View Presentation Video

Álvaro Planchuelo-Gómez^1,2, Maxime Descoteaux³, Santiago Aja-Fernández², Jana Hutter⁴, Derek K. Jones¹, and Chantal M.W. Tax^1,5
1CUBRIC, Cardiff University, Cardiff, United Kingdom, ²Imaging Processing Laboratory, Universidad de Valladolid, Valladolid, Spain, ³SCIL, Université de Sherbrooke, Sherbrooke, QC, Canada, ⁴Centre for Medical Engineering, Centre for the Developing Brain, King's College London, London, United Kingdom, ⁵Image Sciences Institute, University Medical Centre Utrecht, Utrecht University, Utrecht, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Data Acquisition, Quantitative Imaging

Multi-contrast MRI is used to assess the biological properties of tissues, but excessively long times are required to acquire high-quality datasets. To reduce acquisition time, physics-informed Machine Learning approaches were developed to select the optimal subset of measurements, decreasing the number of volumes by approximately 63%, and predict the MRI signal and quantitative maps. These selection methods were compared to a full data-driven and two manual strategies. Synthetic and real 5D-Diffusion-T1-T2* data from five healthy participants were used. Feature selection via a combination of Machine Learning and physics modelling provides accurate estimation of quantitative parameters and prediction of MRI signal.

View Presentation Video

Peyman Shokrollahi¹, Juan M. Zambrano Chaves¹, Avishkar Sharma¹, Jonathan P.H. Lam¹, Debashish Pal², Naeim Bahrami², Akshay S. Chaudhari¹, and Andreas M. Loening^1
1Radiology, Stanford University, Stanford, CA, United States, ²GE Healthcare, Sunnyvale, CA, United States

Keywords: Machine Learning/Artificial Intelligence, Modelling, MR Radiology Workflow

Inaccurate selection of MRI protocols can impede diagnostics and therapeutic workflows, delay appropriate treatment, increase misdiagnosis likelihoods, and increase healthcare costs. Here we depict a machine-learning (ML) based system to accurately predict abdominal MR protocols, trained on the electronic medical records from 11,251 MR exam orders on 6,882 patients. The best model achieved a cumulative F1-score of 95.6% for top-three most-often-ordered protocols and a top-one F1 score of 78.5%. The proposed system can guide radiologists to appropriate protocol selections quickly, optimize workflows, and improve diagnostic accuracy, thereby by serving to support optimal patient outcomes.

View Presentation Video

Chen Luo^1,2, Zhuoxu Cui², Huayu Wang^1,2, Qiyu Jin¹, Guoqing Chen¹, and Dong Liang^2
1Inner Mongolia University, Hohhot, China, ²Shenzhen Institute of Advanced Technology，Chinese Academy of Sciences, Shenzhen, China

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Recently, iterative algorithm driven deep neural network unfolding methods have been successfully applied to MRI. However, the network replaces the original algorithm structure and mathematical properties such as interpretability and convergence of the original algorithm are not guaranteed. Fortunately, the k-space-filled Hankel low rank can naturally be associated with convolutional networks. Given this, we propose an unfolding method for k-space-filling, which guarantees convergence to the unique real MR image. Furthermore, we train this network in a self-supervised manner to cope with scenarios where fully sampled data are difficult to obtain. Finally, numerical experiments validate the effectiveness of the proposed method.

View Presentation Video

Zhaowei Cheng¹, Guangxu Han², Songtao Hu², Ke Fang¹, and Ruiliang Bai^3
1College of Information Science and Electronic Engineering, Zhejiang University, Hangzhou, China, ²Key Laboratory of Biomedical Engineering of Ministry of Education, College of Biomedical Engineering and Instrument Science, Zhejiang University, Hangzhou, China, ³School of Medicine, Zhejiang University, Hangzhou, China

Keywords: Machine Learning/Artificial Intelligence, Data Acquisition

We propose a deep-learning-based method to optimize acquisition protocol and estimate parameter for diffusion exchange spectroscopy. A unified framework has been carefully designed to achieve both goals simultaneously. Using this framework, the acquisition protocol can be directly optimized with an objective function that minimizes the parameter estimation errors, regardless of its degree of freedom. The experimental results from Monte Carlo simulations show that the proposed method outperforms the existing methods in both accuracy and precision of parameters’ estimation. Besides, it speeds up the calculation by 480 times.

