View Presentation Video

Zhengyang Ming^1,2, Arutyun Pogosyan³, Anthony G. Christodoulou^4,5, J. Paul Finn^1,2, Dan Ruan^1,4,6, and Kim-Lien Nguyen^1,2,3,4
1Physics and Biology in Medicine Graduate Program, University of California,Los Angeles, Los Angeles, CA, United States, ²Department of Radiological Sciences, David Geffen School of Medicine at UCLA, Los Angeles, CA, United States, ³Division of Cardiology, David Geffen School of Medicine at UCLA and VA Greater Los Angeles Healthcare System, Los Angeles, CA, United States, ⁴Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States, ⁵Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ⁶Department of Radiation Oncology, David Geffen School of Medicine at UCLA, Los Angeles, CA, United States

Keywords: Arrhythmia, Machine Learning/Artificial Intelligence

Both ECG-gating and self-gated approaches are used for cardiac motion binning in segmented cardiac cine MRI. Typical self-gated methods reduce ECG-dependency by assuming periodic cardiac motion, but may be less reliable in the presence of irregular cardiac motion. We propose a novel clustering algorithm that incorporates regularization in both temporal and cluster dimension to provide robustness against irregular cardiac motion in complex arrhythmias such as atrial fibrillation and premature ventricular contraction. Compared with images from k-means clustering, initial validation using the modified algorithm shows higher image quality scores with comparable single-to-noise ratio (SNR) and image sharpness. 

View Presentation Video

Maroua Mehri^1,2, Pierre Aublin³, Guillaume Calmon¹, Freddy Odille^1,3,4, and Julien Oster^3,4
1Epsidy, Nancy, France, ²University of Sousse, National School of Engineers, LATIS-Laboratory of Advanced Technology and Intelligent Systems, Sousse, Tunisia, ³IADI, INSERM U1254 and Université de Lorraine, Nancy, France, ⁴CIC-IT 1433, INSERM, Université de Lorraine and CHRU Nancy, Nancy, France

Keywords: Heart, Data Analysis, Machine learning/Artificial intelligence

Detecting R-peaks in ECG using deep learning (DL) requires large, annotated datasets. Such datasets with the MRI-specific magneto-hydrodynamic (MHD) effect, do not exist currently. We propose a robust data augmentation framework using records available online, adding realistic MHD artifacts to augment the training dataset. MHD artifacts were modelled from eight 3T MRI-ECG records, and added to 75 non-MRI-ECG records. The R-peak detection was evaluated on six 3T MRI-ECG records. Compared to a DL model trained without data augmentation, the number of false positives and missed detections were reduced by 57.6% and 16.4%, the overall error was decreased by 25%.

Hui Xue¹, Ahsan Javed¹, Rajiv Ramasawmy¹, Azaan Rehman¹, Peter Kellman¹, and Adrienne E Campbell-Washburn^1
1National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD, United States

Keywords: Heart, Machine Learning/Artificial Intelligence

Low field MRI systems are promising to increase the accessibility of cardiac MRI, at the expense of lower SNR. Here, we present the application of a g-factor-savvy denoising for cine imaging at 0.55T to increase useable acceleration rate. We used a model that combines convolutional neural networks and transformer model. The denoising network uses complex 2D+time images in SNR-units and g-factor maps as inputs and was trained with 3T data.  The percent mean myocardial SNR gain at 0.55T across 9 healthy volunteers was 97±31% (R=2) and 122±22% (R=3), with no indication of overt temporal or spatial smoothing.

View Presentation Video

Tina Yao^1,2, Nicole St. Clair³, John Gold³, Gabriel Miller³, David Schidlow³, Sunil Ghelani³, Rahul Rathod³, Jennifer Steeden¹, and Vivek Muthurangu^1
1Institute of Cardiovascular Science, University College London, London, United Kingdom, ²Institute of Health Informatics, University College London, London, United Kingdom, ³Department of Cardiology, Boston Children's Hospital, Boston, MA, United States

Keywords: Heart, Machine Learning/Artificial Intelligence

We have created an end-to-end machine learning pipeline that takes cardiac magnetic resonance scans straight from a registry of single ventricle patients, performs image classification, calculates bounding boxes, and segments the ventricles. The clinical utility of the pipeline is that there is very little human preprocessing required from the clinicians. The pipeline has great robustness as it is trained on multicenter data from different countries, with different scanners, image sizes and aspect ratios, patient ages (relating to heart sizes), and the inherent variability of single ventricle patients. Heart metrics calculated from our pipeline can guide treatment for single ventricle patients.

