View Presentation Video

Haben Berhane¹, Anthony Maroun¹, Mahmoud Ebrahimkhani¹, Ulas Bagci¹, Bradley Allen¹, and Michael Markl^1
1Radiology, Northwestern University, Chicago, IL, United States

Keywords: Flow, Velocity & Flow, CFD

Aortic hemodynamic quantifications are vital for patient management. While 4D Flow MRI provides comprehensive aortic hemodynamics, it is hampered by long-acquisition times and cumbersome pre-processing. In this study, we developed an AI for the prediction of systolic 3D blood flow velocity vector fields with 3D aortic geometry as the only input. We performed testing on 248 BAV and 104 TAV datasets, using the systolic velocity vector fields from the 4D flow MRI as the ground-truth. Generally, we saw very strong agreement between the AI and the 4D flow and resilience to geometric changes (volume, dimension) in the input segmentation.

View Presentation Video

Naoyuki Takei¹, Rafi Brada², Sangtae Ahn³, Graeme McKinnon⁴, Xucheng Zhu⁵, Atsushi Nozaki¹, Shigeo Okuda⁶, Masahiro Jinzaki⁶, and Tetsuya Wakayama^1
1GE Healthcare, Tokyo, Japan, ²GE Research, Herzliya, Israel, ³GE Research, Niskayuna, NY, United States, ⁴GE Healthcare, Waukesha, WI, United States, ⁵GE Healthcare, Menlo Park, CA, United States, ⁶Keio University School of Medicine, Tokyo, Japan

Keywords: Vessels, Machine Learning/Artificial Intelligence

Rapid non-contrast MRA of supra-aortic arteries is necessary to select proper patient for endovascular therapy (EVT) as EVT has become the predominant therapy of acute ischemic stroke. However, conventional 3DTOF has long scan time of 6-7 minutes to cover the entire carotid artery. A highly accelerated MRA with iterative deep learning reconstruction was developed to provide the wide scan coverage and less motion sensitivity within 1 minute. We evaluated the technique with prospectively and retrospectively undersampled data demonstrating high-quality vessel visualization and improved acquisition efficiency at 8x acceleration.

View Presentation Video

Hideki Ota¹, Satoshi Higuchi¹, Yoshiaki Morita², Takashi Nishina³, Sho Tanaka³, Yuichi Yamashita³, Yoshimori Kassai³, Tasuo Nagasaka¹, Mitsue Miyazaki⁴, and Kei Takase^1
1Tohoku University Hospital, Sendai, Japan, ²National Cardiovascular Research Center, Osaka, Japan, ³Canon Medical Systems, Otawara, Japan, ⁴University of California, San Diego, San Diego, CA, United States

Keywords: Vessels, Vessels, workflow, deep learning reconstruction

Efficiency of 3D ky-kz centric fast field echo (FFE) acquisition was acquired using 100% efficiency of real-time motion correction in the sagittal oblique acquisition and compared with conventional 3D FFE acquisition. The thoracic aorta was acquired within 1 minute using centric ky-kz FFE and reconstructed with high-resolution deep learning reconstruction (HR-DLR), providing good image quality. Regular FFE with conventional reconstruction provides fair image quality with prominent noise and artifacts.

Yang Yang¹, songxiong wu², Bingsheng Huang³, Ping Zeng², Mingyu Wang⁴, and Zilong Huang^4
1Department of Radiology, Suining Central Hospital, Suining, China, ²Radiology Department, Shenzhen University General Hospital and Shenzhen University Clinical Medical Academy, shenzhen, China, ³Medical AI Lab, School of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China, ⁴Medical AI Lab, School of Biomedical Engineering, Health Science Center, Shenzhen University, shenzhen, China

Keywords: Vessels, Blood, brain blood vessel segmentation,Multi-Parametric,Multi-view

Accurate segmentation of blood vessels allow neurosurgical navigation and can help neurosurgeons accurate surgical and treatment plans. However, traditional blood vessel segmentation methods based on thresholds have limited performance. To solve problem, we proposed a cascaded DL network (MVPC-Net) that combines three refinements: multi-view learning, multi-parameter input, and a multi-view ensemble module-based strategy. The results of ablation experiments showed that, by adding all the refinements proposed, the performance of the baseline model improved from Dice similarity coefficient 0.865 to 0.922. Thus, our method can provide better segmentation of the brain, and scalp blood vessels and has potential for clinical application.

