Neural network approaches to correct for motion in 7T MRI scans

Summary

Ultra-high field (7T) MRI allows scans with sub-millimetre spatial resolution which reveal subtle signs of disease that are not detectable by conventional hospital MRI scanners (which have 1.5T or 3T magnets). However, many patient scans are degraded by motion - even small involuntary movements can blur the image and lose some of the benefit of 7T MRI.

We have implemented a prospective "FatNav" sequence that takes "navigator" images of fat (mainly located near the skull) interspersed throughout a clinical imaging sequence. These can be used to track and correct for motion in realtime. However, current algorithms for computing the amount of motion at each navigator take around 2s which slows the rate of motion updates that can be achieved.

Project aims

The student in this project will investigate modern deep neural network approaches for the rigid registration, aiming to implement a method that computes the motion update directly from the raw "k-space" data acquired during the scan.

This will build on a recent proof of concept by Hossbach et al (https://aapm.onlinelibrary.wiley.com/doi/10.1002/mp.16119) but extending to 7T MRI and to 3D coverage.

Contact details

Professor Chris Rodgers - ctr28@cam.ac.uk

Opportunities

This project is open to applicants who want to do a:

• PhD