View Presentation Video

Juan Pablo Meneses^1,2,3, Cristobal Arrieta^1,2, Pablo Irarrazaval^1,3,4, Cristian Tejos^1,3, Marcelo Andía^1,2,5, Carlos Sing Long^1,4,6, and Sergio Uribe^1,2,5
1Biomedical Imaging Center, Pontificia Universidad Católica de Chile, Santiago, Chile, ²i-Health Millennium Institute for Intelligent Healthcare Engineering, Santiago, Chile, ³Department of Electrical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁴Institute for Biological and Medical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁵Radiology Department, School of Medicine, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁶Institute for Mathematical & Computational Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile

Keywords: Machine Learning/Artificial Intelligence, Quantitative Imaging, Convolutional Neural Network

Liver PDFF is a biomarker correlated with hepatic pathologies. Recently, several Deep Learning (DL) methods have been proposed to accelerate the necessary post-processing to estimate PDFF. However, none of these techniques had been assessed in terms of bias and precision, as suggested by the ISMRM quantitative MR study group. We propose a two-stages framework denoted Variable Echo Times neural Network (VET-Net), which considers multi-echo MR images and their echo times to estimate PDFF. VET-Net showed a bias of -1.35% when tested over a multi-site phantom dataset, and a within-standard deviation of 0.81% over liver MR images with different TEs.

View Presentation Video

Thanh-Duc Nguyen¹, Saurabh Garg¹, Nasrin Akbari¹, Saqib Basar¹, Sean London², Yosef Chodakiewitz², Rajpaul Attariwala^1,2, and Sam Hashemi^1,2
1Voxelwise Imaging Technology Inc., Vancouver, BC, Canada, ²Prenuvo, Vancouver, BC, Canada

Keywords: Machine Learning/Artificial Intelligence, Cardiovascular, Aorta quantification

We present a validated AI technique for quantification of whole aorta diameter from non-cardiac-gated MRI. The mean absolute error between AI-predicted and radiologist-reported diameter is less than 0.27 and 0.249 cm, while Pearson correlation is 0.72 for the ascending (p-value=0.0002) and 0.74 for the descending aorta (p-values=0.0012). SVM classifiers indicated that a 3.15 cm (AUC=0.93) threshold is used to detect a descending aortic aneurysm while a 4.3 cm (AUC=0.92) threshold is used for ascending aortic aneurysm. Large-scale analysis on 3485 individuals showed aortic diameter to increase with age in male and female populations. 

View Presentation Video

Patrick Bolan¹, Sara Saunders¹, Mitchell Gross¹, Kendrick Kay¹, Mehmet Akcakaya², and Gregory Metzger^1
1Center for MR Research / Radiology, University of Minnesota, Minneapolis, MN, United States, ²Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, United States

Keywords: Machine Learning/Artificial Intelligence, Prostate, Relaxometry

This work uses convolutional neural networks (CNNS) with two training strategies for estimating quantitative T2 values from prostate relaxometry measurements, and compares the results to conventional non-linear least squares fitting. The CNN trained with synthetic data in a supervised manner gave lower median errors and better noise robustness than either NLLS fitting or a CNN trained on in vivo data with a self-supervised loss.

View Presentation Video

Qing Lin¹, Weikun Chen¹, Jian Wu¹, Taishan Kang², Xinran Chen¹, Zhigang Wu³, Shuhui Cai¹, and Congbo Cai^1
1Department of Electronic Science, Xiamen University, Xiamen, China, ²Magnetic Resonance Center, Zhongshan Hospital Afflicated to Xiamen University, Xiamen, China, ³MSC Clinical & Technical Solutions, Philips Healthcare, Shenzhen, China

Keywords: Machine Learning/Artificial Intelligence, Quantitative Imaging, water/fat separation

Most of the water/fat separation techniques need to acquire multiple images with different echo time, which usually take long acquisition time. Multiple overlapping-echo detachment (MOLED) imaging can shorten acquisition time by acquiring multiple MR echo signals in the same k-space. Here, a new method for fast water/fat separation T₂* mapping with MOLED technology was proposed, which can obtain quantitative T₂* maps and M₀ maps of water and fat in a single shot. In vivo experiment demonstrates that the proposed method can obtain accurate quantitative parameter values and accurate separation of water and fat images.

