Home - Brain - New Statistical Method to Predict Alzheimer’s Using Imaging Data
Research Brief

New Statistical Method to Predict Alzheimer’s Using Imaging Data

Principal component analysis of fluorodeoxyglucose positron emission tomography data reveals worrisome brain metabolic pattern in patients with mild cognitive impairment.

By
Janelle Weaver, Contributor
Wednesday, July 25, 2018

SHARE: 

Predicting whether an individual with mild cognitive impairment is likely to develop Alzheimer’s dementia is important for timely interventions. One imaging tool being evaluated for this purpose is fluorodeoxyglucose (FDG)-positron emission tomography (PET), but its predictive value is currently under debate. Some researchers have proposed applying principal component analysis (PCA) to FDG-PET for diagnosing neurodegenerative disorders.

PCA is a standard statistical technique for reducing the dimensionality of large, complex datasets, making them easier to interpret while minimizing information loss. A recent study showed that applying PCA to FDG-PET data is useful for predicting the conversion from mild cognitive impairment to Alzheimer’s dementia, especially when combined with clinical variables. The researchers analyzed FDG-PET scans of 544 patients with mild cognitive impairment using the Alzheimer's Disease Neuroimaging Initiative database. They applied PCA to a training dataset of 272 patients to identify brain metabolic patterns associated with the conversion from mild cognitive impairment to Alzheimer’s dementia.

The researchers then constructed prognostic models with significant predictors and validated the models using the test dataset of the remaining 272 patients. They also analyzed relevant clinical variables such as scores on the Mini-Mental State Examination and APOE4 status -- a genetic risk factor for Alzheimer’s disease. In the test data set, the best prediction power was achieved with a model that combined clinical biomarkers with the brain metabolic pattern identified by PCA.

The findings were presented June 26 at the Society of Nuclear Medicine and Molecular Imaging annual meeting held in Philadelphia, Pennsylvania.