Dr. Mojgan Hodaie

Brain magnetic resonance imaging (MRI) is a non-invasive, versatile imaging technique for visualization of brain structures. Nearly all neurosurgical patients, for instance, require one or several MRIs prior to diagnosis, and then longitudinally after, for the purposes of monitoring.

MR scanner specifications imply that each center uses MR sequences slightly differently. This makes the cross-comparison between sequences performed in different sites difficult, and is a particular challenge for MR- based research studies.

For example, if the initial imaging lacks contrast if a tumor is suspected, new MR sequences with contrast-enhanced images are required. (This necessitates calling back the patient for another full session of MR imaging, typically lasting up to an hour.

If we compare an MR dataset to a series of highly detailed digital photographs, the concept emerges that the correct software would allow us to focus or identify sufficient detail that would extract the necessary information from these photographs without them needing to be retaken.

This project explores an AI-based approach to MR data acquisition that would remove the heterogeneity in data acquired from different machines and decrease the time required for data acquisition by minimizing the need for contrasting agents.

The generative-based approaches of this project will have valuable and immediate impacts in the way patients undergo brain imaging. Imaging data synthesis can significantly reduce the acquisition time and cost of MR procedures, making the process more efficient and improving the patient experience.