A Cornell-led collaboration used machine learning to identify the most accurate ways and timeframes to predict the course of Alzheimer’s disease in cognitively normal or mildly cognitively impaired people.
Modeling has shown that it is easier and more accurate to predict future dementia decline in people with mild cognitive impairment than in cognitively normal or asymptomatic people. At the same time, the researchers found that predictions for cognitively normal subjects are less accurate for longer time horizons, but for people with mild cognitive impairment the reverse is true.
Modeling also demonstrated that magnetic resonance imaging (MRI) is a useful prognostic tool for people in both stages, while tools that track molecular biomarkers, such as positron emission tomography (PET), are more useful for people with mild cognitive impairment.
The team diary, “Machine learning-based multimodal prediction of future decline in Alzheimer’s disease: an empirical study», published on November 16 in PLOS ONE. The lead author is Batuhan Karaman, a PhD student in the field of electrical and computer engineering.
Alzheimer’s disease can take years, sometimes decades, to progress before a person shows symptoms. Once diagnosed, some people decline rapidly, but others may live with mild symptoms for years, making it difficult to predict the rate of disease progression.
“When we can say with certainty that a person has dementia, it is too late. Much damage has already been done to the brain, and it is irreversible damage,” said the lead author. Mert Sabuncuassociate professor of electrical and computer engineering at the College of Engineering and of electrical engineering in radiology at Weill Cornell Medicine.
“We really need to be able to catch Alzheimer’s disease early on,” Sabuncu said, “and be able to tell who is going to progress fast and who is going to progress slower, so we can stratify the different groups. at risk and be able to deploy whatever treatment options we have.
Clinicians often focus on a single “time horizon” – usually three or five years – to predict the progression of Alzheimer’s disease in a patient. The delay may seem arbitrary, according to Sabuncu, whose lab specializes in analyzing biomedical data — particularly imaging data, with a focus on neuroscience and neurology.
Sabuncu and Karaman teamed up with longtime collaborator and co-author Elizabeth Mormino of Stanford University to use neural network machine learning that could analyze five years of data on people who were cognitively normal or had mild cognitive impairment. The data, captured in a study by the Alzheimer’s Disease Neuroimaging Initiative, encompassed everything from an individual’s genetic history to PET and MRI scans.
“What we were really interested in is, can we look at this data and tell if a person is going to progress in the coming years?” said Sabuncu. “And more importantly, can we do a better job of predicting when we combine all the tracking data points we have on individual subjects?”
The researchers found several notable patterns. For example, predicting that a person will go from asymptomatic to mildly symptomatic is much easier for a time horizon of one year, compared to five years. However, predicting whether a person will progress from mild cognitive impairment to Alzheimer’s dementia is more accurate over a longer period, with the “sweet spot” being around four years.
“That might tell us something about the underlying mechanism of the disease and how it evolves over time, but that’s something we haven’t probed yet,” Sabuncu said.
Regarding the effectiveness of different types of data, modeling has shown that MRI scans are more informative for asymptomatic cases and are particularly useful in predicting whether someone will develop symptoms over the next three years, but less useful for predicting people with mild cognitive impairment. . Once a patient has developed mild cognitive impairment, PET scans, which measure certain molecular markers such as amyloid and tau proteins, appear to be more effective.
One of the advantages of the machine learning approach is that the neural networks are flexible enough to be able to operate despite missing data, such as patients who may have skipped an MRI or PET scan.
In future work, Sabuncu plans to further modify the modeling so that it can process full imaging or genomic data, rather than just summary measurements, to harvest more information that will improve predictive accuracy.
The research was supported by the National Institutes of Health, National Library of Medicine and National Institute on Aging, and the National Science Foundation.
Many physicians and scientists at Weill Cornell Medicine maintain relationships and collaborate with external organizations to foster scientific innovation and provide expert advice. The institution makes this information public for the sake of transparency. For this information, see the profile of Dr Sabuncu.