Reduction of recruitment costs in preclinical Alzheimer’s Disease trials: machine learning prescreening for brain amyloidosis from magnetic resonance imaging

Alzheimer’s disease (AD) is a major global health threat by the international medical community with a rapidly accelerating worldwide prevalence, with more than 131 million cases expected by 2050. Without accurate diagnostic tools or therapies that effectively diagnose and stop or reverse the disease, deeper research in these fields remains challenging.

Given the recent string of failures of clinical trials in patients, the industry is shifting towards prevention. AD has a long preclinical stage, characterized by the presence of abnormal amyloid-beta (A) levels with preserved cognition. This stage constitutes a window to test preventive interventions. Still, AD prevention is severely limited at the recruitment stage. Gold-standard techniques for in vivo determination of A are cerebrospinal fluid (CSF) and positron emission tomography (PET). These techniques are unsuited for the screening of the general population given their invasiveness and cost.

We have developed a family of machine learning algorithms on data derived from structural magnetic resonance scans (MRI) that can reduce costs (47%) and participant burden (60%) to detect A-positive cognitively unimpaired individuals.

This technology is not intended to replace gold-standard techniques but as a triaging (pre-screening) method. Since participants scheduled for a PET scan or CSF test must undergo the acquisition of an MRI scan for safety reasons, the marginal cost of this technology is almost zero.