View Presentation Video

Brock W Jolicoeur¹, Leonardo A Rivera-Rivera¹, Grant S Roberts¹, Laura Eisenmenger², and Kevin M Johnson^1
1Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ²Radiology, University of Wisconsin-Madison, Madison, WI, United States

Keywords: Flow, Blood vessels, Machine Learning/Artificial Intelligence

Through a self-supervised deep learning denoising algorithm, more precise cerebrovascular measurements of flow, maximum fluid velocity, and vessel cross-sectional area were obtained from undersampled neurovascular 4D-Flow MRI data. This algorithm was applied to back-to-back scans to illustrate its effectiveness in reducing measurement variance.

View Presentation Video

Zakarya BENTATOU¹, Timothé BOUTELIER¹, Anais BERNARD¹, and Henitsoa RASOANANDRIANINA^1
1OLEA MEDICAL, La Ciotat, France

Keywords: Stroke, Machine Learning/Artificial Intelligence, Large vessel occlusion, TOF, MR angiography

Large vessel occlusion (LVO) in stroke patients is mostly detected using deep-learning-based automated methods on CT angiography but there has been no report on such methods using time-of-flight magnetic resonance angiography (TOF-MRA). Our study includes 460 stroke patients with 230 LVO-positive cases. The first step was vessel segmentation, and the output mask was used for the LVO detection. Both steps were deep-learning based using TOF-MRA. Our model successfully detected 95% LVO-positive and 92% LVO-negative patients. The high detection rate and short processing time (< 60 seconds) suggested that our model is highly adequate in a clinical emergency context.

View Presentation Video

Philipp Boehm-Sturm^1,2,3, Felix Knab^1,2, Stefan Paul Koch^1,2,3, Sebastian Major^1,2, Tracy D. Farr^1,2,3,4, Susanne Mueller^1,2,3, Philipp Euskirchen¹, Moritz Eggers^1,2, Melanie Kuffner^1,2, Josefine Walter^1,2,5, Jens P. Dreier^1,2,6,7, Matthias Endres^1,2,6,8,9,10, Ulrich Dirnagl^1,2,6,8,9,10, Nikolaus Wenger^1,2,6,11, Christian J. Hoffmann^1,2,11, and Christoph Harms^1,2,6,8,9
1Klinik und Hochschulambulanz für Neurologie, Department of Experimental Neurology, Charité-Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin, Berlin, Germany, ²Center for Stroke Research Berlin, Charité-Universitätsmedizin Berlin, Berlin, Germany, ³NeuroCure Cluster of Excellence and Charité Core Facility 7T Experimental MRIs, Charité-Universitätsmedizin Berlin, Berlin, Germany, ⁴School of Life Sciences, University of Nottingham, Berlin, Germany, ⁵QUEST Center for Transforming Biomedical Research, Berlin Institute of Health at Charité – Universitätsmedizin Berlin, Berlin, Germany, ⁶Einstein Center for Neuroscience, Berlin, Germany, ⁷Bernstein Center for Computational Neuroscience, Berlin, Germany, ⁸German Center for Cardiovascular Research (DZHK), partner site Berlin, Berlin, Germany, ⁹NeuroCure Clinical Research Center, Charité-Universitätsmedizin Berlin, Berlin, Germany, ¹⁰German Center for Neurodegenerative Diseases (DZNE), Berlin, Germany, ¹¹Berlin Institute of Health (BIH), Berlin, Germany

Keywords: Stroke, Animals, Machine learning, Mice, MCAO

Prediction of motor-functional outcome in mice could potentially guide researchers in their treatment decisions in preclinical stroke intervention studies and support outcome-dependent stratifications. We pooled 13 studies in which mice underwent identical MRI and behavioral testing protocols. In this large cohort of mice (n=148), we developed and compared machine learning-based predictors of post-stroke recovery.

