High resolution MRI scans of the brain can take around thirty minutes to perform, but in the case of a stroke this can be much too long to wait. Typically, if MRI is used, a stroke patient is rushed through so that fewer imaging slices are taken, resulting in a much lower quality image. Compared to high end scientific studies that produce imaging slices around a millimeter apart, a quick scan can have the slices spaced up to seven millimeters from each other. At this resolution, many of the automated computer vision algorithms that help to understand the images fail to work, and precise diagnosis is a serious challenge. Researchers at MIT working with clinicians at Massachusetts General Hospital have been working on using artificial intelligence techniques to be able to use high resolution scans of different patients taken previously to significantly improve the image quality of MRI scans of incoming stroke victims.
The technique relies on filling in the space between the scanned slices so that an algorithm that has studied large numbers of comparable high quality scans confirms that the generated image looks similar. The data from the original image and the generated data are kept separate so that various measurements can be always compared against the actual scan.
Following up on this, the team will apply their algorithm on 4,000 previously obtained low quality scans of stroke patients from twelve hospitals. Using the higher resolution images they will attempt to study the anatomy of strokes, some of which has remained blurred due to the concerns and limitations when dealing with stroke patients.
The research is being presented next week at the Information Processing in Medical Imaging conference at Appalachian State University in Boone, North Carolina.