View Presentation Video

Dicheng Chen¹, Huiting Liu¹, Yirong Zhou¹, Xi Chen², Zhangren Tu¹, Liangjie Lin³, Zhigang Wu³, Jiazheng Wang³, Di Guo⁴, Jianzhong Lin⁵, and Xiaobo Qu^1
1Department of Electronic Science, National Institute for Data Science in Health and Medicine, Xiamen University, Xiamen, China, ²McLean Hospital, Harvard Medical School, Belmont, MA, United States, ³Philips Healthcare, Beijing, China, ⁴School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, China, ⁵Department of Radiology, The Zhongshan Hospital affiliated to Xiamen University, Xiamen, China

Keywords: Machine Learning/Artificial Intelligence, Brain, Magnetic Resonance Spectroscopy, Quantification

Quantification of ¹H-MRS is difficult because of the overlapping of individual metabolite signals, non-ideal acquisition conditions, and strong background signal interference. We introduced Deep Learning (DL) method to learn these effects to improve the accuracy of the quantification. Results indicate that, compared with the conventional method LCModel, the proposed Qnet (Quantification deep learning network) shows better quantification for both simulated and in vivo acquired MRS data with lower fitting errors and enhanced stability.

View Presentation Video

Daan Kuppens¹, Daisy van den Berg¹, Sebastiano Barbieri², Aart J. Nederveen¹, and Oliver J. Gurney-Champion^1
1Radiology & Nuclear Medicine, Amsterdam UMC, Amsterdam, Netherlands, ²Centre for Big Data Research in Health, University of New South Wales Sydney, Sydney, Australia

Keywords: Machine Learning/Artificial Intelligence, Quantitative Imaging

In quantitative MRI, tissue properties are estimated from MRI data using bio-physical models that relate the MRI signal to the underlying tissue properties via model parameters. Deep learning can improve parameter estimation, but is conventionally dependent on the input being either a fixed set of input signals or a series of regularly sampled signals. Neural controlled differential equations (NCDEs) are models that are independent of the configuration of input data. NCDEs have similar performance to state-of-the-art acquisition-specific deep learning methods in estimating intra-voxel incoherent motion parameters. Therefore, NCDEs are a generic purpose tool for parameter estimation in quantitative MRI.

View Presentation Video

Cecile Maguin¹ and Julien Flament^1
1Université Paris-Saclay, CEA, CNRS, MIRCen, Laboratoire des Maladies Neurodégénératives, Fontenay-aux-roses, France

Keywords: Machine Learning/Artificial Intelligence, CEST & MT, Feature selection

We propose an artificial neural network combined with a feature selection scheme for fast, quantitative CEST imaging, designed for specificity. Our NN was evaluated on glucose phantoms and glutamate/glucose mixed phantoms and goes beyond performances of classical fittings approaches.

Christopher Jiaming Wu¹ and Jia Guo^2
1Biomedical Engineering, Columbia University, New York, NY, United States, ²Columbia University, New York, NY, United States

Keywords: Machine Learning/Artificial Intelligence, Brain, Convolutional Neural Networks, Multi-class Regression, Unsupervised Learning

Quantification of metabolites in the human brain in vivo from magnetic resonance spectra (MRS) has many applications in medicine and psychology, but it remains a challenging task despite considerable research efforts. In this paper, we propose quantification of metabolites from MEGA-PRESS data using deep learning through an unsupervised learning approach. A regression framework based on the Convolutional Neural Networks (CNN) is introduced for estimation of spectral parameters including the relative concentrations of  metabolites, line-broadening, and zero-order phase. The results show that the model is capable of reliably fitting in vivo data.

View Presentation Video

Dayeong An¹ and El-Sayed Ibrahim^1
1Biomedical Engineering, Medical College of Wisconsin, Milwaukee, WI, United States

Keywords: Machine Learning/Artificial Intelligence, Heart

Current gold-standard method for obtaining myocardial strain is based on MRI tagged images, although this increases the MRI exam time and requires special analysis software. We propose to use cine MR images to train a deep neural-network to generate myocardial strain based on target strain maps generated from tagged images acquired at the same locations and timepoints as the cine images. The results showed high agreement between the output and target strain maps. Our method not only saves MRI scan time by acquiring only cine images but also pre and post image processing time by quantifying myocardial strains automatically.

View Presentation Video

Simon Graf¹, Nora Küchler¹, Walter Wohlgemuth¹, and Andreas Deistung^1
1University Hospital Halle (Saale), Halle (Saale), Germany

Keywords: Machine Learning/Artificial Intelligence, Quantitative Susceptibility mapping

 Deploying deep learning models for quantitative susceptibility mapping is driven by optimizing hyper-parameters and using suitable architectures. We investigated the impact of activation functions on network model training. ELU-, leaky ReLU- and ReLU-models with 16 and 32 initial channels were tested for solving dipole inversion on synthetic susceptibility data. All models showed convergence after completing 100 training epochs. However, the 16-channel-ELU-model achieved low losses after only 20 training epochs and showed similar reconstruction performance to the 32-channel-ELU-model. Using the ELU activation allows the use of smaller network models resulting in fewer memory requirements and less training time.

