Predictive signal modeling and multi-rate filtering in accelerated cardiac MRI

Area of Science:

• Medical Imaging
• Cardiovascular MRI
• Computational Imaging

Background:

• Real-time cardiac MRI (CMR) is crucial for dynamic imaging but faces computational challenges.
• Existing methods struggle with capturing cardiac motion and irregular rhythms efficiently.

Purpose of the Study:

• To develop strategies for real-time CMR by enhancing cardiac motion prediction and computational efficiency.
• To reduce artifacts and preserve spatial resolution in dynamic CMR.

Main Methods:

• Modified a predictive signal model (mPMOT) for faster computation of cardiac motion prediction.
• Developed a multi-rate Kalman filter framework for efficient reconstruction of high-resolution CMR data.
• Evaluated reconstructions using multi-rate Kalman filtering and compressed sensing on human and swine datasets.

Main Results:

• mPMOT demonstrated a two-order-of-magnitude speedup in training compared to the original PMOT.
• Achieved high-quality CMR reconstructions at acceleration factors of 9 and 13.5, reducing temporal blurring artifacts.
• Reduced Kalman filter mean-squared error by two orders of magnitude and computation time from hours to seconds.

Conclusions:

• The computationally efficient mPMOT integrates with existing frameworks for robust real-time CMR.
• Enables improved tracking and reconstruction for dynamic cardiac imaging applications.