Image Reconstruction (including machine learning)
Siyeop Yoon, PhD
Post-doc Research Fellow
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Siyeop Yoon, PhD
Post-doc Research Fellow
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Deliliah Davis
Research Assistant
Beth Israel Deaconess Medical Center, Massachusetts, United States
Eiryu Sai, MD, PhD
Postdoctoral Research Fellow
Beth Israel Deaconess Medical Center
Tokyo, Massachusetts, United States
Kei Nakata, MD
Clinical fellow
Mie University Hospital
Tsu, Mie, Japan
Salah Assana, MSc
Research Assistant
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Manuel A. Morales, PhD
Post-doc Research Fellow
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Amine Amyar, PhD
Postdoctoral Research Fellow
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Jennifer Rodriguez
Clinical Trials Specialist
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Julia Cirillo, BSc
Research Assistant
Beth Israel Deaconess Medical Center, Massachusetts, United States
Beth Goddu, RT
MRI Tech
Beth Israel Deaconess Medical Center, United States
Patrick Pierce, RT
MRI Tech
Beth Israel Deaconess Medical Center, Massachusetts, United States
Warren J. Manning, MD
Professor of Medicine and Radiology
Harvard Medical School
Boston, Massachusetts, United States
Reza Nezafat, PhD
Professor
Harvard Medical School
Boston, Massachusetts, United States
Cine CMR enables accurate and precise measures of cardiac function and volume. However, standard 2D cardiac cine imaging requires 10-12 breath-holds (BH) for covering the entire left ventricle (LV) in the short-axis. This study aimed to evaluate an accelerated ECG-segmented cine (FAST-Cine) protocol for imaging multiple (3-4) slices per BH using a resolution enhancement generative adversarial inline neural network (REGAIN).
Methods:
FAST-Cine imaging is achieved by prescribing a low-resolution image (1/4 reduced spatial resolution along phase-encode) and accelerating the imaging using GRAPPA with rate 2. The low-resolution image is then reconstructed using GRAPPA with zero-padding to create an image with full resolution, albeit with significant blurring. Subsequently, REGAIN enhances spatial resolution as analogous to collecting a full-resolution image. Fig 1A shows the schematic of the FAST-Cine reconstruction that uses REGAIN for deblurring of the low-resolution cine images. REGAIN is a modified generative adversarial neural network (1) that allows flexible input size and reduced network parameters. The generator of REGAIN enhances the image resolution (Fig 1B), and the discriminator classifies the generator's output and high-resolution image for the adversarial training (Fig 1C). To allow inline reconstruction, REGAIN was seamlessly integrated with the scanner using FIRE (2). To train the model, we collected k-space data from ECG-segmented cine images from 343 patients undergoing clinical CMR (Fig 1D). The training data were synthesized by discarding outer ky lines (i.e., reduced spatial resolution) and reconstructed using the same pipeline as the inline reconstruction. The FAST-Cine was evaluated in 61 patients with cardiac disease (39 male; 54±17 years) and 15 healthy participants (6 male; 26±4 years). In each subject, we collected ECG-segmented cine images using (a) standard protocol (spatial resolution 1.7×1.7mm2 , 1 slice per 9sec BH), and (b) FAST-Cine protocol (spatial resolution 6.8×1.7mm2, 3-4 slices per 12sec BH) in the short-axis orientation with the following parameters: 3T Siemens Vida, TR/TE = 3.1/1.4ms, flip angle 43°, slice thickness 8mm, matrix size 208×208, and 16 lines per segment. LV function and volume (LVEDV, LVESV, LV mass, and LVEF) were measured and compared using Bland-Altman analysis and linear regression. FAST-Cine images were successfully acquired in all subjects with excellent image quality (Fig 2). All inline reconstruction took approximately 15 sec after scan completion. The FAST-Cine reduced the number of BHs from 11.2±1.4 to 3.7±1.0 times for covering the entire LV in the short-axis. Bland-Altman analyses (Fig 3A) and regression analysis (Fig 3B) revealed excellent agreement for LVEF, LVEDV, LVESV, and LV mass. The FAST-Cine protocol, based on REGAIN, reduces the number of breath-holds by 3-4-fold and yields similar LV function and volume measurements, reducing breath-hold burden on patients and improving scan efficiency.
Results:
Conclusion: