Albert Hsiao

Claudia Prieto

Jennifer Steeden¹, Michael Quail², Alexander Gotschy^2,3, Andreas Hauptmann^1,4, Rodney Jones¹, and Vivek Muthurangu^1
1University College London, London, United Kingdom, ²Great Ormond Street Hospital, London, United Kingdom, ³University and ETH Zurich, Institute for Biomedical Engineering, Zurich, Switzerland, ⁴University of Oulu, Oulu, Finland

Three-dimensional (3D), whole heart, balanced steady state free precession (WH-bSSFP) sequences provides excellent delineation of both intra-cardiac and vascular anatomy. However, they are usually cardiac triggered and respiratory navigated, resulting in long acquisition times (10-15minutes). Here, we propose a machine-learning single-volume super-resolution reconstruction (SRR), to recover high-resolution features from rapidly acquired low-resolution WH-bSSFP data. We show that it is possible to train a network using synthetically down-sampled WH-bSSFP data. We tested the network on synthetic test data and 40 prospective data sets, showing ~3x speed-up in acquisition time, with excellent agreement with reference standard high resolution WH-bSSFP images. 

Niccolo Fuin¹, Giovanna Nordio¹, Thomas Kuestner¹, Radhouene Neji², Karl Kunze², Yaso Emmanuel³, Alessandra Frigiola^1,3, Rene Botnar^1,4, and Claudia Prieto^1,4
1Biomedical Engineering Department, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²MR Research Collaborations, Siemens Healthcare Limited, Frimley, United Kingdom, ³Guy’s and St Thomas’ Hospital, NHS Foundation Trust, London, United Kingdom, ⁴Pontificia Universidad Católica de Chile, Santiago, Chile

Bright- and black-blood MRI sequences provide complementary diagnostic information in patients with congenital heart disease (CHD). A free-breathing 3D whole-heart sequence (MTC-BOOST) has been recently proposed for contrast-free simultaneous bright- and black-blood MRI, demonstrating high-quality depiction of arterial and venous structures. However, high-resolution fully-sampled MTC-BOOST acquisitions require long scan times of ~12min. Here we propose a joint Multi-Scale Variational Neural Network (MS-VNN) which enables the acquisition of high-quality bright- and black blood MTC-BOOST images in ~2-4 minutes, and their joint reconstruction in ~20s. The technique is compared with Compressed-Sensing reconstruction for 5x acceleration, in CHD patients.

Michael Loecher¹, Luigi E Perotti², and Daniel B Ennis^1,3,4,5
1Radiology, Stanford, Palo Alto, CA, United States, ²Mechanical Engineering, University of Central Florida, Orlando, FL, United States, ³Radiology, Veterans Administration Health Care System, Palo Alto, CA, United States, ⁴Cardiovascular Institute, Stanford, Palo Alto, CA, United States, ⁵Center for Artificial Intelligence in Medicine & Imaging, Stanford, Palo Alto, CA, United States

A convolutional neural network based tag tracking method for cardiac grid-tagged data was developed and validated.  An extensive synthetic data simulator was created to generate large amounts of training data from natural images with analytically known ground-truth motion.  The method was validated using both a digital cardiac deforming phantom and tested using in vivo data. Very good agreement was seen in tag locations (<1.0mm) and calculated strain measures (<0.02 midwall Ecc)

Teodora Chitiboi¹, Bogdan Georgescu¹, Jens Wetzl², Indraneel Borgohain¹, Christian Geppert², Stefan K Piechnik³, Stefan Neubauer³, Steffen Petersen⁴, and Puneet Sharma^1
1Siemens Healthineers, Princeton, NJ, United States, ²Magnetic Resonance, Siemens Healthcare, Erlangen, Germany, ³Division of Cardiovascular Medicine, Radcliffe Department of Medicine, University of Oxford, Oxford, United Kingdom, ⁴NIHR Biomedical Research Centre at Barts, Queen Mary University of London, London, United Kingdom

Deep learning enables fully automatic strain analysis from CINE MRI on large subject cohorts. Deep learning neural nets were trained to segment the heart chambers from CINE MRI using manually annotated ground truth. After validation on more than 1700 different patient datasets, the models were used to generate segmentations as the first step of a fully automatic strain analysis pipeline for 460 subjects. We found significant differences associated with gender (strain magnitude smaller for males), height (lower strain magnitude for patients taller than 170 cm) and age (lower circumferential and longitudinal strain for subjects older than 60 years).

Thomas Joyce¹ and Sebastian Kozerke^1
1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland

We present a method for the controllable synthesis of 3D (volumetric) MRI data. The model is comprised of three components which are learnt simultaneously from unlabelled data through self-supervision: i) a multi-tissue anatomical model, ii) a probability distribution over deformations of this anatomical model, and, iii) a probability distribution over ‘renderings’ of the anatomical model (where a rendering defines the relationship between anatomy and resulting pixel intensities). After training, synthetic data can be generated by sampling the deformation and rendering distributions.