Recent years have seen a resurgence of interest in inexpensive low magnetic field (< 0.3 T) MRI systems mainly due to advances in magnet, coil and gradient set designs. Most of these advances have focused on improving hardware and signal acquisition strategies, and far less on the use of advanced image reconstruction methods to improve attainable image quality at low field. We describe here the use of our end-to-end deep neural network approach (AUTOMAP) to improve the image quality of highly noise-corrupted low-field MRI data. We compare the performance of this approach to two additional state-of-the-art denoising pipelines. We find that AUTOMAP improves image reconstruction of data acquired on two very different low-field MRI systems: human brain data acquired at 6.5 mT, and plant root data acquired at 47 mT, demonstrating SNR gains above Fourier reconstruction by factors of 1.5- to 4.5-fold, and 3-fold, respectively. In these applications, AUTOMAP outperformed two different contemporary image-based denoising algorithms, and suppressed noise-like spike artifacts in the reconstructed images. The impact of domain-specific training corpora on the reconstruction performance is discussed. The AUTOMAP approach to image reconstruction will enable significant image quality improvements at low-field, especially in highly noise-corrupted environments.

“Just Accepted” papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review ...

Objective: Magnetic resonance imaging of the breast and the chest wall imaging was performed at ultra-low field (ULF) using optimal radiofrequency (RF) coil designs and, with an eye toward future application in MR-guided proton therapy (MRgPT), simulation studies of a pencil proton beam trajectory in low and ultra-low magnetic fields were performed. Methods: The primary focus in this study is the RF coil design, which is approached as a multi-objective optimization (MOO) problem and effectively solved using a Pareto optimality formalism. Our coil winding pattern algorithm simultaneously optimizes the magnetic field homogeneity of the coil, the attainable MRI signal-to-noise ratio (SNR), and the magnetic field drop-off into the chest wall for improved OAR imaging. Five transmit/receive single channel breast-shaped RF coils were constructed, and 3D phantom MRI was performed at 6.5 mT. The RF magnetic field homogeneity across the imaging volume and SNR were analyzed among the different coils. In vivo breast imaging was performed for the first time at 6.5 mT. Monte Carlo simulations were implemented to study the influence of copper wire on a proton beam, as well as the beam deflection at various static field strengths. Results: Our MOO method was validated - the observed field homogeneity, the relative ratio of attainable SNRs for all five breast coils, and the MRI signal from the chest agree with the simulations. In vivo breast and chest wall images acquired at ULF reveal good breast tissue contrast and can also visualize the heart. Finally, the results of this work indicate that a field strength of ∼50 mT with gradient strengths of ∼10 mT/m to minimize proton beam deflection, and the ability to achieve a ∼2 ´ 2 ´ 5 mm3 spatial resolution within few minutes are essential for making MRgPT clinically feasible. Conclusion: Our MOO method can be used to design breast coils that meet a wide range of imaging requirements.

Medicine is evolving beyond therapy largely predicated on anatomical information and towards incorporating patient-specific molecular biomarkers of disease for more accurate diagnosis and effective treatment. The complementary combination of hyperpolarization by spin-lock induced crossing signal amplification by reversible exchange (SLIC SABRE) and low field magnetic resonance imaging (MRI) can enable accessible metabolic imaging to advance personalized medicine. Hyperpolarized ¹³C-enriched pyruvate has demonstrated promise for imaging metabolism in cancer, heart disease and neurodegenerative disorders; however, broader clinical adoption awaits validated clinical indications, and is further constrained by the cost and limited availability of current hyperpolarization technology. Parahydrogen-based polarization techniques, paired with low-cost high-performance MRI at millitesla fields, offer a means of broadening the reach of metabolic imaging. Here we show results demonstrating in situ hyperpolarization of pyruvate at 6.5 mT by SLIC SABRE, followed by immediate readout without field cycling or sample shuttling. We achieve ¹³C signal enhancements several million times above thermal equilibrium at 6.5 mT, corresponding to polarization levels of approximately 3%. Leveraging this enhancement, we perform ¹³C MRI and acquire NMR spectra with resolution sufficient to distinguish chemical shifts between pyruvate isotopomers. These results show a viable pathway towards accessible metabolic imaging with hyperpolarized ¹³C MRI at ultra-low field.

Accurate image reconstruction is at the heart of diagnostics in medical imaging. Supervised deep learning-based approaches have been investigated for solving inverse problems including image reconstruction. However, these trained models encounter unseen data distributions that are widely shifted from training data during deployment. Therefore, it is essential to assess whether a given input falls within the training data distribution. Current uncertainty estimation approaches focus on providing an uncertainty map to radiologists, rather than assessing the training distribution fit. In this work, we propose a method based on the local Lipschitz metric to distinguish out-of-distribution images from in-distribution with an area under the curve of 99.94% for True Positive Rate versus False Positive Rate. We demonstrate a very strong relationship between the local Lipschitz value and mean absolute error (MAE), supported by a Spearman's rank correlation coefficient of 0.8475, to determine an uncertainty estimation threshold for optimal performance. Through the identification of false positives, we demonstrate the local Lipschitz and MAE relationship can guide data augmentation and reduce uncertainty. Our study was validated using the AUTOMAP architecture for sensor-to-image Magnetic Resonance Imaging (MRI) reconstruction. We demonstrate our approach outperforms baseline techniques of Monte-Carlo dropout and deep ensembles as well as the state-of-the-art Mean Variance Estimation (MVE) network approach. We expand our application scope to MRI denoising and Computed Tomography (CT) sparse-to-full view reconstructions using UNET architectures. We show our approach is applicable to various architectures and applications, especially in medical imaging, where preserving diagnostic accuracy of reconstructed images remains paramount.