MRI Scan Time Reduction Method Using Deep Learning

A reconstruction algorithm based on a generative diffusion model for multi-coil, highly undersampled non-Cartesian MRI allowing drastic reduction of scan time
Problem:
MRI suffers from slow data acquisition due to hardware constraints, leading to prolonged scan times and challenges like reduced patient compliance and motion artifacts. Current acceleration techniques allow to reduce data acquisition time compromise image quality, requiring additional information for image reconstruction algorithms.
Technology:
This invention addresses the lack of diffusion model-based deep learning approaches for image reconstruction from data collected along non-Cartesian sampling trajectories, including the related tasks of scan acceleration, motion correction, and artifact correction. The system allows for significant scan time acceleration much faster than conventional and commonly used sequences, and multiple times faster than comparable proposed deep learning methods. This is achieved due to combining multicoil imaging, spiral scanning, efficient scanning trajectories and sequences and intelligent image reconstruction. Faster scans can improve patient comfort and compliance, enable scanning protocols that are currently infeasible due to long scan times, reduce artifacts.
Advantages:

High quality reconstructed images
Ultrafast scan times compared what are currently possible

Stage of Development:

Proof of Concept
Model trained on about 7,000 brain scans
Spiral MRI sequence which can highly undersample data acquisition

The effect of optimal sampling trajectory without model reconstruction, model reconstruction without frequency guidance, and model reconstruction with frequency guidance. All three contribute to noticeable improvements in image quality, as measured by Structural Similarity Index Measure (SSIM). SSIM values range from 0 to 1, with higher values indicating greater similarity to the fully sampled MRI scan. The combination of choosing an optimal trajectory, performing model reconstruction, and using frequency guidance results in the high SSIM when compared with the reference.
Intellectual Property: