Office of Commercialization and Economic Development
Office of Technology Commercialization

Retrospective Artifact Correction of MRI through Unsupervised Deep Learning

Technology #20-0060

Questions about this technology? Ask a Technology Manager

Download Printable PDF

Researchers
Pew-Thian Yap
Managed By
Matthew Howe
Commercialization Manager 919.966.3929
Patent Protection
Provisional Patent Application Filed

Structural MRI is highly susceptible to motion artifacts, which can be difficult to avoid, especially when the subject must remain relatively motionless in a scanner, such as in pediatric patients or patients that cannot remain still (e.g. patient with Parkinson’s disease). Currently, deep learning based retrospective MRI artifact correction can be employed to improve image quality. Deep learning based systems may require supervised training of a neural network using paired data, e.g., images of a subject with no artifacts and images of the same subject with artifacts. However, acquiring a large amount of paired data for supervised training is impractical, since it is challenging to generate images displaying a wide range of artifacts. An unsupervised learning approach has been developed that can disentangle artifacts or imperfections (e.g. noise, streaking, and ghosting) from artifact-corrupted images to recover high-quality artifact-free counterparts. Importantly, due to the unsupervised nature of the method, it is not required that the artifacts are explicitly specified, which allows for more efficient generation of images than supervised deep learning techniques.