Data Science Tools Can Help Boost Speed, Quality of MRI Reconstruction

Published in the Proceedings of the National Academy of Sciences of the United States of America (PNAS), new research findings from the University of Minnesota indicate that the use of data science tools can enhance traditional methods of magnetic resonance imaging (MRI) reconstruction, resulting in image reconstruction on par with that achieved through deep learning.

In general, traditional MRIs are time-consuming. Although this is a factor that can be worked around, there is a great deal of clinical interest in minimizing the issue.

Using data science tools and ideas gleaned from deep-learning methods, researchers from the University of Minnesota have discovered a way to fine-tune a traditional MRI reconstruction method called compressed sensing that can speed up the MRI process.

“MRIs take a long time because you’re acquiring the data in a sequential manner,” explained Mehmet Akçakaya, PhD, an associate professor in the U of M College of Science and Engineering and senior author of the paper, in a press release. “We want to make MRIs faster so that patients are there for shorter times and so that we can increase the efficiency in the healthcare system.”

The fine-tuned compressing sensing method is nearly as high-quality as deep learning, the researchers said.

Deep learning, a subset of machine learning, is increasingly being used to speed up MRI image reconstruction. The process includes eliminating certain frequencies and using a deep-learning algorithm to perform predictions and fill in the gaps of the missed frequencies.

However, researchers also noted that the use of machine learning might result in biases, skewing results. Thus, improving traditional methods to reach the same levels of success as deep learning could be a viable alternative.

“There’s a lot of hype surrounding deep learning in MRIs, but maybe that gap between new and traditional methods isn’t as big as previously reported,” continued Akçakaya. “We found that if you tune the classical methods, they can perform very well. So, maybe we should go back and look at the classical methods and see if we can get better results. There is a lot of great research surrounding deep learning as well, but we’re trying to look at both sides of the picture to see where we can find the best performance, theoretical guarantees and stability.”

Despite its potential drawbacks, deep-learning methods are increasingly being applied in clinical care.

For example, recent research shows that a deep learning tool trained to identify and classify brain tumors based on standard MRIs may help neuroradiologists diagnose patients.

Further, different types of deep learning can be used to improve clinic operations and provide clinical decision support.