Predicting amyloid risk by machine learning algorithms based on the A4 screen data: Application to the Japanese...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Identifying individuals at high risk for brain amyloid deposition is crucial for Alzheimer's disease (AD) prevention trials.
• Early detection of preclinical AD is essential for effective intervention.

Purpose of the Study:

• To develop and validate machine learning models for predicting amyloid positron emission tomography (PET) standard uptake value ratio (SUVr) in cognitively normal elderly individuals.
• To assess the feasibility of using these models for participant prioritization in Alzheimer's disease prevention studies.

Main Methods:

• Machine learning models were developed using data from the Anti-Amyloid Treatment in Asymptomatic Alzheimer's Disease study.
• Models were applied to 3081 participants in the Japanese Trial-Ready Cohort (J-TRC) webstudy.
• Predictive performance was evaluated using amyloid PET SUVr and confirmed with a subgroup of 37 participants with known amyloid status.

Main Results:

• Key predictors for amyloid deposition included age, family history, and cognitive test scores (Cognitive Function Instrument, CogState).
• The machine learning models demonstrated good performance in predicting amyloid status, with an area under the curve of 0.806 in the J-TRC subgroup.
• Predicted SUVr values correlated well with self-reported amyloid test results.

Conclusions:

• The developed algorithms show potential for automatically prioritizing individuals with higher amyloid risk from the J-TRC webstudy.
• This automated prioritization can enhance the efficiency of recruiting participants for in-person studies, accelerating the identification of preclinical AD cases.