FREE-BREATHING CARDIOVASCULAR MRI USING A PLUG-AND-PLAY METHOD WITH LEARNED DENOISER

Area of Science:

• Medical Imaging
• Cardiovascular System
• Magnetic Resonance Imaging

Background:

• Cardiac magnetic resonance imaging (CMR) is crucial for cardiovascular assessment.
• Long acquisition times limit the clinical utility of CMR.
• Accelerating CMR is essential for improving patient throughput and experience.

Purpose of the Study:

• To propose and validate a novel plug-and-play (PnP) method for accelerated CMR reconstruction.
• To integrate a deep learning (DL) denoiser within the PnP framework for enhanced image quality.
• To evaluate the performance of the PnP-DL method against traditional compressed sensing (CS) techniques.

Main Methods:

• Developed a plug-and-play (PnP) framework for CMR image reconstruction.
• Integrated a deep learning (DL) denoiser trained on spatiotemporal patches from high-quality cine images.
• Compared the PnP-DL method with compressed sensing (CS) on breath-held and real-time (RT) free-breathing datasets.

Main Results:

• PnP-DL demonstrated over 1 dB improvement compared to CS methods on breath-held datasets.
• Qualitative evaluations showed superior performance of PnP-DL on RT free-breathing datasets.
• The PnP-DL method effectively reconstructs high-quality CMR images from undersampled data.

Conclusions:

• The PnP-DL method offers a significant advancement in CMR image reconstruction.
• This technique has the potential to substantially accelerate real-time (RT) free-breathing CMR.
• PnP-DL can enhance diagnostic capabilities and clinical workflow in cardiovascular imaging.