Novel Machine Learning Algorithm Predicts the Development of Alzheimer's Disease
Using MRI scans of cortical thickness, a new machine learning model compares atrophy patterns of a given individual to a group of Alzheimer's patients.
In Alzheimer's disease, increasing numbers of brain cells fail over time, and the symptoms progressively worsen. Even though Alzheimer's currently has no cure, early detection of the disease remains important so that patients can have greater opportunities to slow progression by trying different medications or participating in clinical trials.
Researchers from South Korea have developed a new method for assessing cortical atrophy patterns to predict the likelihood that a given patient will develop Alzheimer's disease. The novel method uses a machine learning algorithm to measure the similarity of an individual's cortical atrophy pattern with that of a representative group of Alzheimer's patients.
The work was published March 7 in the journal Scientific Reports.
The authors scanned the brains of 869 cognitively normal individuals and 473 patients with probable Alzheimer's disease dementia using high-resolution MRI with 3-D volumetric imaging. Laboratory testing -- such as complete blood counts and thyroid function exams -- excluded other causes of dementia in the patients. These subjects' cortical thickness data was analyzed to determine their cortical atrophy patterns, which were then entered into the machine learning model so it could learn the differences between the two groups.
In terms of distinguishing someone with Alzheimer's disease from a healthy individual, the algorithm had sensitivity and specificity values of 87.1 percent and 93.3 percent, respectively.
The researchers also validated the model in two longitudinal cohorts -- one with 79 patients with amnestic-mild cognitive impairment, a condition that can either convert to Alzheimer's disease or revert back to normal cognition, and another with 27 patients with probable Alzheimer’s disease dementia. For the first cohort, patients whose condition converted to Alzheimer's disease had a more Alzheimer's-like cortical atrophy pattern at baseline and after one year when compared to nonconverters. Analysis of the second cohort determined that Alzheimer’s disease patients with faster decline had higher atrophy similarity than those with slower decline. This association was observed at baseline, at a one-year follow-up, and at a three-year follow-up.