David R Rutkowski^1,2, Alejandro Roldán-Alzate^1,2,3, and Kevin M Johnson^1,4
1Radiology, University of Wisconsin, Madison, WI, United States, ²Mechanical Engineering, University of Wisconsin, Madison, WI, United States, ³Biomedical Engineering, University of Wisconsin, Madison, WI, United States, ⁴Medical Physics, University of Wisconsin, Madison, WI, United States

Augmentation of 4D flow MRI data with computational fluid dynamics (CFD) -informed training networks may provide a method to produce highly accurate physiological flow fields. In this preliminary work, the potential utility of such a method was demonstrated by using high resolution patient-specific CFD data to train a neural network, and then using the trained network to enhance MRI-derived velocity fields.

Valery Vishnevskiy^1,2, Hannes Dillinger^1,2, Jonas Walheim^1,2, Lin Zhang¹, and Sebastian Kozerke^1,2
1ETH Zurich, Zurich, Switzerland, ²University of Zurich, Zurich, Switzerland

A novel approach using Cholesky decomposition-based neural networks for the estimation of Reynolds stress tensors in 4D Flow MRI is presented. Evaluation is carried out for simulated MRI signals using particle tracking velocimetry data and tested on in-vivo data obtained in a healthy volunteer and a patient with bioprosthetic aortic valve. The proposed method allows to account for non-Gaussian acquisition noise and guarantees positive-definiteness of the estimated tensors, which yields 68% improvement in turbulent shear stress estimation compared to standard least squares estimation.

Haben Berhane¹, Michael Scott², Takashi Fujiwara³, Lorna Browne³, Joshua Robinson¹, Cynthia Rigsby¹, Michael Markl², and Alex Barker^3
1Lurie Children's Hospital of Chicago, Chicago, IL, United States, ²Northwestern University, Chicago, IL, United States, ³University of Colorado, Anschutz Medical Campus, Aurora, CO, United States

We trained and validated a multi-label convolutional neural network for the segmentation of the aorta and pulmonary arteries from 4D flow MRI for rapid flow analysis across multiple vendors and centers. Using 67 whole-heart 4D flow MRI scans, including 29 with cardiac pathologies, across two institutions and vendors, we trained and tested our CNN using 10-fold cross validation. For flow analysis, We calculated net flow, peak velocity, and Qp-Qs. Across all flow metrics, we found that automated segmentations showed moderate to strong agreement with the manual segmentations, while taking a fraction of the time.

Matthew Van Houten¹, Xue Feng¹, Yang Yang², Austin Robinson³, Craig Meyer¹, and Michael Salerno^1,3
1Department of Biomedical Engineering, University of Virginia, Charlottesville, VA, United States, ²Biomedical Engineering and Imaging Institute and Department of Radiology, Icahn School of Medicine at Mount Sinai, New York, NY, United States, ³Department of Medicine, University of Virginia, Charlottesville, VA, United States

While quantitative first-pass quantitative perfusion imaging is an excellent non-invasive tool for the evaluation of coronary artery disease, current processing shortcomings have kept it from widespread clinical use. In this study, we developed a pipeline which robustly and automatically segments, registers, and quantifies flow with our ultra-high resolution quantitative perfusion sequence.

Einar Heiberg^1,2, Henrik Engblom¹, Marcus Carlsson¹, David Erlinge³, Dan Atar⁴, Anthony H Aletras^1,5, and Hakan Arheden^1
1Department of Clinical Sciences Lund, Clinical Physiology, Skåne University Hospital, Lund University, Lund, Sweden, ²Lund University, Wallenberg Center for Molecular Medicine, Lund, Sweden, ³Lund University, Department of Clinical Sciences Lund, Cardiology, Skåne University Hospital, Lund, Sweden, Lund, Sweden, ⁴Department of Cardiology, Oslo University Hospital Ullevål, and Instititute of Clinical Sciences, University of Oslo, Oslo, Norway, ⁵School of Medicine, Aristotele University of Thessaloniki, Laboratory of Computing, Medical Informatics and Biomedical – Imaging Technologies, Thessaloniki, Greece

The purpose of this study is to systematically evaluate sources to variability in the n-SD from remote method for infarct quantification. Remote ROI position, size, and number of standard deviations all to a large extent affected infarct size. The main driver of infarct variability in the n-SD method are the differences in myocardial SD level, that varies between subjects, site and vendors. Based on the source of variability in infarct size we conclude the n-SD method lack accuracy for infarct quantification, especially in multi-center, multi-vendor setting.