Yichen Wang¹, Xinxin Zhang¹, Sicong Wang², Xinming Zhao¹, and Yan Chen^1
1Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy, Beijing, China, ²GE Healthcare, MR Research, Beijing, China

Keywords: Prostate, Machine Learning/Artificial Intelligence, Deep learning reconstruction

In this prospective study, feasibility of deep learning reconstruction (DLR) in axial FSE-T2WI and axial reduced-FOV DWI (FOCUS DWI) were evaluated compared with standard protocols. Fast protocol with DLR substantially reduced scanning time (axial FSE-T2WI: -32.1%; FOCUS-DWI: -36.8%). Fast FOCUS DWI with DLR showed the highest SNR and CNR for prostate PZ, TZ and lesion. Fast FSE-T2WI with DLR showed the highest SNR and CNR for prostate PZ and TZ. Moreover, fast FOCUS-DWI and FSE-T2WI with DLR demonstrated equivalent or better image quality than standard images. DLR may be useful in prostate multiparametric MRI protocol optimization and high-quality image acquisition.

View Presentation Video

Sixing Liu¹, Lifeng Mei¹, Shoujin Huang¹, and Mengye Lyu^1
1College of Health Science and Environmental Engineering, Shenzhen Technology University, Shenzhen, China

Keywords: PET/MR, Machine Learning/Artificial Intelligence, deep learning

Applying the Periodically Rotated Overlapping Parallel Lines with Enhanced Reconstruction (PROPELLER) technique is one of the strategies to mitigate motion artifacts in MR images. However, due to technical limitations, existing method estimates motion parameters with unsatisfactory results, motion artifacts will still be presented in the final image. Deep learning algorithms are expected to optimize the motion parameter estimation part of PROPELLER technique. We develop a PROPELLER imaging technique incorporating a deep learning model that can provide accurate results and greatly shorten the elapsed time.

Xin Fang¹, Ailian Liu¹, Qingwei Song¹, Guobin Li², and Shuheng Zhang^2
1the First Affiliated Hospital of Dalian Medical University, Dalian, China, ²Shanghai United Imaging Healthcare Co., Ltd, Shanghai, China

Keywords: Brain Connectivity, Brain

Head Computed Tomography (HCT) is a common imaging method for the diagnosis of emergency cerebral infarction. However, it is easy to miss the diagnosis due to the poor performance of the ultra-early or early cerebral infarction lesions. Compared with CT, head Magnetic Resonance Imaging (MRI) has the characteristics of high tissue contrast, no ionizing radiation damage and can make functional imaging sequences, so it has significant advantages in the diagnosis of posterior fossa lesions and ultra-early cerebral infarction

View Presentation Video

R. Marc Lebel^1,2
1GE Healthcare, Calgary, AB, Canada, ²Radiology, University of Calgary, Calgary, AB, Canada

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Deep learning (DL) assisted image reconstructions are becoming state-of-the art, producing better image quality and/or enabling higher acceleration rates than achievable with conventional methods. DL networks are used to mitigate noise amplification while retaining important signal characteristics. However, typical loss functions produce object-dependent noise alterations and non-uniform point-spread functions. Here we present a method for training networks that prioritizes maximizing the point spread function to ensure maximal detail retention.

View Presentation Video

Jörn Huber¹, Klaus Eickel^1,2, and Matthias Günther^1,2,3
1Fraunhofer Institute for Digital Medicine MEVIS, Bremen, Germany, ²mediri GmbH, Heidelberg, Germany, ³Faculty 1 (Physics/Electrical Engineering), University of Bremen, Bremen, Germany

Keywords: Machine Learning/Artificial Intelligence, Artifacts

Arterial Spin Labeling has great potential in clinics as a non-invasive alternative to Dynamic Contrast Enhanced imaging. However, motion sensitivity needs to be tackled by readout techniques such as 3D GRASE PROPELLER, which unfortunately shows a high sensitivity to geometric distortion. Analytical separation of motion and distortion effects is computationally demanding and might fail in some situations. To this aim, a U-NET based convolutional neural network approach is demonstrated which might overcome this limitation.

View Presentation Video

Zi Wang¹, Haoming Fang¹, Chen Qian¹, Boxuan Shi¹, Lijun Bao¹, Liuhong Zhu², Jianjun Zhou², Wenping Wei³, Jianzhong Lin⁴, Di Guo⁵, and Xiaobo Qu^1
1Department of Electronic Science, Biomedical Intelligent Cloud R&D Center, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, National Institute for Data Science in Health and Medicine, Xiamen University, Xiamen, China, ²Department of Radiology, Zhongshan Hospital (Xiamen), Fudan University, Xiamen, China, ³Department of Radiology, Xiamen University, Xiamen, China, ⁴Department of Radiology, Zhongshan Hospital affiliated to Xiamen University, Xiamen, China, ⁵School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, China

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

Recent deep learning is superior in providing high-quality images and ultra-fast reconstructions in accelerated magnetic resonance imaging (MRI). Faithful coil sensitivity estimation is vital for MRI reconstruction. In this work, we propose a Joint Deep Sensitivity estimation and Image reconstruction network (JDSI). During the image artifacts removal, it gradually provides more faithful sensitivity maps, leading to greatly improved image reconstructions. Results on in vivo datasets and radiologist reader study demonstrate that, the proposed JDSI achieves the state-of-the-art performance visually and quantitatively, especially when the accelerated factor is high. Besides, JDSI also owns nice robustness to abnormal subjects.

View Presentation Video

Xinling Yu¹, Jose Serralles², Ilias Giannakopoulos^3,4, Ziyue Liu⁵, Luca Daniel², Riccardo Lattanzi^3,4, and Zheng Zhang^1
1Department of Electrical and Computer Engineering, University of California, Santa Barbara, Santa Barbara, CA, United States, ²Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, ³Center for Advanced Imaging Innovation and Research (CAI2R), Department of Radiology, New York University Grossman School of Medicine, New York, NY, United States, ⁴Bernard and Irene Schwartz Center for Biomedical Imaging (CBI), Department of Radiology, New York University Grossman School of Medicine, New York, NY, United States, ⁵Department of Statistics and Applied Probability, University of California, Santa Barbara, Santa Barbara, CA, United States

Keywords: Machine Learning/Artificial Intelligence, Electromagnetic Tissue Properties

We introduce physics-informed Fourier networks (PIFNs) for Electrical Properties (EP) Tomography (EPT). Our novel deep learning-based method is capable of learning EPs globally from noisy magnetic resonance (MR) measurements, i.e, the magnitude of the magnetic transmit field and the transceive phase. Our proposed method also provides noise-free transmit field reconstructions. Two separate Fourier neural networks are used to efficiently estimate the transmit field and EPs at any location. We show that PIFN EPT accurately infers the EPs distribution of an inhomogeneous phantom from noisy simulated measurements.

View Presentation Video

Thomas Campbell Arnold¹, Serhat V Okar², Danni Tu³, Govind Nair², John T. Pitts⁴, Megan E. Poorman⁴, Karan D. Kawatra², Lisa M. Desiderio⁵, Matthew K. Schindler⁶, Brian Litt⁶, Russell T. Shinohara³, Daniel S. Reich², and Joel M. Stein^5
1Bioengineering, University of Pennsylvania, Philadelphia, PA, United States, ²National Institute of Neurological Disorders and Stroke, Bethesda, MD, United States, ³Biostatistics, University of Pennsylvania, Philadelphia, PA, United States, ⁴Hyperfine, Guilford, CT, United States, ⁵Radiology, University of Pennsylvania, Philadelphia, PA, United States, ⁶Neurology, University of Pennsylvania, Philadelphia, PA, United States

Keywords: Machine Learning/Artificial Intelligence, Low-Field MRI, super resolution

High-field MRI provides superior imaging for diverse clinical applications, but cost and other factors limit availability in various healthcare and lower resource settings. Lower-field strength units promise to expand access but involve tradeoffs including reduced signal, longer scan times, and lower resolution. Here we develop super-resolution methods that can generate high-field quality images from low-field scanner inputs, thus increasing signal and resolution. We use generative adversarial networks to demonstrate image enhancement in T1, T2 and FLAIR sequences.

View Presentation Video

Reina Ayde^1,2, Tobias Senft¹, Marco Fiorito¹, Mauro Spreiter¹, Najat Salameh^1,2, and Mathieu Sarracanie^1,2
1Center for Adaptable MRI Technology (AMT center), Department of Biomedical Engineering, University of Basel, Allschwil, Switzerland, ²AMT center, Institute of Medical Sciences, School of Medicine, Medical Sciences & Nutrition, University of Aberdeen, Aberdeen, United Kingdom

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, undersampling, averaging, data sampling

Low signal-to-noise (SNR) ratios inherent to low-field (LF) MRI challenge its relevance in clinical applications. Accelerating the acquisition by undersampling k-space followed by reconstruction techniques has already shown promising results. Yet, undersampling is usually done by skipping high-frequency information which can lead to misdiagnosis as small lesions can be missed. In this study, we exploited a specificity of low-SNR regimes, that is signal averaging, to explore different acceleration strategies without skipping crucial information in k-space. The DL-reconstructed images arising from those sampling schemes have been evaluated on acquired in-vivo and ex-vivo LF-MRI datasets, showcasing high-frequency preservation and potential for generalization.

View Presentation Video

Yuan Lian¹ and Hua Guo^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Compressed Sensing is widely used for accelerating acquisitions of MR images. Deep learning methods have been introduced to CS-MRI reconstruction to improve image quality and computation speed. Here we introduce a new shrinkage function with adaptive threshold selection for Model-driven deep learning networks to suppress aliasing artifacts by utilizing the information on each feature map. We combine our adaptive threshold selection module with ISTA-Net, and demonstrate that the method can reduce reconstruction errors while preserving structural details effectively.

View Presentation Video

Danyal Bhutto^1,2, Bo Zhu², Jeremiah Zhe Liu^3,4, Stephen Cauley^2,5, Neha Koonjoo^2,5, Bruce R Rosen^2,5, and Matthew S Rosen^2,5,6
1Biomedical Engineering, Boston University, Boston, MA, United States, ²Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA, United States, ³Google, Mountain View, CA, United States, ⁴Biostatistics, Harvard University, Cambridge, MA, United States, ⁵Harvard Medical School, Boston, MA, United States, ⁶Physics, Harvard University, Boston, MA, United States

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Domain-transform manifold learning is a trained reconstruction approach where care needs to be taken to appropriately represent the forward encoding model during training, including for example the numerical properties of the source sensor data, phase relationship of complex sensor data, and field-of-view to prevent artifacts arising in the reconstruction. Here, we study the role that the training corpus and the numeral properties of the training have on the performance of the reconstruction of MRI data and demonstrate reconstruction artifacts that result from inference on out-of-training-distribution data if the training data is not augmented sufficiently.

View Presentation Video

Suryanarayanan Sivaram Kaushik¹, Xucheng Zhu², Robert Marc Lebel³, Kavitha Manickam¹, Ke Li¹, Kailash Saravanan¹, Florian Wiesinger⁴, and Martin Janich^4
1GE Healthcare, Waukesha, WI, United States, ²GE Healthcare, Menlo Park, CA, United States, ³GE Healthcare, Calgary, AB, Canada, ⁴GE Healthcare, Munich, Germany

Keywords: Machine Learning/Artificial Intelligence, Cardiovascular

Deep Learning (DL) based reconstructions help alleviate the longer scan times seen in bSSFP Cine acquisitions by offering higher acceleration factors. This work analyzes the spatial and temporal variations in the noise as a function of acceleration factors in DL reconstructions of highly accelerated bSSFP Cine data.