View Presentation Video

Juan Bai¹, Jiang Wu¹, Ning Li¹, Xuan Li¹, and Kaiyu Wang^2
1Shanxi Cardiovascular Hospital, Taiyuan, China, ²MR Research China,GE Healthcare, Beijing, China

Keywords: Blood vessels, Atherosclerosis, Radiomics；Machine Learning

The stability of  plaques  is the key cause of cerebral ischemic in patients with basilar atherosclerosis. High-resolution vessel wall imaging (HR-VWI) is  a useful technique  for  study of plaques. Radiomics are widely used in  recently years  to discover the radiographic information of diseases and to improve diagnostic performance. In this study, we constructed six  machine learning models based on radiomics features  from  HR-VWI to predict the stability of plaques of the basilar artery, finding support vector machine has the best performance.

View Presentation Video

Di Wu¹, Mengzhou Sun², Yi Li³, Xiaoyun Liang³, Feng Huang³, and Wenzhen Zhu^1
1Radiology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, ²Neusoft Medical Systems Co. Ltd, Shenyang, Liaoning, China, Beijing, China, ³Neusoft Medical Systems Co. Ltd, Shenyang, Liaoning, China, Shanghai, China

Keywords: Stroke, Machine Learning/Artificial Intelligence

Large vessel occlusion detection based on clinical scales is of low sensitivity and that based on CTA needs contrast agent exposure. This study aims to develop a deep learning (DL) algorithm for detecting intracranial large vessel steno-occlusion on contrast agent-free MR techniques including DWI and ASL. The accuracy of the DL algorithm was 88.2% with a sensitivity of 88.0%, comparable to CTA-based DL algorithms with sensitivity ranging from 67% to 94%. The MR-based DL algorithm is feasible to accurately detect intracranial large vessel steno-occlusion without intervention, radiation exposure and contrast agent, which could optimize stroke workflow and guide clinical decision-making.

View Presentation Video

Yuya Hirano¹, Noriyuki Fujima², Kinya Ishizaka¹, Takuya Aoike¹, Jihun Kwon³, Masami Yoneyama³, and Kohsuke Kudo^4,5
1Department of Radiological Technology, Hokkaido University Hospital, Sappro, Japan, ²Department of Diagnostic and Interventional Radiology, Hokkaido University Hospital, Sapporo, Japan, ³Philips Japan, Ltd, Tokyo, Japan, ⁴Department of Diagnostic Imaging, Hokkaido University Graduate School of Medicine, Sapporo, Japan, ⁵Global Center for Biomedical Science and Engineering, Faculty of Medicine, Hokkaido University, Sappro, Japan

Keywords: Blood vessels, Image Reconstruction

SmartSpeed-AI is recently introduced as a physics driven type deep learning-based novel image reconstruction technique. We investigated the utility of SmartSpeed-AI for the better visibility of lenticulostriate artery (LSA) in high spatial resolution TOF-MRA by comparing the compressed-sensing sensitivity-encoding (compressed SENSE) algorithm. Both quantitative and qualitative assessment revealed that the visibility of LSAs were significantly higher under the SmartSpeed-AI reconstruction than compressed SENSE. SmartSpeed-AI can be helpful to provide superb image quality for the depiction of small arteries like LSA in high resolution TOF-MRA.

