Multi-Frame Image Registration for Automated Ventricular Function Assessment in Single Breath-Hold Cine MRI Using...

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Cardiac ventricular function assessment is crucial for diagnosing cardiovascular diseases.
• Current MRI methods can be time-consuming and operator-dependent.
• Accelerated imaging techniques offer potential for faster scans but may compromise image quality.

Purpose of the Study:

• To develop an automated, operator-independent framework for assessing cardiac ventricular function using highly accelerated MRI.
• To improve the reliability and efficiency of cardiac function analysis from MR images.

Main Methods:

• A deep learning framework integrating image registration, motion-compensated reconstruction, and segmentation in a synergistic loop.
• Utilized fully sampled and retrospectively accelerated MR images from healthy and diseased subjects.
• Evaluated the performance of registration and segmentation tasks and their impact on ventricular function parameters.

Main Results:

• The framework demonstrated robustness to undersampling artifacts and required minimal annotation.
• Achieved a 9-10% increase in Dice similarity for ventricular delineation compared to existing methods.
• Predicted ejection fraction strongly correlated (>0.9) with manual measurements.
• Enabled consistent strain measurements across accelerations up to R=24.

Conclusions:

• A comprehensive cardiac ventricular function analysis is feasible with highly accelerated cine MR data and minimal annotation.
• The developed multitasking deep learning strategy enhances accessibility of cardiac function analysis.
• This approach has the potential to enable efficient cardiac function assessment within a single breath-hold.