Whole-Ventricle Cardiac Cine MRI in a Single Breath-Hold With Slice- and Time-Dependent Deep Image Prior at 0.55T.
Authors
Affiliations (5)
Affiliations (5)
- Millennium Institute for Intelligent Healthcare Engineering, Santiago, Chile.
- School of Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile.
- IMPACT, Center of Interventional Medicine for Precision and Advanced Cellular Therapy, Santiago, Chile.
- Institute for Biological and Medical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile.
- School of Biomedical Engineering and Imaging Sciences, King's College London, London, UK.
Abstract
To develop a whole-ventricle cardiac cine MRI radial acquisition at 0.55T in a single breath-hold scan of approximately 12-15 s, reconstructed using a novel self-supervised slice and time-dependent deep image prior (ST-DIP) neural network. A whole-ventricle cardiac cine MRI at 0.55T is acquired in a single breath-hold by collecting eight slices (1.5 s per slice) and reconstructing them jointly using ST-DIP. Acquisition uses a tiny golden angle radial trajectory and a balanced steady-state free precession (bSSFP) readout. Reconstruction uses a frequency-weighted consistency loss in the frequency domain and total variation regularization in the image domain, sharing the same network weights for all slices. The proposed approach was tested in 10 healthy subjects and was compared with time-dependent deep image prior (TD-DIP) and conventional reconstruction methods. ST-DIP achieves an edge sharpness of <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>1.15</mn> <mo>±</mo> <mn>0.32</mn></mrow> </math> [ <math xmlns="http://www.w3.org/1998/Math/MathML"> <mrow><msup><mtext>pixel</mtext> <mrow><mo>-</mo> <mn>1</mn></mrow> </msup> </mrow> </math> ] at the septum, slightly higher than TD-DIP ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>1.08</mn> <mo>±</mo> <mn>0.21</mn></mrow> </math> [ <math xmlns="http://www.w3.org/1998/Math/MathML"> <mrow><msup><mtext>pixel</mtext> <mrow><mo>-</mo> <mn>1</mn></mrow> </msup> </mrow> </math> ]) and significantly higher than GRASP ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>0.72</mn> <mo>±</mo> <mn>0.13</mn></mrow> </math> [ <math xmlns="http://www.w3.org/1998/Math/MathML"> <mrow><msup><mtext>pixel</mtext> <mrow><mo>-</mo> <mn>1</mn></mrow> </msup> </mrow> </math> ]). The LVEF obtained with ST-DIP was <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>58.7</mn> <mo>%</mo> <mo>±</mo> <mn>3.7</mn> <mo>%</mo></mrow> </math> , comparable to the 15-s-per-slice radial reference scan ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>59.1</mn> <mo>%</mo> <mo>±</mo> <mn>5.1</mn> <mo>%</mo></mrow> </math> ). ST-DIP reconstructed 8 slices simultaneously in <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>16</mn> <mo>±</mo> <mn>2</mn></mrow> </math> min, whereas TD-DIP, GRASP, and SENSE required sequential reconstruction times of <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>45</mn> <mo>±</mo> <mn>4</mn></mrow> </math> , <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>12</mn> <mo>±</mo> <mn>2</mn></mrow> </math> , and 3 min, respectively. ST-DIP enables whole-ventricle cine imaging in a single breath-hold and reconstruction of high-quality cine images at 0.55T, outperforming conventional methods in image quality while providing comparable LVEF to a 15-s-per-slice radial reference cine scan.