View Presentation Video

Xiaoqing Alice Zhou¹, David Hike¹, Bei Zhang¹, Zeping Xie¹, Xiaochen Liu¹, Matthew S Rosen¹, Juan Eugenio Iglesias^1,2,3, and Xin Yu*^1
1Massachusetts General Hospital, Boston, MA, United States, ²University College London, London, United Kingdom, ³Massachusetts Institute of Technology, Boston, MA, United States

Keywords: Blood vessels, Alzheimer's Disease

High-resolution cerebrovascular mapping is crucial for the understanding of brain function. While the existing invasive imaging modalities map brain vasculature at the microscopic scale, they are mostly restricted to the cortical surface in small animals. The non-invasive penetrating cerebrovascular mapping for larger animals remains the privilege of MRI. To achieve high spatial resolution with MRI, we trained 3D U-net to super-resolve images from 75µm to 37.5µm isotropic resolution. Our results show a significantly improved segmented micro-vessel network and offer a powerful tool to identify the anatomical distribution of the microvessels in normal and degenerative brains longitudinally.

View Presentation Video

Chenwei Tang¹, Leonardo A Rivera-Rivera^1,2, Laura B Eisenmenger³, and Kevin M Johnson^1,3
1Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ²Department of Medicine, University of Wisconsin-Madison, Madison, WI, United States, ³Department of Radiology, University of Wisconsin-Madison, Madison, WI, United States

Keywords: Blood vessels, Velocity & Flow

Keywords: Wave Encoding, Trajectory Optimization, 4D Flow

Non-Cartesian sampling is often required for 4D Flow imaging because of more efficient sampling. Due to the heuristic nature of the optimization of such trajectories, we propose to parameterize and optimize wave encoded 3D Cartesian sampling using a gradient descent algorithm in a data-driven way. We demonstrate the feasibility of our framework in learning the sampling patterns and the wave parameters and providing high image quality for highly accelerated scans in digital phantoms, phantoms and in vivo with phase contrast.

View Presentation Video

Ingrid Digernes¹, Martin Soria Røvang¹, Terje Nome², Cecilie Nome³, Thor Håkon Skattør², Brian Anthony Enriquez³, Bradley J MacIntosh¹, Anne Hege Aamodt³, and Atle Bjørnerud^1
1Computational Radiology and Artificial Intelligence, Oslo University Hospital, Oslo, Norway, ²Radiology and Nuclear Medicine, Oslo University Hospital, Oslo, Norway, ³Department of Neurology, Oslo University Hospital, Oslo, Norway

Keywords: Stroke, Machine Learning/Artificial Intelligence, segmentation, functional outcome, thrombectomy

Using a standard 3D nnUNET model that was pretrained to segment WMH on FLAIR, we obtained good stroke lesion segmentation accuracy (median Dice = 0.80) when the model was re-trained to segment the stroke lesion on DWI in a smaller stroke sample (N = 82). Applying the segmentation model to DWIs from 307 thrombectomy patients, we found no association between pre-treatment lesion volume and functional outcome at 90 days after thrombectomy. Our results indicate that the established assumption of lesion size being a strong predictor of functional outcome should be investigated further for patients receiving mechanical thrombectomy.

View Presentation Video

Frederick C. Damen¹, Changliang Su², Thomas Anderson¹, Tibor Valyi-Nagy³, Leon Tai⁴, Rifeng Jiang⁵, and Kejia Cai^1,6
1Radiology, University of Illinois at Chicago, Chicago, IL, United States, ²Medical Imaging, Sun Yat-sen University Cancer Center, Guangzhou, China, ³Pathology, University of Illinois at Chicago, Chicago, IL, United States, ⁴Anatomy and Cell Biology, University of Illinois at Chicago, Chicago, IL, United States, ⁵Radiology, Fujian Medical University Union Hospital, Fuzhou, China, ⁶Bioengineering, University of Illinois at Chicago, Chicago, IL, United States

