View Presentation Video

Xiaofeng Yang¹, Yang Lei¹, Tonghe Wang², Sagar Mandava³, Tian Liu⁴, Jonathon Nye¹, and Hui Mao^1
1Emory University, Atlanta, GA, United States, ²Memorial Sloan Kettering Cancer Center, New York, NY, United States, ³GE Healthcare, Atlanta, GA, United States, ⁴Icahn School of Medicine at Mount Sinai, New York, NY, United States

Keywords: PET/MR, Machine Learning/Artificial Intelligence

This study aims to develop an efficient and clinically applicable deep-learning method using routine T1-weighted MRI and non-AC PET to directly synthesize high-quality attenuation-corrected PET without the need of an attenuation coefficient map for simultaneous PET/MRI.

View Presentation Video

Fasil Gadjimuradov^1,2, Laura Pfaff¹, Thomas Benkert², Marcel Dominik Nickel², and Andreas Maier^1
1Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, ²MR Application Predevelopment, Siemens Healthcare GmbH, Erlangen, Germany

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, Transformer, Multiple Instance Learning

Despite its proven clinical value, Diffusion-weighted Imaging (DWI) suffers from several technical limitations associated with prolonged echo trains in single-shot sequences. Parallel Imaging with sufficiently high under-sampling enabled by Deep Learning-based reconstruction may mitigate these problems. Newly emerged architectures relying on transformers demonstrated high performance in this context. This work aims at developing a transformer-based reconstruction method tailored to DWI by utilizing the availability of multiple image instances for a given slice. Redundancies are exploited by jointly reconstructing images using attention mechanisms which are performed across the set of instances. Benefits over reconstructing images separately from each other are demonstrated.

View Presentation Video

Kangping Wang¹, Chentao Cao¹, Zhuoxu Cui¹, Yuanyuan Liu¹, Hairong Zheng¹, Dong Liang¹, and Yanjie Zhu^1
1Shenzhen Institutes of Advanced Technology,Chinese Academy of Sciences, Shenzhen, China

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Diffusion-based generative models have been applied to solve the inverse problem of MR reconstruction and show impressive results. However, the diffusion model requires many iterations to produce high-quality samples, prolonging the reconstruction time. It also may lead to stochastic differential equation (SDE) sequence divergence in MR reconstruction and degrades the reconstruction quality. We proposed a new SDE for diffusion-based MR reconstruction that focuses on the diffusion process in high-frequency of k-space to improve reconstruction robustness and reduce the iterations. We applied the proposed method in MR T1ρ mapping, showing that it can achieve a high acceleration of 14X.

View Presentation Video

Lina Felsner¹, Carlos Velasco¹, Haikun Qi², Karl P. Kunze^1,3, Radhouene Neji^1,3, René M. Botnar¹, and Claudia Prieto^1
1School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²School of Biomedical Engineering, ShanghaiTech University, Shanghai, China, ³MR Research Collaborations, Siemens Healthcare Limited, Camberley, United Kingdom

Keywords: Machine Learning/Artificial Intelligence, Motion Correction

Myocardial T₁ and T₂ mapping play an important role in the assessment of cardiovascular disease. 3D whole-heart joint T₁/T₂ water/fat mapping approaches have been recently proposed, however they require long reconstruction times. Recently a Machine learning based reconstruction was proposed for joint motion correction and motion corrected image reconstruction of undersampled free-breathing single contrast 3D coronary MR angiography. Here, we extend this approach for non-rigid motion-corrected reconstructions for multi-contrast data for joint T₁/T₂ mapping. The proposed approach achieves good agreement with reference techniques and comparable image quality to state-of-the-art methods albeit in 1200 times shorter reconstruction times.

