How to Fix Motion-Corrupted MRI Scans using Deep Learning and Physics

An MRI (magnetic resonance imaging) scan is a test that creates clear images of the structures inside your body using a large magnet, radio waves, and a computer. Healthcare providers use MRIs to evaluate, diagnose, and monitor several medical conditions.

While X-rays and CT scans have their merits, MRI scans offer superior soft tissue contrast and high-quality imaging. While delivering exceptional soft tissue contrast and high-quality imaging, MRI remains susceptible to motion interference, where even slight movements can introduce disruptive image artifacts. These artifacts, which mess up the accuracy of medical images, can mess up how doctors figure out what’s wrong with a patient. This can lead to treatments that aren’t as good because the doctors might miss important details.

Even brief scans can be compromised by minor movements, which uniquely impact MRI images. Unlike camera blurs, MRI motion artifacts can distort entire images.

As indicated by a University of Washington Radiology study, approximately 15 percent of brain MRI scans are impacted by motion, necessitating additional scans. This requirement for repeat imaging contributes to an annual expense of about $115,000 per scanner within hospitals, aimed at obtaining diagnostically reliable images across various MRI modalities.

To fix this problem, the researchers at MIT have taken a significant step forward by harnessing the power of deep learning technology. They used deep learning to find a solution. They mixed deep learning with physics and discovered amazing results.

Their method involves computationally constructing a motion-free image from motion-corrupted data without changing the scanning procedure. The significance of adopting this integrated approach is rooted in its ability to maintain coherence between the resulting images and the factual measurements of the subject matter. 

Failing to achieve this alignment could lead the model to generate what is referred to as “hallucinations” ? seemingly genuine images that, in reality, deviate from the actual physical and spatial attributes. Such discrepancies can potentially alter diagnostic outcomes, underscoring the critical importance of accurate representation in medical imaging.

Looking forward, they highlighted the exciting potential for future studies to delve into more complex forms of head movement and motion affecting various body regions. For instance, in fetal MRI, the challenge lies in coping with rapid and unpredictable action, which goes beyond the capabilities of basic translation and rotation models. This underscores the need to develop more sophisticated strategies that account for intricate motion patterns, offering a promising avenue for enhancing MRI applications across diverse anatomical scenarios.