Machine learning on structural brain scans identifies healthy individuals at risk of Alzheimer's
What is it about?
Here we apply machine learning techniques over magnetic resonance images (MRIs) of the brain of healthy individuals to predict who is harboring abnormal amyloid levels. The method has been trained and tested on two independent cohorts using cerebrospinal fluid levels of amyloid as gold-standard. Predictive capacity is modest (AUC=0.76), but used as a pre-screening tool, it has a notable impact since can cut down to half the burden to detect healthy individuals at risk of Alzheimer's.
Why is it important?
Healthy individuals harboring amyloid protein in the brain are at increased risk of developping Alzheimer's and could benefit from secondary preventive interventions. However, gold-standard techniques for amyloid are not suitable for screening the general population. Here, we present a method that, used as a pre-screening tool, cuts more than half the burden of innecessary tests to detect these individuals. This method comes at no-extra cost as it capitalizes brain scans that need to be acquired anyway for safety reasons.