Project Spotlight: FastMRI

The FastMRI Project seeks to minimize the length of an MRI exam without having to sacrifice quality. In this Project Spotlight blog, we got a chance to hear from the team on where the project currently stands and how the PHDA was able to play a role.

Please share a little about your background and your research experiences.

The FastMRI team is led by Vanessa Schmithorst, PhD, and MRI physicist at the UPMC Children’s Hospital of Pittsburgh, with years of experience in MR pulse sequence design, image reconstruction, data/statistical analysis, and MR clinical and research applications. The team also includes Ashok Panigrahy, MD, and John F. Caffey Endowed Chair in Pediatric Radiology at the University of Pittsburgh, who has years of pediatric neuroimaging experience and in translating magnetic resonance (MR) technology (including arterial spin labeling and spectroscopy) to clinical and research applications; and Rafael Ceschin, PhD, and assistant professor of pediatric radiology, who has expertise in Bayesian analyses, machine learning/deep neural networks, cross-platform software development, and its application to MR data.

What led you to the PHDA?

The team is new to the discovery of commercialization pathways for our patents and ideas. The PHDA helped them with learning how to navigate this process.

Walk us through your project, FastMRI.

At a high level, the goal of FastMRI is to reduce the length of an MRI exam without reducing the technology’s diagnostic value. Magnetic Resonance Imaging (MRI) is a non-invasive medical imaging technology used over the entire body, which generates excellent tissue contrast and is thus of great clinical diagnostic value. However, one drawback of an MRI is that exams can take a long time, sometimes over an hour. This can be frustrating, especially when patients have a hard time laying still (e.g., pediatric patients). A faster exam could lead to faster patient throughout and a more satisfying patient experience.

The team seeks to achieve this goal by reducing scanner “dead time” needed to create the image (part one) and by developing intelligent under-sampling and intelligent image reconstruction (part two). The results would be vendor agnostic, meaning the tool could be used on any MRI machine, regardless of the company that built the machine.

The first aspect of the team’s research is the development of multi-contrast sequences, which speed up the MR exam by limiting the total number of scan sequences necessary. Currently, a separate scan sequence is used for each desired image contrast (T1-weighted, T2-weighted, etc.), which results in time wastage. Optimizing multi-contrast sequences for specific combinations of MR contrast desired in neuro MR exams leads to a protocol-specific sequence or “prequence” for each clinical protocol (e.g., headache, stroke, head trauma).

The second aspect of the team’s research is the development of “intelligent” data under-sampling schemes based on the anatomy and contrast desired as well as the use of non-linear imaging gradients.  Utilizing non-linear imaging gradients and specially tailored excitation pulses better maps the image acquisition space onto the sparse image domain space; better mapping means fewer data points needed, as the conditioning of the transformation is improved. The team has a unique approach in that, via input from clinical radiologists, the intelligent under-sampling does not result in loss of clinically relevant diagnostic information. Incorporating prior knowledge in the image reconstruction results in a closed-form approximate solution which only requires a few iterations for optimization and still allows presentation of images to the clinician in real time. This is much faster than the typical compressed sensing technique for reconstruction, which requires a time-consuming iterative procedure often requiring hundreds of iterations.

 In what ways has UPMC played a role lending clinical expertise and sharing data?

The research team worked with UPMC to pull necessary data, test the technology, and obtain clinician feedback. The ability to leverage the system’s resources was very beneficial and allowed the team to validate the technology in a clinical setting.

What was most exciting or unexpected that you learned during the project?

The team learned that it is important to understand their consumer stakeholders’ needs so that they can craft their product accordingly. It was also exciting for the team to be able to share ideas and communicate with industry partners working on similar projects.

Did you encounter any roadblocks that you did not anticipate?

The team discovered new artifacts in developing their sequences which led them to more developmental work behind the scenes. There were also specific roadblocks related to the different vendor scanners which required developing robust harmonization tools. 

What are the biggest takeaways from your project?

To keep the project relevant, the team will need to stay on top of the new developments in rapid MR imaging and slightly pivot the project to address not only rapid imaging access for vulnerable critical care populations (neonates with newborn injury), but also diverse low-resource populations that might not have access to MRI technology.

What are your project’s next steps? 

Through this research project, the team was able to significantly reduce scan time on a sequence-by-sequence basis for the headache protocol, and the intelligent under-sampling proved to be successful; image quality was retained as was diagnostic utility. As next steps, the team plans to continue working on neuro protocols outside of the headache protocol. They are also developing collaborations with companies in the MR market with the hope of FastMRI becoming the clinical standard-of-care for routine MR protocols. The team is also in the process of submitting SBIR/STTR applications for additional project funds.

Subscribe to our newsletter to get the latest Alliance news, event invitations and updates, and behind-the-scenes spotlights on our researchers, projects, and team members.
Subscribe

By clicking Submit, I agree to the Pittsburgh Health Data Alliance's Terms of Use and Privacy Policy.