Machine Learning and Artificial Intelligence
Yu Wang, PhD
Graduate Research Assistant
University of Virginia
Charlottesville, Virginia, United States
Yu Wang, PhD
Graduate Research Assistant
University of Virginia
Charlottesville, Virginia, United States
Changyu Sun, PhD
Assistant Professor
University of Missouri-Columbia
Columbia, Missouri, United States
Sona Ghadimi, PhD
Research Scientist
University of Virginia, United States
Daniel Auger, PhD
Sr. Client Services Manager
Medical Metrics, Inc
Houston, Texas, United States
Pierre Croisille, MD, PhD
Professor
INSA Lyon, Rhone-Alpes, France
Magalie Viallon, PhD
Senior Researcher
INSA Lyon
Saint-Etienne, Rhone-Alpes, France
Kenneth Mangion, MD PhD
Clinical Lecturer and Cardiology Registrar
University of Glasgow
Glasgow, Scotland, United Kingdom
Colin Berry, MD, PhD
Professor of Cardiology
University of Glasgow
Glasgow, Scotland, United Kingdom
Christopher M. Haggerty, PhD
Associate Professor
Geisinger Health System, Pennsylvania, United States
Linyuan Jing, PhD
Data Scientist
Geisinger Health System
State College, Pennsylvania, United States
Brandon K. Fornwalt, MD, PhD
Associate Professor
Geisinger Health System, Pennsylvania, United States
J. Jane J. Cao, MD, MPH
Catholic System Research Director & Professor of Clinical Medicine
St. Francis Hospital, The Heart Center
Greenvale, United States
Yang J. Cheng, RT
Chief MRI Technologist
St. Francis Hospital, The Heart Center
Greenvale, New York, United States
Andrew D. Scott, PhD
Senior Physicist
Royal Brompton Hospital
London, England, United Kingdom
.jpg "Pedro F. Ferreira, PhD photo")
Pedro F. Ferreira, PhD
Senior Physicist
Royal Brompton Hospital
London, England, United Kingdom
John N. Oshinski, PhD
Professor of Radiology and Biomedical Engineering
Emory University
Atlanta, Georgia, United States
Daniel B. Ennis, PhD
Professor
Stanford University
Stanford, California, United States
Kenneth Bilchick, MD
Professor
University of Virginia Health System
Charlottesville, Virginia, United States
Frederick Epstein, PhD
Professor
University of Virginia
Charlottesville, Virginia, United States
Displacement encoding with stimulated echoes (DENSE) provides accurate global and segmental myocardial circumferential strain (Ecc), but strain-dedicated acquisitions are required [1-3]. Feature tracking (FT) applied to routine cine images also provides strain data, but remains less accurate [2-4]. As DENSE provides both myocardial contours and accurate intramyocardial tissue displacement measurements, we investigated the use of DENSE data to train a supervised deep network (StrainNet) for intramyocardial tissue motion prediction from contour motion. We aimed to show that StrainNet analysis of balanced steady state free precession (bSSFP) cine images could provide the clinical convenience of FT and better agreement with DENSE.
Methods:
Eight centers participated in the study, and 144 healthy volunteers and 161 patients with seven types of heart diseases were included: (a) myocardial infarction (n = 62), (b) heart failure with left bundle branch block (n = 47), (c) hypertrophic cardiomyopathy (n = 17), (d) amyloidosis (n = 13), (e) dilated cardiomyopathy (n = 10), (f) ischemic heart disease without infarction (n = 7), and (g) myocarditis (n = 5). A 3D (2D+t) U-Net was trained to predict intramyocardial displacement from contour data (Fig. 1). The 305 subjects were randomly divided 80:20 into training and testing datasets, and this ratio was applied to each sub-group (healthy volunteers and each disease type). During training, the U-Net inputs were a time series of myocardial contours derived from DENSE magnitude images and the ground-truth output data were the intramyocardial 2D Lagrangian displacement measurements derived from DENSE phase images. For testing, StrainNet was applied to contours derived from FT software (NeoSoft, suiteHeart) applied to standard bSSFP images.
Results:
Fig. 2 shows examples of end-systolic displacement and circumferential strain (Ecc) maps comparing StrainNet analysis of cine images and DENSE for a healthy volunteer and a heart failure patient with left bundle branch block (LBBB). For the healthy subject, StrainNet qualitatively depicted normal displacement and Ecc, in good agreement with DENSE. For the LBBB patient, StrainNet showed simultaneous stretching of the septal segments and contraction of the lateral wall, also with generally good agreement with DENSE. Fig. 3A-B shows correlations and agreement of end-systolic Ecc between StrainNet, FT, and DENSE. Intraclass correlation coefficient (ICC), Pearson CC, coefficient of variation (CV) and the corresponding biases and limits of agreement all show that StrainNet outperformed FT for both global and segmental Ecc (Fig. 3C-D), with DENSE as the reference.
Conclusion:
For both healthy volunteers and patients, StrainNet predicts intramyocardial displacement and strain from myocardial contour motion and shows good agreement with DENSE. For the analysis of routine cine MRI, StrainNet shows better agreement than FT with DENSE for global and segmental circumferential strain.