For the last two decades, the problem that has obsessed Miller is how to mathematically analyze the three-dimensional space of human anatomy. Much as Noam Chomsky and his colleagues at MIT sketched a universal framework for human grammar, Miller would like to create the simplest possible equations and statistical models that can describe human anatomy in all its variety and multitude. That quest has led him into collaborations with a huge range of scholars, including theoretical mathematicians, imaging engineers, radiologists, cardiologists, and neuroscientists.
At this early stage of his center’s development, Miller’s computers generally can’t tell doctors anything that they don’t already know. Enlargement of the left ventricle is associated with heart failure? Old news. A thinning of the brain’s white-matter tracts is a sign of dementia? Been there, done that. But after his team has digested many thousands of images, Miller hopes that the computers will begin to discover subtle anatomical markers and patterns that had previously gone unnoticed by scientists. And that, in turn, could lead to much earlier diagnostic tests for certain diseases-and even point the way toward new treatments.
In the most tantalizing result so far, Miller and his colleagues demonstrated last year a computational technique that can analyze magnetic resonance images of the brain to distinguish between older adults who will soon suffer Alzheimer’s dementia and those who are simply experiencing the normal memory loss that comes with aging.
At the very least, Miller says, techniques like that one have the potential to streamline radiologists’ work. “One of the things that we’d like to build is a machine that can see evidence of something and cue the radiologist, so that the radiologist can more efficiently review the images. We can save them time in their workflow.”
Miller’s ability to shift easily from complex mathematical topics to the practical problems of radiologists might help to explain his success in luring collaborators from across Hopkins into his projects. “Mike’s style is to encourage us and to stimulate us about what we can do together,” says Susumu Mori, a professor of radiology at the School of Medicine who is working with Miller on an analysis of brain scans of young patients. “He’s not the kind of colleague who says, ‘Just give me the data and I’ll do everything.’ He’s not like that at all.”
On this particular project, Miller, Mori, and their colleagues hope to identify physiological changes over time in the brain structures of young patients who are being treated for psychiatric illness. In and of itself, this is not revolutionary-many teams of scholars across the country are studying physical correlates of schizophrenia, for example. But Miller and Mori hope that their computational-imaging system will spot subtle changes that may not have been noticed by others, and that it will build a cumulative body of knowledge about schizophrenia and the growing brain.
“In clinical image-based analysis, the result is free text,” Mori says. That is, the radiologist simply writes a narrative description of what the image shows. “We’d like to supplement that with some quantitative measures,” Mori continues, “similar to what you see with a complete blood cell analysis.”
Miller’s graduate work at Hopkins in the 1970s had nothing directly to do with computational anatomy. He fell into the lab of Murray B. Sachs and Eric D. Young, who were working on pioneering studies of how the auditory nerve processes sounds. (Both Sachs and Young remain on the biomedical engineering faculty today.)
As a doctoral student, Miller helped to identify the “neural spike train”-that is, the series of electrical action potentials-generated by the auditory nerve as it conducts spoken language from the ear to the brain. “We were measuring the signals in the auditory nerve and rebuilding the code that was equivalent in information to the information that’s in these complex acoustic sounds-consonants and plosives and vowels,” Miller says. His doctoral work helped to verify a theoretical model of auditory processing that had been developed by linguists and cognitive scientists at MIT-and it also eventually helped to improve the technology of cochlear implants.
After completing his degree, Miller moved to Washington University to work as a postdoctoral researcher with Donald Snyder, a professor of electrical engineering who pioneered the use of the statistical technique known as “point processes” in analyzing neurons’ behavior. Miller had drawn on Snyder’s technique in his doctoral study, and wanted to work with the man himself. At Wash U, Miller was drawn into a network of scholars who were refining the technology of positron emission tomography (PET), which creates three-dimensional images of biological processes. “It was a big team,” says Snyder, a professor of electrical engineering at Wash U who often collaborated with Miller. “There were computer people, there were people in radiation physics who were developing the hardware. And then there were those of us who wanted to help analyze the images and understand their significance.