Static and Dynamic MRI Reconstruction and Segmentation

 

Channel-wise Attention Network Model (MICCAN)

 

 

Motion Guided Dynamic Reconstruction (MODRN)

 

Joint Reconstruction and Segmentation

 

 

Summary:

Magnetic resonance imaging (MRI) technology is a widely used imaging technology but has several drawbacks including: the amount of time necessary for a scan, the requirement for a breath hold when imaging the heart and post-processing that does not support reconstruction and segmentation as a single automated task, without a need for manual intervention.

To better reconstruct high-quality images and investigate the relationship between reconstruction and segmentation, Researchers at Rutgers University have developed a joint method for MRI that focuses on the use of artificial intelligence and machine learning for the reconstruction of images of the heart and cardiac segmentation. Machine learning based methods jointly learn to reconstruct images from dynamic k-space data (subset of data from MRI) and to generate segmentation masks.   Our solution is comprised of three novel techniques:

  • Reconstruction using a channel-wise attention network model that can attend to salient information by filtering irrelevant features and also concentrates on high-frequency information by enforcing low-frequency information bypassed to the final output.
  • Motion-guided dynamic reconstruction that utilizes deep neural networks with motion information to improve reconstruction quality to decompose the motion-guided optimization problem of dynamice MRI reconstruction into three components: dynamic reconstruction, motion estimation and motion compensation.
  • A joint solution to the inverse problem in medical imaging that takes as input the under sampled k-space data and outputs the reconstructed images and the segmented myocardium. (Currently k-space data is transformed into a reconstructed MRI image and after manual intervention, the myocardium contours are extracted using segmentation.) Our solution performs reconstruction and segmentation simultaneously using a novel deep learning approach which consists of a reconstruction module derived from the fast-iterative shrinkage-thresholding algorithm (FISTA) and a segmentation module.

 

Advantages:

  • Improved techniques for Cardia MRI
  • Reduced time required for MRI scans
  • Motion and organ segmentation that does not require a breath hold to cease motion that obscures images that eliminates the need for manual intervention for segmentation
  • Supports static and dynamic reconstruction

 Market Applications:

  • MRI Scanning & Cardiac MRI Scans

Intellectual Property & Development Status:

 

Patent Pending. Available for licensing and/or research collaboration.

Patent Information:
[%Analytics%]