AI Corrects Motion Artifacts in MRI Scans
MRI scans are known for their high-quality images of soft tissues, unlike X-rays or CT scans. However, these scans are extremely sensitive to motion, which can result in image artifacts. These artifacts can lead to misdiagnoses and inappropriate treatments when crucial details are hidden from the physician. The good news is that MIT researchers have developed a deep learning model that can correct motion in brain MRI scans.
Nalini Singh, a PhD student affiliated with the Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic) at MIT, explains that motion is a common problem in MRI. MRI scans take time, ranging from a few minutes to an hour, and even the smallest movements can severely affect the image. Unlike camera imaging, which results in localized blurs when there is motion, MRI images can be entirely corrupted. Patients may try measures to minimize motion, like being anesthetized or restricting deep breathing. However, these methods are not always feasible, especially for children and patients with psychiatric disorders.
Motion Correction with Deep Learning
The research team at MIT received recognition for their paper titled “Data Consistent Deep Rigid MRI Motion Correction,” winning the best oral presentation award at the Medical Imaging with Deep Learning conference (MIDL) in Nashville, Tennessee. Their approach involves creating a motion-free image from the motion-corrupted data using a combination of physics-based modeling and deep learning. Singh states that their goal was to harness the strengths of both methods and ensure consistency between the image output and the actual measurements. This is crucial to prevent the creation of “hallucinations” – images that may look real but are physically and spatially inaccurate. By addressing motion artifacts, this method can improve diagnoses and patient outcomes.
Eliminating motion artifacts in MRI scans is particularly important for patients with neurological disorders that cause involuntary movement, such as Alzheimer’s or Parkinson’s disease. According to a study by the University of Washington Department of Radiology, motion affects around 15 percent of brain MRIs. The need for repeated scans or imaging sessions that produce high-quality images further increases hospital expenditures, estimated to be around $115,000 per scanner annually.
In the future, the research team aims to expand their work to include more sophisticated motion correction techniques for different parts of the body. For example, fetal MRI faces challenges due to rapid and unpredictable motion, which cannot be easily modeled using simple translations and rotations.
Dr. Daniel Moyer, an assistant professor at Vanderbilt University, sees the potential of this research impacting various clinical cases, including those involving restless children or older patients, pathologies that induce motion, studies of moving tissue, and even imaging healthy patients who may move during the procedure. He believes that these methods are the next step in MRI motion correction and could become standard practice in the future.
The co-authors of the paper include Nalini Singh, Neel Dey, Malte Hoffmann, Bruce Fischl, Elfar Adalsteinsson, Robert Frost, Adrian Dalca, and Polina Golland. The research received support from GE Healthcare, the Massachusetts Life Sciences Center, and various federal agencies and initiatives.