Amyloid-β prediction machine learning model using source-based morphometry across neurocognitive disorders.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Machine learning (ML) models using magnetic resonance imaging (MRI) have been explored for Alzheimer's disease (AD) prediction.
• Limited research has focused on diverse patient populations for these predictive models.
• Accurate prediction of amyloid-beta (Aβ) deposition is crucial for early diagnosis and intervention in neurodegenerative diseases.

Purpose of the Study:

• To develop a clinically useful ML model for predicting Aβ deposition using source-based morphometry (SBM).
• To assess the impact of combining different features (MRI, cognitive tests, apolipoprotein E status) on predictive accuracy.
• To evaluate the model's performance across a diverse cohort including AD, other neurological disorders, psychiatric conditions, and healthy controls.

Main Methods:

• Utilized structural MRI data from 118 participants with various neurological, psychiatric diagnoses, and healthy controls.
• Employed independent component analysis (ICA) within SBM to derive data-driven features from voxel-based gray matter images.
• Used a support vector machine (SVM) classifier and SHapley Additive exPlanations (SHAP) for model interpretability and accountability.

Main Results:

• The comprehensive model integrating MRI, cognitive tests, and apolipoprotein E status achieved 89.8% accuracy (AUC 0.888).
• An MRI-only model showed 84.7% accuracy, highlighting the added value of clinical data.
• The model accurately detected Aβ-positivity in non-AD patients and predicted Aβ-positivity across diverse disorders with moderate-to-high accuracy.

Conclusions:

• An MRI-based, data-driven ML approach, incorporating clinical data, can effectively predict Aβ deposition across a broad spectrum of neurological and psychiatric conditions.
• Source-based morphometry identified a specific gray matter pattern associated with AD, contributing significantly to predictive accuracy.
• This ML model shows promise as a valuable diagnostic aid for identifying Aβ pathology, potentially improving early detection and patient management.