View Presentation Video

Gastao Cruz¹, Yuchi Liu¹, Evan Cummings¹, Jesse Hamilton¹, Vikas Gulani¹, and Nicole Seiberlich^1
1Department of Radiology, University of Michigan, Ann Arbor, MI, United States

Keywords: Heart, MR Fingerprinting

In this work, T1/T2/PDFF mapping with rosette cardiac MRF is improved by exploiting the idea of virtual coils along with subspace constrained reconstruction. The Hermitian symmetry of k-space is explicitly included in the forward model via virtual coils and combined with global low-rank models. This model is further combined with a regularizer leveraging prior information from dictionaries, patch-similarity, and locally low-rank. This virtual-coil + low-rank + patch-based approach is used to jointly reconstruct multi-echo rosette MRF data for improved quality of the final T1, T2, and PDFF maps. 

View Presentation Video

Yunling Li¹, Kai Yang², Xiuyu Chen², Jianxiu Lian³, Yong Sun¹, and Shihua Zhao^2
1Department of Cardiology, The Second Affiliated Hospital of Harbin Medical University, Harbin, China, HarBin, China, ²Department of Magnetic Resonance Imaging, Fuwai Hospital, State Key Laboratory of Cardiovascular Disease, National Center for Cardiovascular Diseases, Beijing, China, ³Beihang University, Beijing, China

Keywords: Heart, Cardiovascular, Acute Cardiac Allograft Rejection

Acute cardiac allograft rejection is a major cause of morbidity and mortality after heart transplantation. Reliable non-invasive diagnostic techniques for acute cardiac allograft rejection (ACAR) are unavailable. Cardiac magnetic resonance feature tracking can detect global functional changes in the longitudinal, radial, and circumferential directions during the early stages of graft rejection. We found that cardiac magnetic resonance feature tracking (CMR-FT)-derived global longitudinal strain (GLS), global circumferential strain (GCS), and global radial strain (GRS) are safe, noninvasive, and gadolinium contrast-free parameters for ACAR diagnosis. CMR-FT may provide a method that can reduce the dependency on invasive EMBs after heart transplantation.

View Presentation Video

Satonori Tsuneta¹, Satoru Aono², Rina Kimura¹, Jihun Kwon³, Takuya Aoike², Masami Yoneyama³, Kinya Ishizaka², Noriyuki Fujima¹, and Kohsuke Kudo^4
1Department of Diagnostic and Interventional Radiology, Hokkaido University Hospital, Sapporo, Japan, ²Department of Radiological Technology, Hokkaido University Hospital, Sapporo, Japan, ³Philips Japan, Ltd., Tokyo, Japan, ⁴Department of Diagnostic Imaging, Hokkaido University Graduate School of Medicine; Global Center for Biomedical Science and Engineering, Faculty of Medicine, Hokkaido University, Sapporo, Japan

Keywords: Myocardium, Data Processing

Cine cardiac magnetic resonance (CMR) imaging is an optimal cardiac volumetric analysis method because of its high contrast resolution. However, its spatial resolution is limited owing to prolonged scanning and breath-holding. Although compressed sensing–sensitivity encoding (Compressed SENSE; CS) and its deep-learning-based advancement (SmartSpeed AI; SSAI) can reduce the scan time, the spatial resolutions remain unchanged. Herein, we investigated the effect of a deep-learning-based super-resolution technique (SmartSpeed Precise Image; SSPI) on the cine CMR visual image quality in comparison with CS and SSAI with conventional zero-filling interpolation; resultantly, the SSPI significantly improved the visual image quality scores.

Dominik Daniel Gabbert¹, Lennart Petersen^1,2, Abigail Burleigh¹, Simona Boroni Grazioli¹, Sylvia Krupickova³, Reinhard Koch², Anselm Sebastian Uebing¹, Monty Santarossa², and Inga Voges^1
1Department of Congenital Heart Disease and Pediatric Cardiology, University Hospital Schleswig-Holstein, DZHK (German Center for Cardiovascular Research), partner site Hamburg/Kiel/Lübeck, Kiel, Germany, ²Multimedia Information Processing Group, Kiel University, Kiel, Germany, ³Departments of CMR and Paediatric Cardiology, Royal Brompton Hospital, London, United Kingdom

Keywords: Heart, Machine Learning/Artificial Intelligence, Congenital Heart Disease

We present a new method for detection of hypoplastic left heart syndrome (HLHS) based on the spatial arrangement of 7 distinctive anatomical landmarks in CMR images. The method was applied to the axial SSFP CMR scans of 46 patients with HLHS and 33 healthy controls. A tailor-made U-net-like deep convolutional network (CNN) with a shared 3D-convolutional encoder backbone and 7 segmentation heads was used for prediction of landmarks. Classification based exclusively on the coordinates of the detected landmarks had an accuracy of 98.7%. In future studies, the method may be applied to HLHS subgroups or other cardiac diseases.

View Presentation Video

Yu Ding¹, Yingmin Liu¹, Chong Chen¹, Ning Jin², Rizwan Ahmad¹, and Orlando Simonetti^1
1The Ohio State University, Columbus, OH, United States, ²Siemens Healthineers, Columbus, OH, United States

Keywords: Myocardium, Low-Field MRI, Late Gadolinium Enhancement

Low-field cardiac MR imaging has intrinsically low SNR that degrades the image quality, especially for LGE imaging. Using multiple acquisitions and motion-corrected averaging is a well-established method to improve SNR. However, the ineffectiveness of MOCO at low field leads to blurring. We proposed a novel method that integrates motion correction into compressed sensing and reconstructs a single motion-corrected image.  Quantitative sharpness measurements demonstrate that the proposed method improves boundary sharpness, independent of differences in SNR.

View Presentation Video

Marc Vornehm^1,2, Jens Wetzl², Daniel Giese^2,3, Jianing Pang⁴, Rizwan Ahmad⁵, and Florian Knoll^1
1Computational Imaging Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, ²Magnetic Resonance, Siemens Healthcare GmbH, Erlangen, Germany, ³Institute of Radiology, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, ⁴Siemens Medical Solutions USA Inc., Chicago, IL, United States, ⁵Biomedical Engineering, The Ohio State University, Columbus, OH, United States

Keywords: Heart, Low-Field MRI

Acceleration of cardiac cine MRI is highly desirable in order to decrease the required breath-hold duration. On low-field MRI systems in particular, this could help make cardiac MRI more widely available. We present a method based on the Variational Network for reconstruction of cardiac cine MRI, trained on data from 1.5 and 3T systems. Reconstructions of retrospectively and prospectively undersampled acquisitions at 0.55T with an acceleration rate of eight are shown and compared to Compressed Sensing reconstructions. Despite the domain shift to low-field data, the neural network achieved an SSIM of 94.6%, which is comparable to the Compressed Sensing results.

View Presentation Video

Karl Ludger Radke¹, Janina Hußmann^1,2, Lena Röwer^1,2, Dirk Voit³, Jens Frahm³, Gerald Antoch¹, Dirk Klee¹, Frank Pillekamp^1,2, and Hans-Jörg Wittsack^1
1University Düsseldorf, Department of Diagnostic and Interventional Radiology, Düsseldorf, Germany, ²Department of General Pediatrics, Neonatology and Pediatric Cardiology, University Children's Hospital Düsseldorf, D-40225 Düsseldorf, Germany, Düsseldorf, Germany, ³Biomedical NMR, Max Planck Institute for Multidisciplinary Sciences, D-37070 Göttingen, Germany, Göttingen, Germany

Keywords: Heart, Software Tools

In this work, we propose a new open-source framework for cardiac functional analysis. Our proposed framework can be used in a pipeline with the commercial software Circle cvi42 as well as the free software ITKSnap and enables a step towards a fully automated, reproducible, and quantitative assessment of MR images in clinical routine.

