Aortic and mitral flow quantification using dynamic valve tracking and machine learning: Prospective study assessing static and dynamic plane repeatability, variability and agreement

Julio Garcia; Kailey Beckie; Ali F Hassanabad; Alireza Sojoudi; James A White

doi:10.1177/2048004021999900

Aortic and mitral flow quantification using dynamic valve tracking and machine learning: Prospective study assessing static and dynamic plane repeatability, variability and agreement

JRSM Cardiovasc Dis. 2021 Feb 27:10:2048004021999900. doi: 10.1177/2048004021999900. eCollection 2021 Jan-Dec.

Authors

Julio Garcia^{1

2

3

4

5}, Kailey Beckie^{1

2

3

4}, Ali F Hassanabad^{1

2}, Alireza Sojoudi⁶, James A White^{1

3}

Affiliations

¹ Department of Cardiac Sciences, University of Calgary, Calgary, AB, Canada.
² Department of Radiology, University of Calgary, Calgary, AB, Canada.
³ Stephenson Cardiac Imaging Centre, University of Calgary, AB, Canada.
⁴ Libin Cardiovascular Institute, University of Calgary, Calgary, AB, Canada.
⁵ Alberta Children's Hospital Research Institute, Calgary, AB, Canada.
⁶ Circle Cardiovascular Imaging, Advanced Technologies, Calgary, AB, Canada.

Abstract

Background: Blood flow is a crucial measurement in the assessment of heart valve disease. Time-resolved flow using magnetic resonance imaging (4 D flow MRI) can provide a comprehensive assessment of heart valve hemodynamics but it relies in manual plane analysis. In this study, we aimed to demonstrate the feasibility of automate the detection and tracking of aortic and mitral valve planes to assess blood flow from 4 D flow MRI.

Methods: In this prospective study, a total of n = 106 subjects were enrolled: 19 patients with mitral disease, 65 aortic disease patients and 22 healthy controls. Machine learning was employed to detect aortic and mitral location and motion in a cine three-chamber plane and a perpendicular projection was co-registered to the 4 D flow MRI dataset to quantify flow volume, regurgitant fraction, and a peak velocity. Static and dynamic plane association and agreement were evaluated. Intra- and inter-observer, and scan-rescan reproducibility were also assessed.

Results: Aortic regurgitant fraction was elevated in aortic valve disease patients as compared with controls and mitral valve disease patients (p < 0.05). Similarly, mitral regurgitant fraction was higher in mitral valve patients (p < 0.05). Both aortic and mitral total flow were high in aortic patients. Static and dynamic were good (r > 0.6, p < 0.005) for aortic total flow and peak velocity, and mitral peak velocity and regurgitant fraction. All measurements showed good inter- and intra-observer, and scan-rescan reproducibility.

Conclusion: We demonstrated that aortic and mitral hemodynamics can efficiently be quantified from 4 D flow MRI using assisted valve detection with machine learning.

Keywords: 4D-flow magnetic resonance imaging; Bicuspid aortic valve; machine learning; mitral valve; valve tracking.