Keywords: Stroke, Diffusion/other diffusion imaging techniques

The ischemic stroke cascade is a complex menagerie of processes causing disruption of neuronal function and willful and/or reactive breakdown of the neural vascular unit (NVU), and, prevails to recovery, repair, or patient demise. To better understand the underpinnings of the stroke cascade we applied our Multimodal Apparent Diffusion (MAD) method to multi b‑value diffusion weighted magnetic resonance imaging, up to b‑value of 10K s/mm², on 41 consecutive stroke patients. Using Fuzzy C Means we were able to discern 13 normal appearing tissue types and 16 lesion types. There are several findings that should be contemplated in current clinical imaging.

Na Han¹, Wanjun Hu¹, Yurong Ma¹, Laiyang Ma¹, Yu Zheng¹, Jing Zhang¹, and Kai Ai^2
1Department of Magnetic Resonance, Lanzhou University Second Hospital, Lanzhou, China, ²Philips Healthcare, Xi’an, China

Keywords: Stroke, Radiomics, Carotid Plaque, Vessel Wall High-Resolution MRI

The study of the relationship between carotid atherosclerotic plaque and ischemic stroke must go beyond the assessment of the basic imaging characteristics of plaque or the degree of vascular stenosis, and should adopt a new model which carotid plaque imaging combined with artificial intelligence. This study established a radiomics model of carotid plaque based on vessel wall high-resolution MRI (VW-HRMRI) to accurately predict the risk of ischemic stroke. Our study showed that the radiomics_T1CE model and combined_T1CE model outperformed other models, whereas the combined model showed no significant difference with the radiomics model.

View Presentation Video

Siying Xu¹, Marcel Früh¹, Kerstin Hammernik^2,3, Sergios Gatidis^1,4, and Thomas Küstner^1
1Medical Image and Data Analysis (MIDAS.lab), Department of Diagnostic and Interventional Radiology, University Hospital of Tuebingen, Tuebingen, Germany, ²Lab for AI in Medcine, Technical University of Munich, Munich, Germany, ³Department of Computing, Imperial College London, London, United Kingdom, ⁴Max Planck Institute for Intelligent Systems, Tuebingen, Germany

Keywords: Image Reconstruction, Machine Learning/Artificial Intelligence, Self-Supervised Learning, Image Reconstruction, Heart

Existing methods for deep-learning reconstruction require abundant fully-sampled images as labels, which is challenging or impractical in practice. To address this issue, we propose a self-supervised learning (SSL) approach that can be trained on subsampled data, avoiding the need for fully-sampled datasets. Specifically, we pre-train a feature extractor by contrastive learning as the first step. In the second step, the pre-trained feature extractor assists the self-supervised network during reconstruction by feature embedding. Results indicate that the proposed SSL method can effectively reconstruct cardiac CINE images without fully-sampled data. It outperforms existing SSL networks and shows comparable results to supervised learning.

View Presentation Video

Changyu Sun^1,2, Kenneth Bilchick³, Michael Salerno⁴, and Talissa A. Altes^2
1Biomedical, Biological and Chemical Engineering, University of Missouri Columbia, Columbia, MO, United States, ²Radiology, University of Missouri Columbia, Columbia, MO, United States, ³Department of Medicine, University of Virginia Health System, Charlottesville, VA, United States, ⁴Department of Medicine, Stanford University, Stanford, CA, United States

Keywords: Image Reconstruction, Heart, Self-supervised Learning

Physics-guided self-supervised learning (PG-SSL) of MRI reconstruction may provide high spatiotemporal fidelity and fast reconstruction of highly accelerated first-pass myocardial perfusion MRI without ground truth. We sought to develop a SSL model with self-supervised regularization (SSR) using Siamese architecture with stop gradient and re-undersampling block to generate physics-based data augmentation and regularization. PG unrolled network was used as the sub-network in Siamese structure. Self-supervised learning with self-supervised regularization (SSLR) outperformed low rank plus sparse on retrospective rate-8 undersampling single-band data and showed improved SNR and temporal fidelity on prospective multiband whole-heart coverage high resolution perfusion imaging.

