Machine learning prediction of future amyloid beta positivity in amyloid-negative individuals

The pathophysiology of Alzheimer’s disease (AD) involves \(\beta\) $\beta$ -amyloid (A \(\beta\) $\beta$ ) accumulation. Early identification of individuals with abnormal \(\beta\) $\beta$ -amyloid levels is crucial, but A \(\beta\) $\beta$ quantification with positron emission tomography (PET) and cerebrospinal fluid (CSF) is invasive and expensive.

We propose a machine learning framework using standard non-invasive (MRI, demographics, APOE, neuropsychology) measures to predict future A \(\beta\) $\beta$ -positivity in A \(\beta\) $\beta$ -negative individuals. We separately study A \(\beta\) $\beta$ -positivity defined by PET and CSF.

Cross-validated AUC for 4-year A \(\beta\) $\beta$ conversion prediction was 0.78 for the CSF-based and 0.68 for the PET-based A \(\beta\) $\beta$ definitions. Although not trained for the clinical status-change prediction, the CSF-based model excelled in predicting future mild cognitive impairment (MCI)/dementia conversion in cognitively normal/MCI individuals (AUCs, respectively, 0.76 and 0.89 with a separate dataset).

Standard measures have potential in detecting future A \(\beta\) $\beta$ -positivity and assessing conversion risk, even in cognitively normal individuals. The CSF-based definition led to better predictions than the PET-based definition.