Discriminating Aging Cognitive Decline Spectrum Using PET and Magnetic Resonance Image Features

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Population aging is increasing the incidence of neurodegenerative diseases like Alzheimer's disease (AD).
• Early detection of individuals at risk for cognitive decline is vital for maintaining quality of life.
• Brain imaging techniques, including PET and MRI, provide key biomarkers for assessing cognitive decline risk.

Purpose of the Study:

• To investigate ensemble models for classifying individuals across the cognitive decline spectrum.
• To combine features from single and multiple imaging modalities (FDG+AMY+MRI, PET ensemble).
• To evaluate the performance of various machine learning algorithms in discriminating cognitive decline stages.

Main Methods:

• Utilized imaging data from 131 individuals across four cognitive assessment groups.
• Employed leave-one-out cross-validation with Decision Tree, Random Forest (RF), LGBM, and Categorical Boosting (CAT) algorithms.
• Assessed model performance using balanced accuracy, enhanced by Shapley Additive exPlanations with Recursive Feature Elimination (SHAP-RFECV).

Main Results:

• Feature selection using CAT or RF algorithms demonstrated superior performance in discriminating the early cognitive decline spectrum.
• Magnetic Resonance Imaging (MRI) features were predominantly utilized for effective discrimination.
• The combination of CAT or RF algorithms with SHAP-RFECV showed robust discrimination of early aging cognitive decline stages.

Conclusions:

• CAT and RF algorithms, coupled with SHAP-RFECV, effectively discriminate early stages of cognitive decline, primarily leveraging MRI features.
• Further research is needed to explore the correlation between selected brain regions and the cognitive decline spectrum.