Jenae Vancura


2019-11-04 08:01:47 (GMT) This past year during a shadowing experience in neurosurgery, a family came in for a consult regarding lesions that were discovered in their son’s brain stem. To everyones surprise, when we sat down to look at the follow-up MRI from that morning, the lesions were gone.


And then we understood- those lesions, the ones that likely kept the parents awake at night, terrified, weren’t lesions at all. Those “lesions” on the first scan were simply MRI artifacts, not anatomically present, but computer generated.


My solution is a program called ArtIfact, which would aid doctors in flagging artifacts that may be mistaken for pathology. It would use machine learning to do two things- first identifying abnormal anatomical structures on an MRI scan. This function would be done through exposing the program to data sets of regular anatomical scans that would allow the program to learn which images are different enough to constitute an “abnormal structure”.


The second thing ArtIfact would do is take a scan that has been identified as having an atypical structure, and compare that structure to the rest of the image, identifying other areas of the image that are similar in size and shape. Hopefully after the machine learning has progressed enough, ArtIfact will be able to flag a potential artifact and the original structure it may have been generated from.


This program is not meant to replace pathologist or be the final word in determining if something is an artifact, especially when there are many potential artifacts generated from MRIs that are easily identifiable by a physician. ArtIfact is intended to act as simply an aid in identifying potential artifacts that may be mistaken as pathologies, so that no person has to hear they are living with a disease that isn’t really there.