View Presentation Video

Jeewon Kim¹, Wonil Lee¹, Beomgu Kang¹, Seohee So², and HyunWook Park^1
1Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, Republic of, ²Korea Institute of Science and Technology, Seoul, Korea, Republic of

Keywords: Machine Learning/Artificial Intelligence, Parallel Imaging

We propose a new Parallel Imaging scheme using a deep neural network which performs well with fewer ACS line in noisy environments. The proposed scheme includes ACS loss which is used in RAKI and cycle interpolation loss that we newly propose in our work. RAKI generalized GRAPPA in noisy environments by applying non-linear k-space interpolation with a deep neural network. However, it requires additional ACS lines to output satisfactory performance. Here, we suggest a new scheme to overcome the reconstruction performance in a noisy environment with fewer ACS lines.

Xinxin Zhang¹, Yichen Wang¹, Sicong Wang², Min Li², Yan Chen¹, and Xinming Zhao^1
1National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, 100021, China, Beijing, China, ²GE Healthcare, MR Research China, Beijing, Beijing, China

Keywords: Urogenital, Bladder, Deep Learning Reconstruction

The application of DLR significantly shortened scan times and improved the overall image quality score and image artifacts score and SNR and CNR of FSE-T2WI. DLR fast FSE-T2WI demonstrated significantly higher SNR (256.7±102.9 VS 94.7±40.8, p < 0.05) and CNR (168.0±77.3 VS 59.6±29.8, p < 0.05) and overall image quality scores (median, 4.0 vs. 3.0 for reader1 and 4.0 vs. 3.5 for reader2) than those of conventional FSE-T2WI. DLR may be useful in reducing the acquisition time of bladder MRI without compromising image quality.

View Presentation Video

Felix Krueger¹, Christoph Stefan Aigner¹, Max Lutz¹, Layla Tabea Riemann¹, Katja Degenhardt¹, Bernd Ittermann¹, Tobias Schaeffter^1,2, Kerstin Hammernik^3,4, and Sebasian Schmitter^1,5,6
1Physikalisch-Technische Bundesanstalt, Berlin and Braunschweig, Germany, ²Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom, ³Technical University of Munich, Munich, Germany, ⁴Imperial College London, London, United Kingdom, ⁵Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, ⁶Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States

Keywords: RF Pulse Design & Fields, Parallel Transmit & Multiband

In this work we utilize deep learning to estimate multi-slice whole-brain B₁⁺-maps in sub-seconds from initial localizer scans at 7T. The investigated neural networks use the receive profiles of the individual coil elements of an 8Tx/8Rx transceiver head coil as input information. The networks are trained on seven volunteers and tested in 2 unseen subjects for transversal/coronal/sagittal slices by comparing the prediction with the acquired B₁⁺-maps. Subsequently, the feasibility of using the DL-based B₁⁺-maps in a subject-specific calibration pipeline is demonstrated.

View Presentation Video

Zheyuan Hu^1,2, Zihao Chen^1,2, Hsu-Lei Lee¹, Yibin Xie¹, Debiao Li^1,2, and Anthony Christodoulou^1,2
1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ²Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States

Keywords: Machine Learning/Artificial Intelligence, Quantitative Imaging

Cardiovascular MR (CMR) Multitasking can quantify various parameter combinations without breath-holding or ECG monitoring. Clinically practical reconstruction time is viable using supervised deep subspace learning, but it depends on sequence-specific training. Here we explore whether universal, sequence-invariant CMR Multitasking deep learning reconstruction is practical by trading temporal awareness (breadth) for added depth in spatial domain. We evaluated the performance and generalizability of both strategies by training on T1 mapping data only and testing on two datasets: a) a matched-sequence T1 mapping data; and b) a novel-sequence T1-T2 mapping data.