Google for Neurologists
Javed Mostafa, Ph.D., is an information junkie. As a professor with joint appointments in Information Science and the Biomedical Research & Imaging Center at UNC-Chapel Hill, Mostafa likes to explore innovative ways to dig up data, documents and images.
His latest project is a computer-based tool that could help physicians analyze the brain changes associated with Alzheimer’s disease. The program searches for brain scans with similar patterns of damage to give doctors a basis of comparison when diagnosing, treating and predicting the course of each patient’s disease. Think Google for neurologists.
Javed Mostafa, Ph.D., is an information junkie. As a professor with joint appointments in Information Science and the Biomedical Research & Imaging Center at UNC-Chapel Hill, Mostafa likes to explore innovative ways to dig up data, documents and images.
His latest project is a computer-based tool that could help physicians analyze the brain changes associated with Alzheimer’s disease. The program searches for brain scans with similar patterns of damage to give doctors a basis of comparison when diagnosing, treating and predicting the course of each patient’s disease. Think Google for neurologists.
The standard approach employed by neurologists doesn’t take such a “cohort” of similar patients into account. Rather, physicians typically address brain scans one patient at a time for atrophy or loss of brain cells. Mostafa’s system – called Viewfinder Medicine – tries to change that paradigm by giving physicians an automated system that could guide them in decision-making with regard to their patients.
As one neurologist, Daniel Kaufer, M.D., sees it, the realm of Alzheimer’s disease diagnosis could use the help. Kaufer, director of the Memory and Cognitive Disorders Program at UNC-Chapel Hill, says that MRIs are primarily used to rule out strokes or tumors or other structural lesions. The changes associated with Alzheimer’s are rather subtle, and, as a result, it can be difficult to make a reliable diagnosis based on the MRI alone.
“Currently there are limitations with regard to how much a brain scan can predict the clinical picture,” Kaufer said. “These automated systems are trying to improve the ability to take what is a very subjective process and make it more objective.”
The Viewfinder Medicine system presents neurologists with relevant images on the computer screen in a pattern known as a “fisheye.” By placing a patient’s brain scan in the center or “eye” of the screen, the physician can prompt the system to retrieve a cohort of similar patients, which it scatters around that center focal point. Importantly, it uses not just brain scans but also age group, gender, cognitive scores and other supporting data to determine the best matches for the patient.
“I am not a physician, I am an information retrieval guy, so my interest in this project is to study how both visual images and textual information can be combined to make searching an effective process,” explained Mostafa, who also serves as one of the leaders of the NC Translational and Clinical Sciences Institute (NC TraCS). “This is a domain where it is very important to be accurate. The more clues about a patient you can use to determine their diagnosis and treatment, the better.”
Mostafa was first drawn into this domain by the ADNI (Alzheimer’s Disease Neuroimaging Initiative), an NIH-funded project that has developed a database of images from 200 elderly controls, 400 subjects with mild cognitive impairment and 200 subjects with Alzheimer’s disease. Such comprehensive databases are hard to come by, so when Mostafa heard of ADNI he was eager to sign up for access to the collection.
“In medical science it is very hard to find image collections that can support a research project like this,” said Mostafa. “I could do it on my own, but that would mean I would have to get IRB approval, access MRI scans, make sure they are consistent, then I would have to store it somewhere, maintain it, manage it, make sure that the cognitive score data is available. Everything needs to be curated; this notion of having good data for research becomes a huge barrier for a researcher. There is a reason people call this kind of data gold standard -- it is because they are expensive to produce!”
Mostafa’s research was also aided by funding from NC TraCS, UNC’s home of the NIH Clinical and Translational Science Awards (CTSA). The national consortium was created to make biomedical research faster, cheaper and more efficient. Mostafa says having NC TraCS, with its grants support and programs like Carolina KickStart, which helps commercialize and move discoveries into the marketplace for potentially widespread use, has been key to his success.
“I couldn’t say enough positive things about them,” he said. “I am a pragmatic type of a scientist, if my work just got published and cited, I wouldn’t consider that my most valuable contribution. It has to get out to the world, and someone, hopefully someone, will find it useful. NC TraCS has been fabulous about making that happen.”
Translating scientific advances into health care improvements is a passion for Mostafa, who presented the vfM system at the 9th International Workshop on Content-Based Multimedia Indexing held in June in Madrid, Spain. He and his co-author Mayank Agarwal found that the “classification performance” of their computer-based tool matched the best result reported in the medical imaging literature, with up to 87% of patients correctly classified in their respective groups of normal, mild cognitive impairment or Alzheimer’s disease. Mostafa and Agarwal, who recently graduated with a doctorate from the School of Information and Library Science, are now making modifications to the system in hopes of getting its accuracy closer to the 100% mark.
“We are not sure if physicians are going to like this way of analyzing, but we hope they do,” said Mostafa. “For us, this is not an image analysis project; it is an image retrieval project. We didn’t invent any new image analysis algorithms or software, we invented a new way of searching and browsing MRI images and analyzing those images from a physicians’ perspective. The goal is to help create a tool that helps the physician to understand and analyze the information.”
The tool could prove particularly useful to physicians who are not just focused on memory disorders but who see different types of patients every day. Kaufer thinks that automated programs like Viewfinder Medicine could improve the care of such patients who aren’t seen at a specialized clinic.
“My advantage is I can look at the brain scan and, based on experience of looking at thousands of them, I can extract subtleties that help inform my differential diagnosis,” said Kaufer the neurologist. “These automated programs could help operationalize that expertise so that it would be more widely available.”
Mostafa believes automation could also lead to earlier detection of Alzheimer’s disease. Though there is currently no cure and few effective treatments for the illness, diagnosing it early in its progression could give physicians a better chance of testing new interventions and buying patients more time. Mostafa hopes to develop his application so that it could enable physicians to stage a patient’s disease on image data alone, without having to wait for the results of cognitive testing. But Kaufer thinks that may be overreaching a bit.
“There is so much overlap between groups, it makes it difficult for these imaging techniques to accurately classify a single subject every time,” said Kaufer. “I see people with terrible memory whose brains look normal, and other people whose brains look pretty atrophic, but their memory is still preserved. We are never going to get to a point where pure imaging can detect a diagnosis, but I do think that in the next five years, quantitative brain imaging methods are going to play a larger role in helping to facilitate the diagnosis of dementia in conjunction with other clinical factors.”
Mostafa, who with the support of NC TraCS recently co-founded a company concentrating on patient-centric decision support (Keona Health), thinks this latest project has commercial potential. But first the researchers will have to test the merits of the system with doctors in the Alzheimer’s domain, like Kaufer, who says he is more than willing to serve as a guinea pig.