View Presentation Video

Jialong Li¹, Qiqi Lu¹, Yanqiu Feng¹, and Xinyuan Zhang^1
1Department of Biomedical Engneering, Southern Medical University, Guangzhou, China

Keywords: Machine Learning/Artificial Intelligence, Diffusion Tensor Imaging

Diffusion tensor imaging (DTI) is widely used in clinical applications and neuroscience. Its practical utility is limited by the need for multiple scans. Here, we integrate deep learning and model-based optimization methods to estimate diffusion tensor using only one non-diffusion-weighted images and six diffusion-weighted images. The data fidelity term is the weighted linear least squares fitting (WLLS) and the regularization term is Regularization by Denoising (RED). The Alternating Direction Method of Multiplier (ADMM) is adopted to iteratively optimize the model. Experiment results demonstrate that the proposed model-based strategy has great potential to improve the accuracy of diffusion tensor estimation.

View Presentation Video

Guohua Zhao¹, Mengyang He², Yizhou Su³, Yusong Lin³, Eryuan Gao¹, Jie Bai¹, Xiaoyue Ma¹, Huiting Zhang⁴, Xu Yan⁴, Guang Yang⁵, and Jingliang Cheng^1
1The Department of Magnetic Resonance Imaging, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, China, ²School of Cyber Science and Engineering, Zhengzhou University, Zhengzhou, China, ³Collaborative Innovation Center for Internet Healthcare, Zhengzhou, China, ⁴MR Scientific Marketing, Siemens Healthcare, Shanghai, China, ⁵Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China

Keywords: Machine Learning/Artificial Intelligence, Tumor

Preoperative differentiation between glioblastomas (GBM) and solitary brain metastases (SBM) would aid in appropriate treatment planning and follow-up. Neurite orientation dispersion and density imaging (NODDI) can identify diverse tissue components within tumors. Texture analysis can be used to extract and quantify these tissue inhomogeneities. In this study, we extracted texture features using NODDI and validated the NODDI metric map models and a combination model to discriminate between GBM and SBM. Finally, the combined NODDI model achieved the best discriminative power. Texture analysis based on NODDI has great potential for distinguishing between GBM and SBM.

View Presentation Video

Xiaohua Chen¹, Zhiqiang Chen², Zhuo Wang¹, Shaoru Zhang¹, Yunshu Zhou¹, Shili Liu¹, Ruodi Zhang¹, Yuhui Xiong³, and Aijun Wang^4
1Clinical medicine school of Ningxia Medical University, Yinchuan, China, ²Department of Radiology ,the First Hospital Affiliated to Hainan Medical College, Haikou, China, ³GE Healthcare MR Research, Beijing, China, ⁴Department of Radiology, General Hospital of Ningxia Medical University, Yinchuan, China

Keywords: Machine Learning/Artificial Intelligence, Brain

This study aims to propose a fully automatic approach based on convolutional neural networks (CNNs) to predict the O6-Methylguanine-DNA-methyltransferase (MGMT) methylation status of gliomas using conventional pre-operative MR images. It was shown that the Markov Random Field-U-Net network can accurately segment the tumor region, and the improved 34-layer Resnet network can predict the MGMT methylation status effectively. This model has the potential to be a practical tool for the non-invasive characterization of gliomas to help the individualized treatment planning.

View Presentation Video

Amirmohammad Shamaei^1,2 and Rudy Rizzo^3,4
1Institute of Scientific Instruments of the Czech Academy of Sciences Research institute in Brno, Brno, Czech Republic, ²Department of Biomedical Engineering, Brno University of Technology, Brno, Czech Republic, ³Magnetic Resonance Methodology, Institute of Diagnostic and Interventional Neuroradiology, University of Bern, Bern, Switzerland, Bern, Switzerland, ⁴Translational Imaging Center, sitem-insel, Bern, Switzerland

Keywords: Machine Learning/Artificial Intelligence, Spectroscopy, deep learning, machine learning, convolutional neural network, metabolite quantification, Bayesian neural network, variational autoencoder, epistemic uncertainty, aleatoric uncertainty

While deep learning (DL)-based approaches have been employed to quantify MRS signals, the interpretability of deep-model results and assessing what a DL model knows is a crucial component of DL-based approaches. We present a physics-informed DL-based algorithm for MRS data quantification with simultaneous uncertainty estimation, which uses the advantages of linear combination model fitting and the capabilities of ensembles of variational autoencoders. We acknowledge the need for further investigation with in-vivo datasets and other approaches. Furthermore, a more thorough analysis that includes scores that take estimation variation and uncertainty for both the proposed DL technique and traditional model fitting is required.

