Using Machine Learning to Predict Dementia from Neuropsychiatric Symptom and Neuroimaging Data

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Machine learning (ML) shows promise for predicting mild cognitive impairment (MCI) and dementia.
• Neuropsychiatric symptoms (NPS) have been underutilized in ML models for cognitive decline prediction.

Purpose of the Study:

• To determine if baseline mild behavioral impairment (MBI) status and brain morphology predict future diagnoses in individuals with normal cognition (NC) or MCI.
• To assess the prognostic utility of NPS in conjunction with neuroimaging for ML-based diagnostic prediction.

Main Methods:

• Utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• Transformed Neuropsychiatric Inventory (NPI) items into MBI domains.
• Employed a logistic model tree classifier with information gain for feature selection to predict diagnosis.

Main Results:

• A binary classification model (NC vs. MCI/AD) achieved 84.4% accuracy using only MBI total score and left hippocampal volume.
• A three-class model (NC vs. MCI vs. dementia) reached 58.8% accuracy with seven features.
• MBI total score, impulse dyscontrol, and affective dysregulation were key predictors.

Conclusions:

• Baseline NPS, particularly MBI domains, offer prognostic value for predicting diagnostic changes.
• Integrating NPS with brain morphology enhances ML model performance for cognitive decline prediction.
• MBI scores are significant predictors of future cognitive status.