Research Profiles: Robotics Institute part 2

An insight into Yanxi Liu’s Alzheimer’s disease research

Saravana Sivasankaran Feb 13, 2006

The Medical Robotics Technology Center (MRTC) and the Vision & Autonomous Systems Center (VASC) at Carnegie Mellon have been collaborating on what promises to be an innovative solution to a widely demonstrated problem in aging people. Under the leadership of Associate Research Professor Yanxi Liu, a project titled “Predicting Risk of Alzheimer’s Disease From Shape Features” is trying to use computer vision techniques to help early detection of Alzheimer’s disease.

So what is this disease whose famed victims include Ronald Reagan, Charlton Heston, and Rosa Parks? Alzheimer’s disease (AD) is a neurodegenerative disease, commonly causing dementia. It is characterized by progressive intellectual deterioration, declining daily activities, and behavioral changes. Memory loss, usually manifested as minor forgetfulness, is the most noticeable symptom. Amnesia progresses steadily as the illness does, although older memories can remain relatively well-preserved.

Age is the most important risk factor for AD. The chance of developing Alzheimer’s doubles every five years for people over age 65. Experts estimate that the number of Americans with Alzheimer’s disease will continue to grow, and by 2050 the number of individuals with Alzheimer’s could range from 11.3 million to 16 million, according to www.alz.org.

Unfortunately, AD is not easily detected until late in its progression. The drugs to slow progression might be more effective if they were administered early. Brain scans can be used to confirm diagnosis, but they are unreliable in the early stages, though later they will show significant, clear loss of brain tissue. Detecting AD before symptoms start to appear is the holy grail of the research community.

“The basic idea is very simple,” explained Liu. “We are developing computer algorithms to search for discriminative features in the MR images of normal elderly brains versus Alzheimer’s-disease brains. In collaboration with the NIH Alzheimer’s Disease Research Center at UPMC, we were given a set of MR images containing patients with AD diagnosis ... and corresponding normal controls. We carried out a sequence of experiments to map the MR structural images to different types of image features, search for discriminative subspaces, and validate the results using cross-validation.”

Carnegie Mellon is not the first to carry out research in this field. There are currently many projects funded by the government and the Alzeimer’s Association to facilitate research into how artificial intelligence can aid neurologists.

“The main difference from other projects of this kind perhaps is the aggressiveness in our treatment of types and dimension of the feature spaces we have looked at, ranging from 3D volume and image features to features extracted from deformation fields that are generated by deforming each subject’s MR image to a reference neuroimage, and the tensor fields,” said Liu.

Researchers use quantified brain asymmetry for different feature types and under different dimensions. To deal with such a large volume of features, which range in the millions, they also have to develop effective feature screening and feature selection tools to make the task computationally feasible. They are also taking full advantage of the computer’s capability to search through the whole 3D brain for discriminative regions that may or may not be indicated in the literature.

Liu and her team have already made significant breakthroughs. Recently, they showed that by using image features around the hippocampus, it was possible to find discriminative feature subsets for AD, mild cognitive impairment, and normal controls.

“Our immediate next step is to use a large, longitudinal MR image dataset with rich collateral information, to learn those features from the MR images of a normal subject in hope of predicting his or her AD tendency as early as possible,” explained Liu.

Using computer vision and machine learning to predict AD is not meant to replace current scanning methods. Instead, it would work to complement existing PET, CT, and MRI scans.

“In 2004, scientists at UPMC made a breakthrough for AD research by completing the first human study of a compound that, through positron emission tomography, enables them to peer into the brains of people to see the telltale plaque deposits which are believed to be at the root of the disease,” Liu said. “We are in collaboration with the UPMC group led by Dr. Julie Price to learn how the structural, as well as the functional, MRI and PET imaging modalities can complement each other. It is our belief that multi-modality imaging and image analysis will lead to more effective, safe, and economical technology for AD diagnosis of the future.”

The need for research like Liu’s cannot be stressed enough. Finding a treatment that could delay onset by five years could reduce the number of individuals with Alzheimer’s disease by nearly 50 percent after 50 years, according to www.alz.org. A report commissioned by the Alzheimer’s Association states that Alzheimer’s disease costs American business $61 billion a year. Of that figure, $24.6 billion covers Alzheimer health care and $36.5 billion covers costs related to caregivers of individuals with Alzheimer’s, including lost productivity, absenteeism, and worker replacement. With that in mind, this could easily qualify as one of Carnegie Mellon’s most important contributions to the medical world.

Besides working in the Robotics Institute, Liu is associated with the Center for Automated Learning and Discovery, part of SCS, and she is an adjunct associate professor in the radiology department at the University of Pittsburgh. The Alzheimer’s disease project has been funded by grants from ADRC of UPMC, National Institutes of Health, and the Health Department of Pennsylvania.