Identifying incipient dementia individuals using machine learning and amyloid imaging.

Area of Science:

• Neuroscience
• Biomarkers
• Machine Learning

Background:

• Alzheimer's disease (AD) clinical trials require accurate prediction of dementia progression.
• Amyloid-β protein is key in AD, but neuronal degeneration biomarkers better predict short-term clinical progression.
• Mild cognitive impairment (MCI) patient populations in trials often have imbalanced stable vs. progressive cases.

Purpose of the Study:

• To develop a machine learning (ML) probabilistic method for predicting progression to Alzheimer's dementia within 24 months.
• To utilize regional information from a single amyloid positron emission tomography (PET) scan for prediction.
• To address the challenge of imbalanced data in short-term MCI progression prediction.

Main Methods:

• A novel ML-based probabilistic algorithm was developed.
• The algorithm analyzes regional information from amyloid PET scans.
• The method was designed to handle imbalanced datasets of stable and progressive MCI.

Main Results:

• The algorithm achieved 84% accuracy in predicting dementia progression within 24 months.
• It obtained an area under the receiver operating characteristic curve (AUC) of 0.91.
• Performance surpassed existing algorithms using similar biomarkers and prior multi-modal biomarker studies.

Conclusions:

• The developed ML algorithm accurately predicts short-term progression to Alzheimer's dementia using amyloid PET scans.
• This tool offers significant potential for enriching clinical trial populations for disease-modifying therapies.
• The method provides a valuable approach for identifying individuals likely to benefit from early intervention in AD.