The goal is for any doctor to be able to upload a patient's MRI scan, then wait for the computer to spit out results as it searches for images in the databank with similar patterns and known diagnoses.
The databank, which has 7,000 brain images of Hopkins patients and counting, should be publicly available in three years, said Dr.
When the program goes global, any radiologist would be able to tap the expertise of the Hopkins specialists who have been expanding the databank, image by image. Details of each patient's medical condition are included, but names and other personal information are stripped out.
In addition to helping with diagnosis, the databank could help detect new diseases or better classify them, Huisman said. It could also shorten the time that even experienced doctors need to pinpoint rarely seen disorders. And when scans in the library show changes in brain structure early in a disease, it could give doctors a head start on treatment before the onset of clinical symptoms.
"The beauty of the databank is we have so many patients in there, and we have categorized different groups of diseases: trauma, infection, inflammation, tumors, metabolic diseases, malformations," he said. "This is a growing project. The more cases we have in there, the more it will become an automated process."
The researchers have begun building a similar library of MRI, or magnetic resonance imaging, scans that focuses on brain disorders commonly found in elderly patients. Hopkins neurology professor
Key to the databank is sophisticated software whose development has been led by
Miller likened it to Picasa and other photo-editing programs that use facial recognition technology -- but vastly more complicated to design. The structures in the human brain, he said, "are so much more complex than the nice beautiful, simple symmetry of the structure of your face."
The databank's software takes as many as a thousand structural measurements in 250 regions of the brain to create an anatomical index. Creation of the digital library is supported by a three-year
"This is really a brain cloud," said Miller, adding that he is not aware of any similar projects.
The technology comes from a field Miller pioneered in the 1990s. Called computational anatomy, it makes it possible to break down the human brain into a myriad of labels and create an index that can be used to compare images.
"A radiologist could take a scan and say, 'What other children have I seen whose scans look this way?' " he said. "That's a different question than saying, 'Show me all the scans of kids who have some abnormality.' That's a linguistic search of medical records."
Miller said the databank project represents an important transition from research to clinical practice. He has collaborated with Dr.
Huisman, who came to Hopkins from
Miller credited Huisman with recognizing early the value of adopting modern imaging protocols, routinely logging brain scans of children seen at Hopkins and indexing them by what the images show. Huisman favors MRI scans because, unlike CT scans, they don't expose children to radiation.
On a recent afternoon at the Children's Center, Huisman sat in a room with seven viewing stations, each equipped with several flat screens to view scans. He brought up the image of the brain of a youngster who went to Hopkins after the child began to lose cognitive function.
Huisman pointed to an area where the white matter was too bright and tiger-striped. "This means this brain is abnormal; something is going wrong," he said.
Thanks to his expertise, he could quickly tell the child has an inherited degenerative disorder, information that will help the treating physician.
But he said other radiologists aren't brain specialists and might not recognize such a condition from scans, particularly because many early-presenting cases don't match the classical textbook examples. That is where the databank can help.
"This allows those people ... to use the database to get basically the same kind of expertise for a child they're taking care of as if the child would be in a center focusing on this," Huisman said.
The larger the number of scans, he said, the more variability will be included for a given disease, adding that it is important to have "a big bulk of data" for each disease. He said he expects the project to be put online "in another two to three years," once the digital library reaches 10,000 to 15,000 images.
(c)2014 The Baltimore Sun
Visit The Baltimore Sun at www.baltimoresun.com
Distributed by MCT Information Services