View Presentation Video

Kevin Koch¹, Nikolai Mickevicius¹, Robin Ausman¹, and Andrew S Nencka^1
1Radiology, Medical College of Wisconsin, Milwaukee, WI, United States

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

We present a feasibility study exploring the use deep learning (DL) methods to enhance the image quality of isotropic 3D-multi-spectral metal artifact suppressed images. Conventional high resolution 2D fast/turbo spin echo images are used as network training labels by masking regions corrupted by metal artifacts.  A pilot set of 3 cases deploying the presented concept to T2-weighted imaging of the instrumented spine are analyzed.   The quantitative results of this analysis demonstrate improved spinal cord contrast, image resolution, and general agreement with 2D-FSE images when the DL enhancement model is inferred on the complete 3D-MSI datasets. 

View Presentation Video

Quentin Uhl¹, Tommaso Pavan¹, Thorsten Feiweier², Erick Jorge Canales-Rodríguez³, and Ileana Jelescu^1
1Dept. of Radiology, Lausanne University Hospital (CHUV), Lausanne, Switzerland, ²Siemens Healthcare GmbH, Erlangen, Germany, ³Signal Processing Lab 5 (LTS5), École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

Keywords: Microstructure, Modelling, Acquisition Protocol

View Presentation Video

Jonathan Trevathan¹, Jonathan Scott¹, Joshua Trzasko¹, Armando Manduca¹, John Huston¹, Richard Ehman¹, and Matthew Murphy^1
1Mayo Clinic, Rochester, MN, United States

Keywords: Data Acquisition, Data Acquisition, MRE,Elastrography

To improve the clinical value of brain stiffness measurements, increased resolution is desired to accurately map the mechanical signatures of disease processes. However, MR elastography-based stiffness maps must be estimated by inverting the measured displacement fields. It is not well-established that increased acquisition resolution will directly translate to improved accuracy in the final mechanical property maps. In this simulation study, using two neural network inversions, we show 2-mm data outperforms 3-mm data for stiffness accuracy and precision across a range of signal-to-noise ratios.

View Presentation Video

Lifeng Mei¹, Celia.M. Dong², Sixing Liu¹, Shoujin Huang¹, and Mengye Lyu^1
1College of Health Science and Environmental Engineering, Shenzhen Technology University, Shenzhen, China, ²University Research Facility in Behavioral and Systems Neuroscience, The Hong Kong Polytechnic University, Hong Kong, China

Keywords: Image Reconstruction, Parallel Imaging

In this study, we combine the readout-concatenation framework with VarNet based deep learning method to achieve SMS reconstruction of low noise amplification. It can flexibly handle arbitrary slice aliasing patterns with in-plane acceleration. Moreover, the modified network architecture allows coil calibration using a prescan of different contrasts, which is preferred in many scenarios.

View Presentation Video

Edengenet Mashilla Dejene^1,2 and Christoph Kolbitsch^1
1Physikalisch - Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, ²Charité Universitätsmedizin Berlin, Berlin, Germany

Keywords: Liver, DSC & DCE Perfusion

Respiratory motion can impair the accurate estimation of physiological parameters in DCE-MR of the liver. A Deep learning network is proposed to quantitatively investigate the impact of respiratory motion on the estimation of physiological parameter maps. The proposed network provides quantitative parameters for DCE-MR and uncertainty estimates for these parameters. Here we could show that the estimated epistemic uncertainty of  k_trans is sensitive to motion. This could provide important information about how well motion correction worked and how reliable the obtained quantitative DCE parameters are.

