Researchers at Worcester Polytechnic Institute (WPI) have used artificial intelligence to analyze brain anatomy and predict Alzheimer's disease with an accuracy rate of nearly 93 percent. Their findings, published in the journal Neuroscience, show that loss of brain volume related to Alzheimer's varies by age and sex.
Alzheimer's disease is a neurodegenerative disorder that damages neurons and leads to cognitive decline and death. In the United States, about 6.9 million people aged 65 or older are living with this condition.
The WPI team, including PhD student Senbao Lu and master's graduate Bhaavin Jogeshwar, analyzed MRI scans from the Alzheimer's Disease Neuroimaging Initiative, which contains images from individuals aged 69 to 84. These scans represent people with normal cognition, mild cognitive impairment, and diagnosed Alzheimer's disease.
To manage the large amount of data in MRI images, researchers first applied machine learning techniques to measure volumes in 95 regions across 815 scans. They then used an algorithm to compare these measurements between healthy subjects and those showing signs of cognitive decline or Alzheimer's.
Their approach was able to distinguish Alzheimer's cases from normal brains and those with mild impairment with a reported accuracy of 92.87 percent.
Among all groups studied, volume reduction in three specific brain areas—the hippocampus, amygdala, and entorhinal cortex—was most closely linked to Alzheimer’s disease. The hippocampus is involved in memory and learning; the amygdala controls emotions; while the entorhinal cortex plays a role in memory and perception.
Both men and women aged 69 to 76 showed early loss of volume in the right hippocampus—a finding that may be important for early diagnosis. "The critical challenge in this research is to build a generalizable machine-learning model that captures the difference between healthy brains and brains from people with mild cognitive impairment or Alzheimer's disease," said Nephew. "A generalizable model means that the biomarkers we found are not unique to this dataset but could be universal to all patients with mild cognitive impairment or Alzheimer's."
The study also identified differences based on sex: females experienced more volume loss in the left middle temporal cortex (linked to language and visual processing), while males showed greater loss in the right entorhinal cortex. Nephew noted that these differences might relate to hormonal changes associated with aging.
Nephew’s group plans further research using deep learning models and examining other health factors such as diabetes that could affect brain health. The project has drawn interest from students across several disciplines at WPI.
"This research exemplifies the strength of neuroscience at WPI, which is interdisciplinary and computational," Nephew said. "The brain is an extremely complicated organ, and we need to think broadly about how to better understand, predict, and treat the diseases that afflict the brain."