View Presentation Video

Pedro Louro Costa Osório¹, Markus Henningsson^2,3,4, Alberto Gomez Herrero^5,6, Rita G. Nunes¹, and Teresa M. Correia^6,7
1Institute for Systems and Robotics - Lisboa and Department of Bioengineering, Instituto Superior Técnico - Universidade de Lisboa, Lisbon, Portugal, ²Division of Cardiovascular Medicine, Department of Medical and Health Sciences, Linköping University, Linköping, Sweden, ³Centre for Medical Image Science and Visualization (CMIV), Linköping University, Linköping, Sweden, ⁴MR Physics, Perspectum Ltd, Oxford, United Kingdom, ⁵Ultromics Ltd, Oxford, United Kingdom, ⁶School of Biomedical Engineering Imaging Sciences, King’s College London, London, United Kingdom, ⁷Centre for Marine Sciences - CCMAR, Faro, Portugal

Keywords: Heart, Machine Learning/Artificial Intelligence, View Planning

Cardiac Magnetic Resonance (CMR) is a powerful technique which can be used to perform a comprehensive cardiac examination. However, its adoption is often limited to specialised centres, in part due to the need for highly trained operators to perform the complex procedures of determining the 4 standard cardiac planes: 2-, 3-, 4-chamber and short axis views. To automate view planning, a deep learning-based tool (DeepCardioPlanner) has been proposed to regress the view defining vectors from a rapidly acquired 3D image. It successfully takes advantage of multi-objective learning to allow accurate, fast and reproducible view prescriptions without any operator input.

View Presentation Video

Lingping Ran¹, Lu Huang¹, Xianghu Yan¹, Yun Zhao¹, and Liming Xia^1
1Department of Radiology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China

Keywords: Heart, Heart

Routine cardiac functional assessment with conventional balanced steady state free precession (bSSFP) cine sequence generally requires multiple breath-holding (BH) and long examination time. The purpose of this study was to evaluate the image quality and bi-ventricular functional analysis using a novel free-breathing (FB) artificial intelligence (AI) cine method with motion corrected (MOCO) in a clinical context. Compared to conventional multiple BHs cine, the proposed FB-MOCO AI cine method achieved comparable image quality with shortened scan times. No significant difference in ventricular volumetric and functional parameters were found, suggesting its potential applicability for clinical applications.

View Presentation Video

Habib Rebbah¹, Guillaume Gautier¹, and Timothé Boutelier^1
1Research & Innovation, Olea Medical, La Ciotat, France

Keywords: Heart, Segmentation

In the case of U-Net based segmentation algorithms, we propose to explore a common question of the medical image research teams: using already existing algorithm or train a new one specific to the target data.  We compared for the cardiac CINE case three approaches: using an algorithm trained with external data, train a new one with the taget data from scratch, and fin-tune the first one with the target data.

View Presentation Video

Yajing Zhang¹, Guohui Ruan¹, Tianyu Han², Christiane Kuhl³, Masami Yoneyama⁴, Daniel Truhn³, and Shuo Zhang^3,5
1MR Clinical Science, Philips Health Technology, Suzhou, China, ²RWTH University Aachen, Aachen, Germany, ³Diagnostic and Interventional Radiology, University Hospital RWTH Aachen, Aachen, Germany, ⁴Philips Japan, Tokyo, Japan, ⁵Philips Market DACH, Hamburg, Germany

Keywords: Myocardium, Cardiovascular

Ultrafast imaging with high acceleration in cardiac MRI is of great clinical interest, but so far often results in inferior image quality that prevents its use in routine diagnosis. In this work, we aim to establish an unsupervised deep learning neural network based on vector quantized variational autoencoder for noise reduction and image quality improvement. Initial results on both public and clinical data promise a new approach to the existing methods. Further investigations with focus on its effectiveness of performance in real world applications are warranted.

View Presentation Video

Haruno Ito¹, Masaki Ishida¹, Masafumi Takafuji¹, Shinichi Takase¹, Yoshiaki Komori², Davide Piccini^3,4, Jessica A.M. Bastiaansen⁴, Jérôme Yerly^4,5, Matthias Stuber^4,5, and Hajime Sakuma^1
1Radiology, Mie University Hospital, Tsu, Mie, Japan, ²Siemens Healthcare K.K., Tokyo, Japan, ³Advanced clinical imaging technology, Siemens Healthineers International AG, Lausanne, Switzerland, ⁴Department of Diagnostic and Interventional Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ⁵CIBM Center for Biomedical Imaging, Lausanne, Switzerland

Keywords: Heart, Heart, Cine MRI

Cine MRI derived from 3T contrast-enhanced free-running 5D whole-heart coronary MRA (free-running cine MRI) was compared with standard 2D cine MRI in 30 patients with suspected CAD. In a selected patient, temporal width was optimized to 50 ms. Then, in the remaining 29 patients, free-running cine MRI provided good agreement in LV volume and function quantification with standard 2D cine MRI with a good inter-observer reproducibility. The results suggest that free-running cine MRI allows for a shorter scan protocol by skipping the standard 2D cine MRI.