View Presentation Video

Xinyi Wang^1,2, Zihao Tang^1,2, Mariano Cabezas², Arkiev D’Souza², Fernando Calamante², Dongnan Liu^1,2, Michael Barnett^2,3, Sicong Tu², Weidong Cai¹, and Chenyu Wang^2,3
1School of Computer Science, University of Sydney, Sydney, Australia, ²Brain and Mind Centre, University of Sydney, Sydney, Australia, ³Sydney Neuroimaging Analysis Centre, University of Sydney, Sydney, Australia

Keywords: Brain Connectivity, Tractography & Fibre Modelling, Fiber Orientation Distribution

Modern structural brain connectome pipelines and tractography techniques heavily rely on the quality of the diffusion weighted image acquisition (angular resolution) and the subsequent estimation of the fiber orientation distributions (FODs) for each voxel. Generating reliable connectomes from low angular single-shell acquisitions in clinical scenarios remains a challenging task. This work presents an end-to-end deep learning framework to enhance FOD estimates according to multi-shell acquisitions from low angular single-shell acquisitions to guarantee high quality tractography and connectomes within acceptable time and resources.

View Presentation Video

Michael Dubiner¹, Jan Sedlacik^1,2,3,4, Tom Wilkinson^1,2,3, Pip Bridgen^1,2,3, Franck Mauconduit⁵, Alexis Amadon⁵, Sharon Giles^1,2,3, Radhouene Neji^1,3,6, Joseph V. Hajnal^1,2,3, Shaihan J. Malik^1,2,3, and Raphael Tomi-Tricot^1,2,3,6
1Biomedical Engineering Department, School of Biomedical Engineering & Imaging Sciences, King's College London, London, United Kingdom, ²Centre for the Developing Brain, School of Biomedical Engineering & Imaging Sciences, King's College London, London, United Kingdom, ³London Collaborative Ultra high field System (LoCUS), London, United Kingdom, ⁴Great Ormond Street Hospital for Children, London, United Kingdom, ⁵Paris-Saclay University, CEA, CNRS, BAOBAB, NeuroSpin, Gif-sur-Yvette, France, ⁶MR Research Collaborations, Siemens Healthcare Limited, Frimley, United Kingdom

Keywords: RF Pulse Design & Fields, High-Field MRI

A fast B₁⁺ mapping sequence such as saturation-prepared turbo FLASH (satTFL) is desirable for online pulse design at ultra-high field, but it often results in inaccuracies. Previous work performed linear fitting of the B₁⁺ magnitude obtained with the satTFL over that of the longer but more accurate Actual Flip angle Imaging (AFI) sequence to obtain calibration parameters that would correct the satTFL B₁⁺ on the fly for brain imaging. In this work we introduce a new machine-learning-based method that uses additional features to create a more precise and less location-dependent accuracy.

View Presentation Video

Yang Gao^1,2, Tianyi Ding², Martijn Cloos², and Hongfu Sun^2
1Central South University, Changsha, China, ²The University of Queensland, Brisbane, Australia

Keywords: Contrast Mechanisms, MR Fingerprinting, relaxometry reconstruction, B0 estimation, self-supervised deep learning

Deep neural networks are increasingly employed in MRF reconstruction recently, as an alternative for conventional dictionary matching. However, most previous deep leanring MRF networks were trained based on the dictionary pairs (generated using theoretical bloch equation) or in vivo acquisitions paired with dictionary matching reconstructions (as training labels), whose reconstruction accuracy relies heavily on the dictionary construction and the number of sampling arms. In this work, we propose a self-supversied deep learning MRF method, namely MRF-Mixer, which can result in more accurate MRF reconstructions compared with dictionary matching.   