Maurice Pradella¹, Sven Knecht², Manuela Moor¹, Shan Yang¹, Constantinos Anastasopoulos¹, Gian Voellmin², Philip Haaf², Stefan Osswald², Michael Kuehne², Christian Sticherling², Bram Stieltjes¹, and Jens Bremerich^1
1Department for Radiology, University Hospital Basel, Basel, Switzerland, ²Department for Cardiology, University Hospital Basel, Basel, Switzerland

Deep learning based, automatic segmentation of the whole left atrium in cine MRI makes detailed geometrical analysis possible by fitting of an ellipsoid into the contours of the left atrium. Therefore, we could identify the ellipsoidal volume at the time-point before atrial contraction as an independent predictor of recurrence of atrial fibrillation after catheter ablation in a multivariable analysis.

Fan Huang¹, Vince Varut Vardhanabhuti¹, Pek-Lan Khong¹, Ming-Yen NG¹, and Peng Cao^1
1Department of Diagnostic Radiology, The University of Hong Kong, Hong Kong, Hong Kong

We proposed a segmentation method, which is based on a level-set reformulated via a deep recurrent network (RLSNet). This network takes the advantage of U-Net in terms of medical pattern recognition and level-set algorithm in terms of keeping the enclosed and smooth shape of the segmentation contour. We evaluate the network by the segmentation of the left and right ventricles of the heart on cardiac cine Magnetic Resonance Images, which gives greater performance than using U-Net only.

Agisilaos Chartsias¹, Haochuan Jiang¹, Giorgos Papanastasiou^2,3, Chengjia Wang^2,3, Colin Stirrat^2,3, Scott Semple^2,3, David Newby^2,3, Rohan Dharmakumar⁴, and Sotirios A Tsaftaris^1,5
1School of Engineering, University of Edinburgh, Institute of Digital Communications, Edinburgh, United Kingdom, ²Edinburgh Imaging Facility QMRI, Edinburgh, United Kingdom, ³Centre for Cardiovascular Science, Edinburgh, United Kingdom, ⁴Cedars Sinai Medical Center, Los Angeles, CA, United States, ⁵The Alan Turing Institute, London, United Kingdom

We propose a novel deep learning method, Multi-modal Spatial Disentanglement Network (MMSDNet), to segment anatomy in medical images. MMSDNet takes advantage of complementary information provided by multiple sequences of the same patient. Even when trained without annotations, it can segment anatomy (e.g., myocardium) in Late Gadolinium Enhancement (LGE) images, which is essential for assessing myocardial infarction. This is achieved by transferring knowledge from the simultaneously acquired cine-MR data where annotations are easier to be obtained. MMSDNet outperforms classical methods including non-linear registration, and simple copying of contours, as well as the state-of-the-art U-Net model.

Niccolo Fuin¹, Giorgia Milotta¹, Thomas Kuestner¹, Aurelien Bustin¹, Gastao Cruz¹, Rene Botnar^1,2, and Claudia Prieto^1,2
1Biomedical Engineering Department, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²Pontificia Universidad Católica de Chile, Santiago, Chile

T2 mapping is a promising technique for the characterization of myocardial inflammation and oedema. We recently proposed a quantitative 3D whole-heart sequence (qBOOST-T2) which provides co-registered 3D high-resolution bright-blood and T2 map volumes from a single free-breathing scan. However, high-resolution qBOOST-T2 requires long scan times of ~10 min. Here we propose a joint Multi-Scale Variational Neural Network (jMS-VNN) to enable the acquisition of 3D high-resolution bright-blood and accurate T2 map volumes in ~3 mins, and their reconstruction in ~30s. The proposed jMS-VNN jointly reconstructs data from multiple contrasts and efficiently apply dictionary-based signal matching for fast T2 map generation.

Omer Burak Demirel^1,2, Sebastian Weingärtner^1,2,3, Steen Moeller², and Mehmet Akçakaya^1,2
1Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, United States, ²Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States, ³Department of Imaging Physics, Delft University of Technology, Delft, Netherlands

Simultaneous Multi-slice (SMS) imaging has great potential for high acceleration rates with minimal loss in SNR. However, due to the unfavorable coil geometry only moderate acceleration rates have been achieved in cardiac applications. Outer volume suppression (OVS) has been proposed to overcome this limitation by suppressing unwanted signal from extra-cardiac tissues such as chest and back. Despite OVS, existing reconstruction techniques may suffer from residual leakage artifacts or noise amplification. In this work, we sought to extend SPIRiT to a readout-concatenated space to improve image quality in highly accelerated perfusion and cine imaging while mitigating leakage and noise amplification.