radial MRI

Motivation: Hankel-based reconstruction distorts the image's center less than its periphery. This prompts us to examine Hankel-based reconstruction for neural network training data preparation.

Goal(s): To train the model-agnostic neural network on Hankel-based reconstruction data to improve image center reconstruction.  

Approach: A neural network trained on Hankel-based reconstruction data was compared to an equivalent network trained on NUFFT-based reconstruction data.

Results: In the context of radial dynamic imaging, where the ROI can be placed in the center of the image, our approach achieved better results than when using NUFFT-based data preparation for reconstruction of undersampled radial data.

Impact: This study might influence dynamic radial-MRI reconstruction. Our data preparation for training and testing the network improved cardiac-MRI qualitative outcomes, especially in the heart region. The radial-MRI society may find the proposed solution appealing when paired with DL-based approaches.

Keywords: Image Reconstruction, Image Reconstruction

Motivation: Deep-learning-based image reconstruction methods for accelerated magnetic resonance imaging are optimized for global image quality metrics, but lack focus on diagnostically relevant features during training and reconstruction.

Goal(s): A novel approach combining reconstruction and segmentation during training is investigated, incorporating feedback on clinically relevant features for reconstruction.

Approach: Pretrained reconstruction (E2E-VN) and segmentation models (nnUNet) are connected. The reconstruction model is trained with a weighted combination of reconstruction and segmentation loss. Training and evaluation are performed on fastMRI+ data.

Results: The proposed method resulted in improved image quality of reconstructed images at 8x acceleration compared to baseline E2E-VN, along-with improved downstream segmentation. 

Impact: Training deep-learning-based image reconstruction methods for accelerated MRI with additional feedback on diagnostic content improves image quality in the overall image and the region of interest, and subsequently the diagnostic utility of reconstructed images.

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

Motivation: Clinical routine rectal cancer imaging requires high resolution and hence increased number of excitations to achieve sufficient signal to noise ratio (SNR).

Goal(s): The deep learning method is applied to improve the image quality and imaging speed of rectal magnetic resonance imaging.

Approach: We investigate the use of deep learning based reconstruction in shortening the scan time of the T2-weighted TSE imaging (T2DL) sequence in rectal cancer imaging.

Results: The results show that the DL reconstruction improves the SNR and CNR of the images. Also, the image acquisition time can be reduced by reconstructing images with reduced number of excitations by deep learning.

Impact: Deep learning reconstruction may lead to unprecedented improvements in SNR and CNR compared to conventional reconstruction algorithms, which may be used to obtain higher quality images. In addition, deep learning methods can indirectly shorten image acquisition time.

Keywords: Image Reconstruction, Machine Learning/Artificial Intelligence

Motivation: Image distortions caused by encoding perturbations and slow MR acquisitions compromise real-time MRI-guided radiotherapy treatments.

Goal(s): We aim to develop and investigate a diffusion model-based method to accelerate MR reconstruction with encoding perturbations.

Approach: The diffusion model was trained by 180,670 T1-weighted brain images from a public MR dataset and nonuniform fast Fourier transform was applied to operate forward encoding process with perturbations.

Results: Imaging results showed that the proposed network enabled fast MR image reconstruction with corrected geometric distortions for any subsampling patterns.

Impact: The developed diffusion model to accelerate MR reconstruction with perturbations. The results demonstrated that the proposed method enabled fast distortion-corrected image reconstruction for any subsampling patterns. 

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, knee, c-spine

Motivation: Deep learning (DL) is a powerful tool for MR image formation tasks and MR data at ultra-low-field (ULF) strength has significantly lower SNR than high-field.

Goal(s): Enhancing the image quality of ULF knee and c-spine data at 0.05T via DL reconstruction.

Approach: We extend our recently developed 3D DL partial Fourier reconstruction and superresolution (PF-SR) method on PF-sampled low-resolution noisy brain data to knee and c-spine data.

Results: The preliminary results demonstrate PF-SR, trained on synthetic ULF data simulated from high-field data, can reduce noise and artifacts, and enhance spatial resolution in experimental ULF knee and c-spine data, acquired from 0.05T MRI platform.

Impact: Through leveraging the homogeneous human knee and spine anatomy available in high-field data to enhance the image quality of ultra-low-field knee and spine MRI at 0.05T via deep learning reconstruction in a low-cost and shielding-free 0.05T MRI platform.

Keywords: Synthetic MR, Neuro, Multi-Contrast, Data Acquisition, Machine Learning, Synthetic MR Neuro

Motivation: As the public demand for MRI grows exponentially, there is an increasing need for a one-click 3D MR exam that can generate multiple image contrasts and parametric maps as an effective way to improve patient throughput.

Goal(s): Our goal was to implement an acquisition and reconstruction method that makes high-resolution whole brain multi contrast examination possible in less than 3 minutes. 

Approach: We implemented a deep learning-guided vast undersampled MR acquisition and a time efficient recon algorithm that uses a densely connected unrolled neural network. 

Results: Our proposed method preserved image quality and quantitative accuracy of the multicontrast and multiparametric images.

Impact: This study demonstrates that with highly undersampled 3D QALAS acquisition combined with the DL recon algorithm, a 3-minute one-click exam is feasible that generates whole-brain high-resolution brain volumes of multiple contrasts and quantitative maps, which can enhance patient workflow in a busy clinical practice.

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

Motivation: MRI scans are often sensitive to subject motion, impacting image quality. One challenge is that it is difficult to detect and mitigate motion-related issues until after the scan has completed.

Goal(s): To create a quantitative method for real-time evaluation of MRI scans, identifying events that may corrupt images and enabling prompt decision-making.

Approach: The study used simulated data from the ACDC dataset, training a ResNET18 neural network to predict image quality using SSIM scores.

Results: Our method can quickly and accurately assess MRI image quality. This could aid  motion event detection. However,  validation on actual data is needed.

Impact: This study introduces an automated, deep-learning based method for real-time assessment of motion-related image quality for cardiac MRI. This innovation can potentially enhance the reliability and efficiency of MRI scans.