Tricuspid valve flow measurement using a deep learning framework for automated valve-tracking 2D phase contrast
Lamy, Jérôme ; Gonzales, Ricardo A. ; Xiang, Jie ; Seemann, Felicia ; Huber, Steffen ; Steele, Jeremy ; Wieben, Oliver ; Heiberg, Einar LU
and Peters, Dana C.
(2024)
In Magnetic Resonance in Medicine
92(5).
p.1838-1850
- Abstract
Purpose: Tricuspid valve flow velocities are challenging to measure with cardiovascular MR, as the rapidly moving valvular plane prohibits direct flow evaluation, but they are vitally important to diastolic function evaluation. We developed an automated valve-tracking 2D method for measuring flow through the dynamic tricuspid valve. Methods: Nine healthy subjects and 2 patients were imaged. The approach uses a previously trained deep learning network, TVnet, to automatically track the tricuspid valve plane from long-axis cine images. Subsequently, the tracking information is used to acquire 2D phase contrast (PC) with a dynamic (moving) acquisition plane that tracks the valve. Direct diastolic net flows evaluated from the dynamic PC... (More)
Purpose: Tricuspid valve flow velocities are challenging to measure with cardiovascular MR, as the rapidly moving valvular plane prohibits direct flow evaluation, but they are vitally important to diastolic function evaluation. We developed an automated valve-tracking 2D method for measuring flow through the dynamic tricuspid valve. Methods: Nine healthy subjects and 2 patients were imaged. The approach uses a previously trained deep learning network, TVnet, to automatically track the tricuspid valve plane from long-axis cine images. Subsequently, the tracking information is used to acquire 2D phase contrast (PC) with a dynamic (moving) acquisition plane that tracks the valve. Direct diastolic net flows evaluated from the dynamic PC sequence were compared with flows from 2D-PC scans acquired in a static slice localized at the end-systolic valve position, and also ventricular stroke volumes (SVs) using both planimetry and 2D PC of the great vessels. Results: The mean tricuspid valve systolic excursion was 17.8 ± 2.5 mm. The 2D valve-tracking PC net diastolic flow showed excellent correlation with SV by right-ventricle planimetry (bias ± 1.96 SD = −0.2 ± 10.4 mL, intraclass correlation coefficient [ICC] = 0.92) and aortic PC (−1.0 ± 13.8 mL, ICC = 0.87). In comparison, static tricuspid valve 2D PC also showed a strong correlation but had greater bias (p = 0.01) versus the right-ventricle SV (10.6 ± 16.1 mL, ICC = 0.61). In most (8 of 9) healthy subjects, trace regurgitation was measured at begin-systole. In one patient, valve-tracking PC displayed a high-velocity jet (380 cm/s) with maximal velocity agreeing with echocardiography. Conclusion: Automated valve-tracking 2D PC is a feasible route toward evaluation of tricuspid regurgitant velocities, potentially solving a major clinical challenge.
(Less)
- author
- Lamy, Jérôme
; Gonzales, Ricardo A.
; Xiang, Jie
; Seemann, Felicia
; Huber, Steffen
; Steele, Jeremy
; Wieben, Oliver
; Heiberg, Einar
LU
and Peters, Dana C.
- organization
- publishing date
- 2024-11
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- deep learning, phase contrast, regurgitation, slice-following, tricuspid valve
- in
- Magnetic Resonance in Medicine
- volume
- 92
- issue
- 5
- pages
- 13 pages
- publisher
- John Wiley & Sons Inc.
- external identifiers
-
- scopus:85194834294
- pmid:38817154
- ISSN
- 0740-3194
- DOI
- 10.1002/mrm.30163
- language
- English
- LU publication?
- yes
- id
- 733f9ddf-127b-45f4-84fa-5762ef871ffb
- date added to LUP
- 2024-10-14 15:15:22
- date last changed
- 2025-07-08 13:52:10
@article{733f9ddf-127b-45f4-84fa-5762ef871ffb,
abstract = {{<p>Purpose: Tricuspid valve flow velocities are challenging to measure with cardiovascular MR, as the rapidly moving valvular plane prohibits direct flow evaluation, but they are vitally important to diastolic function evaluation. We developed an automated valve-tracking 2D method for measuring flow through the dynamic tricuspid valve. Methods: Nine healthy subjects and 2 patients were imaged. The approach uses a previously trained deep learning network, TVnet, to automatically track the tricuspid valve plane from long-axis cine images. Subsequently, the tracking information is used to acquire 2D phase contrast (PC) with a dynamic (moving) acquisition plane that tracks the valve. Direct diastolic net flows evaluated from the dynamic PC sequence were compared with flows from 2D-PC scans acquired in a static slice localized at the end-systolic valve position, and also ventricular stroke volumes (SVs) using both planimetry and 2D PC of the great vessels. Results: The mean tricuspid valve systolic excursion was 17.8 ± 2.5 mm. The 2D valve-tracking PC net diastolic flow showed excellent correlation with SV by right-ventricle planimetry (bias ± 1.96 SD = −0.2 ± 10.4 mL, intraclass correlation coefficient [ICC] = 0.92) and aortic PC (−1.0 ± 13.8 mL, ICC = 0.87). In comparison, static tricuspid valve 2D PC also showed a strong correlation but had greater bias (p = 0.01) versus the right-ventricle SV (10.6 ± 16.1 mL, ICC = 0.61). In most (8 of 9) healthy subjects, trace regurgitation was measured at begin-systole. In one patient, valve-tracking PC displayed a high-velocity jet (380 cm/s) with maximal velocity agreeing with echocardiography. Conclusion: Automated valve-tracking 2D PC is a feasible route toward evaluation of tricuspid regurgitant velocities, potentially solving a major clinical challenge.</p>}},
author = {{Lamy, Jérôme and Gonzales, Ricardo A. and Xiang, Jie and Seemann, Felicia and Huber, Steffen and Steele, Jeremy and Wieben, Oliver and Heiberg, Einar and Peters, Dana C.}},
issn = {{0740-3194}},
keywords = {{deep learning; phase contrast; regurgitation; slice-following; tricuspid valve}},
language = {{eng}},
number = {{5}},
pages = {{1838--1850}},
publisher = {{John Wiley & Sons Inc.}},
series = {{Magnetic Resonance in Medicine}},
title = {{Tricuspid valve flow measurement using a deep learning framework for automated valve-tracking 2D phase contrast}},
url = {{http://dx.doi.org/10.1002/mrm.30163}},
doi = {{10.1002/mrm.30163}},
volume = {{92}},
year = {{2024}},
}