Yuhui Xiong¹, Xiaocheng Wei¹, Jiankun Dai¹, Yang Fan¹, Jie Lu², and Bing Wu^1
1GE Healthcare MR Research, Beijing, China, ²Department of Radiology, Xuanwu Hospital, Capital Medical University, Beijing, China

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

This study aims to shorten the diffusion spectral imaging (DSI) scan time to a clinically acceptable level while providing satisfactory complex white matter fiber structure description as well as accurate diffusion metric quantification by applying deep learning-based reconstruction. Images were acquired using conventional (≈30 min) and rapid (≈8 min) DSI sequences, and reconstructed using conventional and DL-based methods, respectively. Atlas-based fiber-tracking and diffusion metrics quantification from various advanced models were conducted. The results demonstrated that the 8-minute rapid DSI sequence combined with DL-recon can provide complex fiber structure tractography and advanced diffusion metric quantification of satisfactory quality.

Yishuang Wang¹, Tianxiang Huang¹, Meining Chen², XU YAN², and Longlin Yin^1
1Radiology, Sichuan Provincial People's Hospital, Chengdu, China, ²MR Scientific Marketing, Siemens Healthcare, Shanghai, China

Keywords: Machine Learning/Artificial Intelligence, Liver

MR multiparametric quantitative techniques have an important role in liver-related diseases.We achieved simultaneous proton density fat fraction(PDFF), R2*and R1 quantification of different liver diseases using a simultaneous multi-relaxation-time Imaging(TXI) technique.We observed differences of PDFF,R2* and R1 mapping between healthy volunteers and patients with cirrhosis,HCC,metastatic carcinoma of liver,and liver transplantation.Rapid scanning increases usage of this method in the clinic.These quantitative values can be biomarkers for assessing liver function and liver tissue characterization.

View Presentation Video

Piyush Kumar Prajapati¹, Rakesh Kumar Gupta², and Anup Singh^1,3,4
1Centre for Biomedical Engineering, Indian Institute of Technology Delhi, New Delhi, India, ²Department of Radiology, Fortis Memorial Research Institute, Gurugram, India, ³Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi, New Delhi, India, ⁴Department of Biomedical Engineering, All India Institute of Medical Sciences, New Delhi, India

Keywords: Machine Learning/Artificial Intelligence, Brain

Tracer Kinetic (TK) parametric maps are obtained from Dynamic Contrast Enhanced (DCE) - MRI which aid in the detection and grading of brain tumors. Conventionally, TK maps are obtained using Non-Linear-Least-Square (NLLS) fitting approach, which is time consuming and data noise. In the current study, we implemented a deep learning framework whose backbone is attention networks to estimate TK parametric maps. Transfer learning was performed to extend work from synthetic data to high-grade glioma (HGG) patients’ DCE-MRI data to obtain better quality TK maps in lesser time.

View Presentation Video

Hadas Ben-Atya¹ and Moti Freiman^1
1Faculty of Biomedical Engineering, Technion, Haifa, Israel

Keywords: Machine Learning/Artificial Intelligence, Relaxometry

Recovery of T₂ distribution of tissue from MRI data acquired at multiple echo times has the potential to be used as a biomarker for the assessment of various pathologies, including stroke and epilepsy, investigation of neurodegenerative diseases, and tumor characterization. Current deep neural networks (DNN) for T₂ distribution recovery are highly sensitive to variations in the acquisition parameters such as different echo times. We present a new physically-driven DNN model that encodes the TE acquisition parameters as part of its architecture. Our model accurately recovers the T₂ distribution, regardless of variations in SNR and in the acquisition parameters.

View Presentation Video

Wonil Lee¹, Beomgu Kang¹, Jongyeon Lee¹, Georges El Fakhri², Chao Ma², Yeji Han³, Jun-Young Chung⁴, Young Noh⁵, and HyunWook Park^1
1The School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Yuseong-gu, Korea, Republic of, ²Massachusetts General Hospital, Boston, MA, United States, ³Department of Biomedical Engineering, Gachon University, Incheon, Korea, Republic of, ⁴Department of Neuroscience, College of Medicine, Gachon University, Incheon, Korea, Republic of, ⁵Department of Neurology, Gil Medical Center, Gachon University College of Medicine, Incheon, Korea, Republic of

Keywords: Machine Learning/Artificial Intelligence, Quantitative Imaging, Intravoxel incoherent motion

Recently, various methods have been proposed to quantify intravoxel incoherent motion parameters. Many studies have shown that quantification methods using deep learning can accurately estimate IVIM parameters. Unsupervised learning is useful when quantifying IVIM parameters for in-vivo data because it does not require label data. However, in some cases, loss function does not converge as iteration increases. Constraint functions can be used to solve these problems by limiting the range of estimated outputs. In this study, we investigated the effects of constraint function to limit the range of estimated output.

View Presentation Video

Kornelius Podranski¹, Kerrin J. Pine¹, Timoteo Colnaghi², Andreas Marek², Patrick Scheibe¹, Nico Scherf^1,3, and Nikolaus Weiskopf^1,4
1Neurophysics, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, ²Max Planck Computing and Data Facility, Garching (Munich), Germany, ³Neural Data Science and Statistical Computing, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, ⁴Felix Bloch Institute for Solid State Physics, Faculty of Physics and Earth Sciences, Leipzig University, Leipzig, Germany

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, Brain

Approaches for accelerating multi-echo gradient echo (ME-GRE) acquisitions as a basis for multi-parameter mapping (MPM) were explored. Fully sampled ME-GRE data were retrospectively undersampled to equispaced Cartesian, CAIPIRINHA and Poisson disc patterns. Echoes were jointly reconstructed with the iterative ENLIVE algorithm and the machine learning/artificial intelligence adapted DeepcomplexMRI (DCMRI) approach. The approaches result in comparable peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM), but show different types and different levels of artifacts. The DCMRI approach promises fast reconstruction and flexibility in the choice of undersampling patterns for ME-GRE imaging in the future.

View Presentation Video

Congcong Liu¹, Zhuo-Xu Cui¹, Yuanyuan Liu¹, Chentao Cao¹, Jing Cheng¹, Yanjie Zhu¹, Haifeng Wang¹, and Dong Liang^1,2
1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, ²Pazhou Lab, Guangzhou, China

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Traditional hand-craft designed methods to accelerated T1rho mapping have limited characterisation capabilities, while deep learning methods lack the interpretability. On the other hand, the joint distribution is the most direct and accurate way to characterize the correlation between different images. Therefore, we attempt to propose a joint distribution estimation method and use it to construct a T1$$$\rho$$$ reconstruction model. In particular, we use a score-based diffusion model to model the joint distribution of acquired T1rho-weighted images. Moreover, the corresponding reconstruction model is solved using the Langevin gradient descent method. Finally, numerical experiments validate the effectiveness of the proposed method.

Zhaonan Sun¹, Xiaoying Wang², and Kexin Wang^3
1Department of Radiology, Peking University First Hospital, Beijing, China, ²Department of Radiology, Peking University First Hospital, Beijing, China, ³School of Basic Medical Sciences, Capital Medical University, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Prostate

A total of 557 mpMRI data were retrospectively collected from three hospitals to build an external validation dataset, with 245 csPCa cases and 312 non-csPCa cases. The csPCa lesions were annotated based on pathology records by two experienced radiologists. AI algorithms were used to automatically detect and localize the suspicious csPCa areas on the T2WI and ADC maps. The metrics of sensitivity, specificity, and accuracy were used to evaluate the diagnostic efficacy of the AI algorithms at the lesion level, the sextant level, and the patient level.

View Presentation Video

Sukrit Sharma¹, Cornelius Cadrien², Philipp Lazen¹, Hangel Gilbert¹, Roxane Licandro³, Wolfgang Bogner¹, and Georg Widhalm^2
1High-field MR Center, Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna, Austria, Medical University of Vienna, Vienna, Austria, ²Medical University of Vienna, Vienna, Austria, ³Department of Biomedical Imaging and Image-guided Therapy, Computational Imaging Research Lab (CIR) and Laboratory for Computational Neuroimaging, A.A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital / Harvard Medical School, Charlestown, MA, US., Medical University of Vienna, Vienna, Austria

Keywords: Machine Learning/Artificial Intelligence, Brain, Glioma

To contribute to better tumour classification and thus enhancing patient outcomes, we statistically analysed metabolic maps of 37 glioma patients obtained using high resolution 7T MRSI. We tested and optimised different semi-supervised learning based classification approaches. Random forest classification of IDH mutation status and tumour grade in clinical imaging based segmented tumour regions yielded high diagnostic accuracy with AUC of 86% and 99% respectively. We found Glu, Gln, GSH, tCho, Ins, Gly and tCr as important determining features. These are similar to comparable SVS studies while providing the advantage of whole-brain coverage.

View Presentation Video

Patrick Kinz¹ and Lothar R Schad^1
1Computer Assisted Clinical Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany

Keywords: Machine Learning/Artificial Intelligence, Oxygenation

We developed a CNN for OEF mapping from QSM+qBOLD data, which utilizes utilizes 3D convolutional layers. Two dimensions for the spatial components of an image and one dimension for the temporal component in the qBOLD data. The results are an improvement over our previous 2D CNN, but even the more advanced network architecture struggles with voxels, that have a very low deoxyhemoglobin content. In this abstract we also study with simulated data when the CNN produces reliable results and when it predicts default values for the reconstructed parameters instead.