Accelerated deep-learning reconstructed cine balanced steady-state free precession cine imaging: Potential for workflow improvement. (RF_TH_249)

Thursday, January 30, 2025

3:00 PM – 3:10 PM East Coast USA Time

Location: Blue Room PreFunction

Presenting Author(s)

AV

Aishwariya Vegunta, MD

MRI Fellow
U Wisconsin Madison

Co-Author(s)

AV

Aishwariya Vegunta, MD

MRI Fellow
U Wisconsin Madison
Jitka Starekova, MD

Scientist
University of Wisconsin-Madison
TT

Tilo Thottakara, MD

Postdoctoral Scholar
UCSF
JS

Jakub Siembida, MD

Assistant Professor
University of Wisconsin
TG

Thomas M. Grist, MD

Professor
University of Wisconsin
MS

Mark Schiebler, MD

Professor
University of Wisconsin
SR

Scott Reeder, MD, PhD

University of Wisconsin-Madison
David Bluemke

Primary Author(s)

PN

Prashant Nagpal, MD

Associate Professor
University of Wisconsin-Madison

Co-Author(s)

PN

Prashant Nagpal, MD

Associate Professor
University of Wisconsin-Madison

Background: With the rising number of patients requiring cardiac MRI (CMR), the demand for both faster and higher-quality imaging is increasing. A commercially available deep learning (DL) application to accelerate conventional balanced steady-state free precession (bSSFP) cine imaging to achieve shorter scan times without impacting diagnostic value is available. However, the real-world efficacy of this deep learning reconstruction technology is unknown. The purpose of this study is to assess the time savings in clinically indicated CMR exams for cine imaging using DL bSSFP.

Methods: This retrospective study included CMR exams indicated for cardiomyopathy evaluation between October 2023 and August 2024. An accelerated DL cardiac cine bSSFP sequence (Sonic DL, GE Healthcare) was implemented and compared to CMR examinations using conventional bSSFP cine. Total time (in seconds) for cine imaging, including planning slices (axial, approximate 2- and 4-chamber images) and diagnostic slices (cine short-axis (SAX) and long-axis (LAX) images) was determined. In addition, time per slice in seconds was determined for the diagnostic SAX and LAX series. Visual analysis was conducted using a 5-point Likert score (1 – Extremely poor to 5 – Excellent with no artifacts) to evaluate image quality of the exams. For continuous variables two-tailed t-test was used and for categorical a chi-squared test was used.

Results: A total of 80 exams (50% male) were included in this study: 46 with DL bSSFP cine and 34 with conventional bSSFP cines (of which 9 also had DL cine). There was significantly lower scan time (p< 0.001) for the planning sequences, as well as in the scan time for the SAX and LAX sequences (Table 1) when using DL bSSFP cine. At per slice level, the time (in seconds, s) to acquire planning slices was lower for DL bSSFP (4.7 ± 1.5 s) as compared to conventional bSSFP (8.1 ± 1.9 s) (p < 0.001). For diagnostic slices, the time per slice for DL bSSFP was also lower, both for the SAX slices (6.3 ± 2.0 s with DL bSSFP vs 8.4 ± 2.2 s for conventional bSSFP, p < 0.001) and the LAX slices (6.2 ± 1.9 s with DL bSSFP vs 8.2 ± 2.1 s for conventional bSSFP, p < 0.001).

In the 9 patients with both conventional and DL bSSFP cines, scan time with DL bSSFP cine was reduced by an average of 4.2 ±1.7 s (p< 0.001). Overall image quality scores for SAX and LAX cine series were higher for DL bSSFP compared to conventional bSSFP (SAX: 4.0 vs 2.6; LAX: 4.0 vs 2.4) based on a 5-point Likert scale (1 = Extremely Poor, 5 = Excellent), with p < 0.001.

Conclusion: A commercially available deep-learning cine sequence allowed 25-40% lower cine CMR scan time and greater image quality. This time savings allowed for workflow improvements and enables exams for patients who have difficulty holding their breath.