The technology behind MR (magnetic resonance) imaging has origins in research conducted by Nikolai Tesla more than 130 years ago. Since that time, scientists have been adding to the research, slowly developing what would become the MRI machine throughout the 20th century. Today, more than 22,000 MRI units exist across the world,1 but the evolution of this technology has not slowed down. If anything, computing capabilities have actually sped it up.
MR imaging, when paired with an experienced radiologist, is currently capable of dramatic, life-saving results. From cancer to diabetes, these machines are being used as part of proactive diagnostic and treatment planning processes because they provide a safe, non-invasive way for medical providers to see details inside a patient's body. But the incorporation of machine learning and artificial intelligence may drive MR imaging to even newer heights.
Artificial intelligence and reduced MR imaging times
One of the longstanding drawbacks of MR imaging is how long the machines take to complete a scan. While MR imaging times have dropped over the years, it can still take 10 to 60 minutes on average. Sometimes, patients have to return almost immediately for a second scan because the machine didn't capture the appropriate images required for diagnostics.2
It sounds like a small inconvenience in the larger picture of serious illness or injury — and it is. But that doesn't mean the timetable related to a scan doesn't pose a challenge, especially for some patients. Scans may require individuals to hold very still for a long time, all while in a loud, tight space. The ability to reduce scan time would improve patient comfort while also allowing radiology departments to meet greater productivity demands without impacting quality.
This is an area in which AI is showing great promise. In fact, a research team published findings in 2018 after developing an automated reconstruction process that might be able to substantially reduce the time it takes MRI machines to build accurate, appropriate images. The research was funded by the National Institutes of Health and resulted in a technology called AUTOMAP.
Researchers noted that the AI-based system delivered a better signal-to-noise ratio, was more accurate and was faster at necessary MRI adjustments than the current standard manual process.
The eventual outcomes of this research might mean faster MRI procedures, immediate image feedback during any type or procedure (which is not always available now) and faster access to diagnostics.3
At the forefront of AI-based research in this space is an effort to use machine learning's ability to leverage enormous banks of data to reduce the impact of under-sampling on MRI. Currently, MRI scans take so long because images that are acquired quickly are "noisy" and grainy. They usually can't be read accurately by the human eye. But by coupling huge data banks and machine learning, "smart" scans may be able to reconstruct better images in less time.4
Machine learning and diagnostic accuracy
However, AI isn't just important to the speed of the scan itself. It's playing an increasing role in diagnostics. Deep learning is currently able to approximate human accuracy in diagnosing a number of conditions and may be even more accurate within specific scenarios.5
It's not that AI is going to replace radiologists. Machine learning is capable of drawing on tremendous banks of knowledge, including previous scans and outcomes. It will always beat out a human mind in this capacity. But the human brain, especially those trained in radiology, is able to approach every issue from different and creative angles. It's when you combine these tools that MR imaging is likely to take an enormous step into the future.
AI is going to make it more possible for radiologists and others to come to the right diagnosis or treatment plan faster than ever, reducing patient wait times, enhancing hospital resource management and supporting more positive outcomes.
1. "Featured History: Magnetic resonance imaging." Department of Radiology, University of Washington. Web. 21 December 2018. <https://rad.washington.edu/blog/featured-history-magnetic-resonance-imaging/>.
2. "Magnetic Resonance Imaging (MRI)." Inside Radiology. 21 December 2018. <https://www.insideradiology.com.au/mri/>.
3. "Artificial Intelligence enhances MRI scans." NIH Research Matters. 21 December 2018. <https://www.nih.gov/news-events/nih-research-matters/artificial-intelligence-enhances-mri-scans>.
4. Hugh Harvey. "Can AI enable a 10 Minute MRI?" Torwards Data Science. 21 December 2018. <https://towardsdatascience.com/can-ai-enable-a-10-minute-mri-77218f0121fe>.
5. Koichiro Yasaka. "Deep learning and artificial intelligence in radiology: Current applications and future directions." PLOS. 30 November 2018. 21 December 2018. <https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1002707>