CMR-Analysis (including machine learning)

1348492 - The Effect of Respiratory Position for Free Breathing Cine CMR LVEF Quantification

Wednesday, January 25, 2023

Location: E-POSTER

Presenting Author(s)

MA

Marina M. Awad

Undergraduate Student
University of Virginia
Charlottesville, Virginia, United States

Primary Author(s)

MA

Marina M. Awad

Undergraduate Student
University of Virginia
Charlottesville, Virginia, United States

Co-Author(s)

Junyu Wang, PhD

Postdoctoral Fellow
Stanford University
Palo Alto, California, United States
Christopher M. Kramer, MD

Chief
University of Virginia Health System
Charlottesville, Virginia, United States
Michael Salerno, MD, PhD, MSc

Professor
Stanford University, California, United States

Disclosure(s):

Michael Salerno, MD, PhD, MSc: No relevant disclosure to display

Background: Free-breathing real-time cine CMR imaging is useful for patients with limited breath-hold capacity and/or arrhythmias¹. However, quantifying the left ventricular ejection fraction (LVEF) from free-breathing real-time imaging is limited by the lack of an EKG signal to define the cardiac cycle, and through-plane cardiac motion resulting from free-breathing. We developed a technique that determines end-expiration based on the center of mass of the segmentation mask and determines end-systole (ES) and end-diastole (ED) based on the temporal changes in area of the segmentation mask. Here, we sought to analyze the effects of respiratory positions on the LVEF quantification of free breathing cine CMR.

Methods: The LVEF was quantified from 10 free-breathing spiral cine patient cases, reconstructed using the DESIRE framework² and our DL-based segmentation technique³, and compared to manual contouring of breath-held segmented SSFP images as the reference. Data was acquired using a continually rotating golden-angle spiral trajectory with 8 seconds of data per slice location. The spatial resolution was 1.3 mm², and the temporal resolution was 35 ms. We compared quantification of LVEF during free-breathing using the following strategies: (1) end-expiration, (2) end-inspiration (3) single heartbeat (HB) (4) average over 3 HB and (5) average over 5 HB. The contour’s centroid at end-expiration and end-inspiration was determined from the peaks and troughs of the filtered vertical location of centroid vs frame graph respectively (Fig. 1). The average ED and ES mask area for each slice was multiplied by the voxel volume and summed over the slices to obtain the EDV and ESV. LVEF was calculated using the formula LVEF=(EDV-ESV)/EDV × 100%. For the 1, 3, and 5 HB strategies, the ED and ES volumes associated with the first one, three, and five heartbeats were used without considering the respiratory position.

Results: Using end-expiration (mean bias 0.28 ml (p=0.78)), end-inspiration (mean bias 0.99 ml (p=0.44)), 1 HB (mean bias -1.2 ml (p=0.63)), 3 HB (mean bias 1.6 ml (p=0.17)), and 5 HB (mean bias 1.7 ml (p=0.18)) per slice, had no significant difference (p >0.05) in their mean LVEF as compared to the breath-help SSFP. For the end-expiration, end-inspiration, 1, 3, and 5 HB strategies, the limits of agreement (LOA) range were 12.1, 15.3, 30.1, 13.9, and 15.2 respectively indicating less variability for the end-expiration quantification. Based on a Tukey Test, only the pairwise LOA comparisons with 1 HB had a significant difference (Fig. 3).

Conclusion: The most accurate approach is using end-expiration because it has the lowest mean bias and narrowest LOA, indicating it has the best agreement with the reference⁴. We report that there is no difference in obtaining the LVEF from 3 HB/slice versus 5 HB/slice (p=0.53) according to Tukey’s test. This finding enables the LVEF to be autonomously obtained with less data without losing quantification accuracy.