Tricuspid valve flow measurement using a deep learning framework for automated valve-tracking 2D phase contrast

Area of Science:

• Cardiovascular Imaging
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Measuring tricuspid valve flow velocities using cardiovascular magnetic resonance (CMR) is difficult due to the valve's rapid movement.
• Accurate assessment of tricuspid valve dynamics is essential for evaluating diastolic function.
• Current methods face limitations in directly quantifying flow through the dynamic tricuspid valve.

Purpose of the Study:

• To develop and validate an automated 2D valve-tracking method for measuring flow through the dynamic tricuspid valve.
• To overcome the challenges associated with direct flow evaluation in the moving tricuspid valve plane.
• To improve the assessment of diastolic function through accurate tricuspid valve flow measurements.

Main Methods:

• A deep learning network (TVnet) was employed to automatically track the tricuspid valve plane from cine images.
• A dynamic 2D phase contrast (PC) sequence was utilized, with the acquisition plane synchronized to the tracked valve motion.
• Measurements were compared against static 2D-PC scans and ventricular stroke volumes derived from planimetry and great vessel PC imaging.

Main Results:

• The 2D valve-tracking PC method demonstrated excellent correlation with right-ventricle stroke volume (ICC=0.92) and aortic PC (ICC=0.87).
• Compared to static 2D-PC, the valve-tracking method showed significantly less bias (p=0.01) when compared to right-ventricle stroke volume.
• The technique successfully measured tricuspid regurgitant velocities, including a high-velocity jet in one patient that agreed with echocardiography.

Conclusions:

• Automated 2D valve-tracking using phase contrast imaging is a feasible approach for evaluating tricuspid valve flow and regurgitant velocities.
• This method offers a potential solution to the clinical challenge of accurately assessing tricuspid valve function.
• The technique shows promise for enhancing the evaluation of diastolic function and valvular heart disease.