Machine Intelligence for Accelerated Magnetic Resonance Imaging
Approved Research ID: 61943
Approval date: August 7th 2020
In this research project, we address the current main limitation of magnetic resonance imaging (MRI): long acquisition times. Despite the numerous advantages of MRI, including absence of radiation, high spatial details, and excellent soft-tissue contrast, it is not as easily accessible as modalities such as ultrasound imaging or CT, and it is not the modality of choice for imaging moving organs such as the heart. Speed is the single most critical bottleneck to MRIs taking a dominant position and offering its full advantages in the clinical setting.
To address this challenge, we propose to use machine intelligence to assist us in producing diagnostically accurate and detailed MRI scans even when we acquire only a fraction of data. We know that, as in JPEG images and MPEG movies, a lot of data is redundant and can be thrown away. Now, we are turning to machine intelligence to help us identify what and how much can be thrown away without compromising diagnostic integrity. Preliminary work already suggests an acceleration factor of 8 is possible. We will work towards achieving an acceleration factor of 30 for dynamic imaging scenarios, such as when we image the heart or our gastrointestinal organs.
Our mission is expansive and we plan on developing the core technology platform in 3 years. We will tackle accelerating not only static image acquisitions but, more importantly, dynamic image acquisitions (i.e. a time-series of images). By doing so, we are including every application of MRI that is seen in the clinic. If successful, patients may finally have access to MRI screening and repeated post-operative follow-up. More importantly, MRIs may become a go-to diagnostic tool that can be delivered to the patient shortly after a scan is requested.