Dec 19, 2016 | Atlanta, GA
MRI machines are powerful tools used widely in conventional medicine, including tracking the progression of mild cognitive impairment (MCI) and early Alzheimer’s disease (AD).
However, the machines do not always produce error-free results. Additionally, the high cost of data acquisition on these machines prohibits enough data collection for analysis of these errors.
Kamran Paynabar, ISyE Assistant Professor, and Chitta Ranjan, an ISyE Ph.D. candidate in statistics, worked with collaborators at Harvard Medical School to determine the measurement error rate for MRI machines and its effects on patient diagnosis.
The team used data from the Alzheimer’s Disease Neuroimaging Initiative database, which included information for 741 patients, most of whom were 70 years or older. This information is concerned with observations of patients’ brains to assess for MCI and early AD over a period of up to four years. They were specifically interested in the decline of the hippocampal volume — a brain structure associated particularly with long-term memory.
Such longitudinal data reveals whether cognitive ability of patients “is getting worse over time or it is stable. There often exist some measurement errors associated with MRI systems, and we show that these errors, if not detected and decoupled, may mask a significant declining trend in the cognitive ability of Alzheimer’s patients,” Paynabar explained.
Paynabar, Ranjan, and their collaborators developed a new method to determine the MRI machine errors through what is called “repeatability analysis”: The measurement for a patient is taken over time, and the collected data is put together. If a patient could undergo an MRI with the process repeated in short intervals (for example, every hour or so), it would be easy to determine the measurement error, because the procedure is repeated.
However, because of cost and patient safety considerations, MRIs are not administered more than once a month.
The challenge for the team was how to determine the measurement errors without repeatability. In response, the team developed a statistical model, specifically a model based on mixed-effect regression integrated with a newly developed EM-variogram technique, to determine the measurement error and decouple it from the overall model error.
The statistical model — which was validated in the experimental testing phase of the study — when applied to MRI data created increased sensitivity, allowing for earlier diagnosis and treatment of cognitive disease. As per government studies, AD is the fifth major cause of death for senior citizens in the U.S. Early treatment of these patients will lead to longer and healthier lifetimes for thousands of people.
Further, Paynabar said, the statistical model has applications beyond Alzheimer’s disease. It is not limited to MRIs but can be applied to “any longitudinal data studies with unrepeated measures.” It can be used for patients with brain tumors and other types of cancers and can also be used in drug testing, quality inspection, and other health care applications where limited data is available due to various practical constraints.
Click on image(s) to view larger version(s)
Stewart School of Industrial & Systems Engineering