View Presentation Video

Xucheng Zhu¹, Suryanarayanan Kaushik², Frandics Chan³, Melany Atkins⁴, Prashant Nagpal ⁵, Reed Busse², and Martin Janich^6
1GE Healthcare, Menlo Park, CA, United States, ²GE Healthcare, Waukesha, WI, United States, ³Radiology, Stanford University, Palo Alto, CA, United States, ⁴Radiological Consultants, Inova Fairfax Hospital, Fairfax, VA, United States, ⁵Radiology, University of Wisconsin–Madison, Madison, WI, United States, ⁶GE Healthcare, Munich, Germany

Keywords: Heart, Image Reconstruction, Deep learning, reconstruction

Cardiac bSSFP Cine is widely used clinically; however, it is time consuming and requires multiple breath-holds. Deep learning-based accelerated Cine (DLCine) is a novel technique combining accelerated variable density sampling and deep learning regularized reconstruction that allows much higher acceleration compared to conventional Cine with parallel imaging. The purpose of this work was to compare image quality and global cardiac function utilizing DLCine versus conventional Cine by three expert readers. The results demonstrate that DLCine can be used to reduce the scan time while maintaining image quality and providing accurate global cardiac function measurement.

View Presentation Video

Isabel Montón Quesada¹, Augustin C. Ogier¹, Jérôme Yerly^1,2, Jonas Richiardi¹, Christopher W. Roy¹, Juerg Schwitter³, Matthias Stuber^1,2, and Ruud B. van Heeswijk^1
1Department of Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, ²CIBM Center for BioMedical Imaging, Lausanne, Switzerland, ³Cardiology Service, Cardiovascular Department, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland

Keywords: Heart, Image Reconstruction

Several cardiac and respiratory motion-resolved free-running acquisition techniques have recently been developed, but automated self-gated motion extraction with principal component analysis (PCA) alone remains challenging and sometimes fails to correctly identify the components related to cardiac motion. We combined three blind source separation algorithms (fastICA, ICASSO and SOBI) with PCA and compared them with gold-standard ECG R-wave triggers in 10 healthy volunteers to determine the most precise and robust self-gated cardiac trigger extraction. Of these tested techniques, PCA combined with SOBI resulted in the lowest variability of the cardiac intervals relative to ECG intervals (p=0.03) and the sharpest images (p=0.03).

View Presentation Video

Shigeo Okuda¹, Ryo Tsukada², Manabu Arai¹, Sari Motomatsu², Atsushi Nozaki³, Xucheng Zhu⁴, and Masahiro Jinzaki^1
1Radiology, Keio University School of Medicine, Tokyo, Japan, ²Keio University Hospital, Tokyo, Japan, ³GE Healthcare Japan, Tokyo, Japan, ⁴GE Healthcare, Menlo Park, CA, United States

Keywords: Heart, Cardiovascular, Accelerated cine

Accelerated cardiac cine was obtained with deep learning reconstructed technique on ten patients (DL Cine). Three series of DL Cine with different parameters were acquired, including reduction factor (RF) of 12 under free breathing (FB), RF of 12 during one breathhold (R12) and RF = 9 dividing the left ventricular short axis into two slabs during each breathhold (R9). The two readers evaluate image quality (IQ) score and measure the cardiac functional parameters and compared them with the conventional ASSET cine. Although the IQ score was smaller than ones of the conventional cine, they are clinically acceptable. We found good correlations between volumetry on the conventional and DL cine.

View Presentation Video

Runyu Yang¹, Haozhong Sun¹, and Huijun Chen^1
1Center for Biomedical Imaging Research, Tsinghua University, Beijing, China

Keywords: Heart, Image Reconstruction

Obtaining high spatial and temporal resolution image in dynamic magnetic resonance imaging is a huge challenge. It is effective to use low-rank and sparse prior jointly for dMRI reconstruction. However, the models represented the tensor data into a matrix, which can not completely explore the spatiotemporal correlation information. Besides, using nuclear norm as convex surrogate of the rank function to enforce the low-rank, which may lead to the sub-optimal result. Hence, we proposed the piecewise frequency tensor nuclear norm in the low-rank and joint sparse prior to reconstructed data. The proposed method was tested in cardiac cine and perfusion data.