Professor Wismuellers lab uses advanced pattern recognition and machine learning technology to explore and visualize huge files of medical data in novel ways. Computational radiology helps bridge the gap between fundamental engineering research and clinical practice.
Radiology of the Future
When people talk about Twitter as a source of Big Data, Axel Wismueller simply smiles.
Those 400 million tweets a day (as of June 2012) represent about 56 gigabytes of information.
By comparison, the daily production of biomedical images by a single radiology practice is approximately a terabyte, Wismueller says. (One terabyte equals 1,000 gigabytes.) And there are hundreds of radiology practices.
“So this is real Big Data,” observes Wismueller, professor of imaging sciences, biomedical engineering, and electrical and computer engineering.
This abundance of medical images is already creating bottlenecks in obtaining timely diagnoses.
Wismueller understands this all too well. Not only does he head a biomedical imaging research group at the University of Rochester, he is a practicing diagnostic radiologist at its Medical Center. When he began practicing in the 1990s, Wismueller says, he read 100 to 150 images a day. Now, wielding a mouse at his computer, he scrolls through 30,000 to 50,000.
Looking at such numbers “exceeds information-processing capabilities of the human brain and could eventually make it difficult for us to maintain the highest professional standards that we are committed to provide to our patients,” Wismueller says.
And the problem is only going to get worse, he believes. A surge in aging baby boomers is creating additional demand for biomedical images to detect and treat such diseases as Alzheimer’s, breast and prostate cancer, osteoarthritis, and osteoporosis. But the number of radiologists is stable or declining.
“The only remedy is computer-aided analysis of biomedical images,” Wismueller says. Indeed, he is convinced that computer-aided analysis will transform the field within 10 to 20 years.
Imagine, for example, that all those images being taken every day, along with all the related health reports, lab tests, and diagnoses, could be stored in an accessible database.
A radiologist confronted with a hard-to-diagnose condition—one of the interstitial lung diseases, for example—could use computational methods to “mine” the database for similar-looking images with similar associated test results for which specific diagnoses had already been determined.
It would help the radiologist narrow the range of possibilities and arrive at a speedier diagnosis for his own patient.
However, several hurdles must be overcome before computer-aided analysis of biomedical images becomes reality. Computer scientists face unresolved questions about how to extract, characterize, and classify all that data from biomedical images. Any subsequent changes in radiology procedures or patient care would have to prove their feasibility in carefully orchestrated, highly regulated clinical settings. And even then, there are the hurdles of government approvals and acceptance by health insurance companies.
Wismueller is not daunted by this. He quotes Chicago architect Daniel Burnham: “Make big plans; aim high in hope and work”—and keeps pushing the frontiers of engineering and medicine, seeking pioneering advances in “computational radiology.” His team is applying computational methods in ways that could make it easier to visualize huge files of data and in ways that could significantly advance the diagnosis and treatment of multiple sclerosis, interstitial lung disease, breast cancer, Alzheimer’s disease, HIV, and osteoporosis.
For example, Wismueller, with a grant jointly funded by NIH and the German government, is developing a computational framework that uses resting-state functional MR images to examine brain connectivity— how different parts of the brain “talk” to each other. The framework will be used to investigate how brain connectivity changes when antiretroviral therapy is begun for patients who suffer cognitive impairment as a result of HIV.
Another example: Wismueller and his team are applying methods that astrophysicists use to describe the distribution of galaxies to a much more confined space—the mesh-like trebacular structures inside bones. He and the team are converting biomedical images of bone tissue affected by osteoporosis into three-dimensional maps that can identify likely fracture sites and the amount of load these sites can bear. This could help monitor the progression of the disease in a patient and whether the fracture risk is sufficient to require more costly treatments.
Wismueller is convinced that, even as the tools of Big Data transform radiology, “the radiologist will remain at the center of what we are doing just as the patient will remain at the center of why we are doing it.”