The Need

Cardiovascular MRI (CMR) data acquisition is often time-consuming and relies on complex reconstruction methods that do not fully exploit the spatial and temporal structure of CMR data. There is a need for a more efficient, unsupervised approach that can accelerate CMR imaging while maintaining diagnostic quality, reducing the dependency on large training datasets.

The Technology

M-DIP (motion-guided deep image prior) is an innovative framework for unsupervised accelerated CMR reconstruction. It employs three deep learning generators to recover dynamic imaging series from undersampled k-space data. These generators create static image dictionaries, generate time-dependent weights, and produce deformation fields to warp images, capturing cardiac and respiratory motions as well as contrast dynamics.

Benefits/Advantages

• Unsupervised approach reduces the need for large training datasets.
• Efficiently captures complex cardiac and respiratory motions.
• Adaptable to a broad range of CMR applications.
• Potential to significantly reduce imaging time.
• Maintains high diagnostic image quality.