Researchers from MIT have developed a deep learning model that they hope could correct motion-corrupted MRI scans, with the aim of reducing risk of misdiagnosis or “inappropriate treatment when critical details are obscured from the physician”, along with reducing costs produced by the need for repeated scanning.
MIT notes that due to sensitivity of MRI, “small movements can have dramatic effects on the resulting image” which can be a particular challenge with “populations particularly susceptible to motion, including children and patients with psychiatric disorders”.
The researchers have developed a deep neural network capable of producing a reconstruction of a scan from two inputs – motion parameters and corrupted k-space data (representing “the spatial frequency information in two or three dimensions of an object”). The team trained the network using simulated, motion-corrupted k-space data that has been generated from known motion parameters, which “computationally constructs a motion-free image from motion-corrupted data without changing anything about the scanning procedure.”
Nalini Singh, lead author and Harvard-MIT PhD student, says that the aim was to “combine physics-based modelling and deep learning to get the best of both worlds.”
MIT highlights that along with the potential to improve patient outcomes, correcting motion-corrupted MRI scans could also benefit organisations financially. A study from the University of Washington Department of Radiology estimated that “motion affects 15 percent of brain MRIs”, with researchers equating this to the cost of this to “approximately $115,000 in hospital expenditures per scanner on an annual basis”.
Daniel Moyer, assistant professor at Vanderbilt University, said: “Not only is it excellent research work, but I believe these methods will be used in all kinds of clinical cases: children and older folks who can’t sit still in the scanner, pathologies which induce motion, studies of moving tissue, even healthy patients will move in the magnet.”
In other news on machine learning, this week we covered how researchers at the Moorfields Eye Hospital and the UCL Institute of Ophthalmology have used artificial intelligence to help identify “markers that indicate the presence of Parkinson’s disease in patients on average seven years before clinical presentation”.
The FDA also recently released a discussion paper designed to generate discussion about artificial intelligence and machine learning in drug development and manufacturing, noting that they have “the potential to accelerate the drug development process and make clinical trials safer and more efficient”.