View Presentation Video

Pengfang Qian^1,2, Junwei Yang³, Zijian Zhou^1,2, Peng Hu^1,2, and Haikun Qi^1,2
1School of Biomedical Engineering, ShanghaiTech University, Shanghai, China, ²Shanghai Clinical Research and Trial Center, Shanghai, China, ³Department of Computer Science and Technology, University of Cambridge, Cambridge, United Kingdom

Keywords: Image Reconstruction, Machine Learning/Artificial Intelligence

Various deep learning methods have been proposed for cardiac cine MRI, including accelerated image reconstruction, cardiac motion estimation and segmentation, which are traditionally considered as separate tasks without exploiting the inter-task correlation. In this study, we propose a unified deep learning model to perform accelerated cine image reconstruction, motion estimation and segmentation simultaneously in an iterative framework, where correlations between tasks are exploited by compensating motion in reconstruction, semi-supervising segmentation using pseudo-labels generated by motion and improving motion estimation using intermediately reconstructed images. Experiment results show that the multi-task model outperformed single-task networks.

View Presentation Video

Hae Sol Moon¹, Lindsay Heffron², Ali Mahzarnia³, Barnabas Obeng-Gyasi³, Matthew Holbrook³, Cristian T. Badea^1,3, Wuwei Feng⁴, and Alexandra Badea^1,3,4,5
1Biomedical Engineering, Duke University, Durham, NC, United States, ²Orthopaedic Surgery, Duke University School of Medicine, Durham, NC, United States, ³Radiology, Duke University School of Medicine, Durham, NC, United States, ⁴Neurology, Duke University School of Medicine, Durham, NC, United States, ⁵Brain Imaging and Analysis Center, Duke University School of Medicine, Durham, NC, United States

Keywords: Multimodal, Data Analysis, Deep learning; Segmentation; Lesion Load; Stroke

We compared the ability of 2D and 3D U-Net Convolutional Neural Network (CNN) architectures to segment ischemic stroke lesions and predict patient outcome using single-contrast (DWI) and dual-contrast images (T2w FLAIR and DWI). The predicted lesion segmentation metrics and location relative to corticospinal tract correlated with post-stroke patient outcome measured by National Institutes of Health Stroke Scale (NIHSS). The 2D multi-modal CNN achieved the best results with mean Dice of 0.74. The highest correlation was for weighted-lesion load with both baseline and 90-days NIHSS (80%, p<0.001). Our results support that multi-contrast MR helps automate lesion segmentation and predict post-stroke outcomes.

View Presentation Video

Xiaoxu Yang¹, Pengxin Yu², Haoyue Zhang², Rongguo Zhang², Yuehong Liu¹, Xiuqin Jia¹, Penghui Sun¹, Xin Liu¹, Xunming Ji³, Qi Yang¹, and Chen Zhang^4
1Beijing Chaoyang Hospital, Beijing, China, ²Infervision Medical Technology Co., Ltd, Beijing, China, Beijing, China, ³Beijing Xuanwu Hospital, Beijing, China, ⁴MR Scientific Marketing, Siemens Healthineers, Beijing, China, Beijing, China

Keywords: Stroke, Brain

A deep learning algorithm for detecting cerebral venous thrombosis using routine brain MRI achieved higher patient-level sensitivity than radiologists and reduced the number of overlooked thrombosed segments. The proposed deep learning (DL) algorithm achieved area under the receiver operating characteristic curve of 0.96 for detecting patient with cerebral venous thrombosis. The sensitivity of DL algorithm was higher than that of radiologists and obtained high specificity on patient-level.Compared to radiologists, DL algorithm found more thrombosed segments, indicating greater sensitivity and a sufficient specificity at the segment-level.