View Presentation Video

Ruiyang Zhao^1,2, Yahang Li^1,3, Zepeng Wang^1,3, Aaron Anderson^1,4, Paul Arnold^1,4, Graham Huesmann^1,4,5, and Fan Lam^1,2,3
1Beckman Institute for Advanced Science and Technology, Urbana, IL, United States, ²Department of Electrical and Computer Engineering, University of Illinois Urbana-Champaign, Urbana, IL, United States, ³Department of Bioengineering, University of illinois Urbana-Champaign, Urbana, IL, United States, ⁴Neuroscience Institute, Carle Foundation Hospital, Urbana, IL, United States, ⁵School of Molecular and Cellular Biology, University of Illinois Urbana-Champaign, Urbana, IL, United States

Keywords: Spectroscopy, Machine Learning/Artificial Intelligence

We introduced a data-driven denoiser trained in a self-supervised fashion as a novel spatial-temporal constraint for MRSI reconstruction. Our proposed denoiser was trained using noisy data only via the Noise2void framework that trains an interpolation network exploiting the statistical differences between spatiotemporally correlated signals and uncorrelated noise. The trained denoiser was then integrated into an iterative MRSI reconstruction formalism as a Plug-and-Play prior. An additional physics-based subspace constraint was also included into the reconstruction. Simulation and in vivo results demonstrated impressive SNR-enhancing reconstruction ability of the proposed method, with improved performance over a state-of-the-art subspace method.

View Presentation Video

Mingming Wu¹, Arum Somasundaram¹, Selina Rupp¹, Jessie Han¹, Stella Naebauer¹, Daniela Junker¹, Anna Reik², Meike Wiechert², Hans Hauner^2,3, Christina Holzapfel², and Dimitrios C. Karampinos^1
1Department of Diagnostic and Interventional Radiology, School of Medicine, Technical University of Munich, School of Medicine, Technical University of Munich, Munich, Germany, ²Institute for Nutritional Medicine, School of Medicine, Technical University of Munich, Munich, Germany, ³Else Kroener-Fresenius-Center of Nutritional Medicine, School of Life Sciences, Technical University of Munich, Freising, Germany

Keywords: Fat, Metabolism, Obesity

A deep-learning algorithm based on the nnU-Net and using water-fat images enabled robust automatic segmentation of abdominal organs including visceral and subcutaneous adipose tissue, liver, iliopsoas and erector spinae muscle groups. Each organ's volume and fat content were examined in a weight loss study comprising 127 subjects with BMI of 30-39.9kg/m², who followed a low caloric diet (LCD). Dixon water-fat images were acquired before and after diet. Differences in fat distribution among abdominal organs and fat content was assessed among both sexes. Differences in the changes of organ volume and fat fraction as a response to the LCD were revealed.

View Presentation Video

Madeline Hess¹, Sharmila Majumdar¹, and Valentina Pedoia^1
1University of California San Francisco, San Francisco, CA, United States

Keywords: Osteoarthritis, Osteoarthritis

We find that the presence or absence of ROAMES morphological phenotypes as predicted from MRI is a useful indicator of osteoarthiris progression over 8 years as measured by cartilage thinning.

View Presentation Video

Brian Toner¹, Simon Arberet², Fei Han³, Mariappan Nadar², Vibhas Deshpande⁴, Diego Martin⁵, Maria Altbach⁶, and Ali Bilgin^7,8
1Applied Mathematics, The University of Arizona, Tucson, AZ, United States, ²Digital Technology & Innovation, Siemens Healthineers, Princeton, NJ, United States, ³Siemens Healthineers, Los Angeles, CA, United States, ⁴Siemens Healthineers, Austin, TX, United States, ⁵Radiology, Houston Methodist, Houston, TX, United States, ⁶Medical Imaging, The University of Arizona, Tucson, AZ, United States, ⁷Electrical and Computer Engineering, The University of Arizona, Tucson, AZ, United States, ⁸Biomedical Engineering, The University of Arizona, Tucson, AZ, United States

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Unrolled networks with data consistency layers have been shown to be effective in reconstructing MRI data. Radial turbo spin echo sequences enable acquisition of multi-contrast k-space data, which can be used to generate multi-contrast images at different echo times together with a co-registered T2 map. In this work, we will show that the cascading unrolled network architecture is effective in reconstructing images from radial turbo spin echo data. In order to do this, data consistency layers must be implemented to be able to combine data from multi-coil acquisitions and from multiple echo times.