Novel Alzheimer's Disease Stating Based on Comorbidities-Informed Graph Neural Networks.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Alzheimer's Disease (AD) is the leading cause of dementia, necessitating early detection for effective intervention.
• The Alzheimer's Disease Neuroimaging Initiative (ADNI) provides valuable data for AD research.
• Accurate multi-class classification of cognitive states (Cognitive Normal, Mild Cognitive Impairment, Alzheimer's Disease) is crucial.

Purpose of the Study:

• To develop and evaluate Graph Neural Network (GNN) models for multi-class Alzheimer's Disease (AD) classification.
• To investigate the impact of incorporating comorbidity data from electronic health records into GNN models.
• To compare the performance of GNN models with attention mechanisms against existing state-of-the-art techniques.

Main Methods:

• Constructed a patient-clinical graph network to model relationships between cognitive normal (CN), mild cognitive impairment (MCI), and AD patients.
• Trained various GNN-based prediction models using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Integrated comorbidity data from electronic health records as features within the GNN framework.

Main Results:

• The GNN model incorporating comorbidity data achieved the highest classification performance.
• The GNN model with attention mechanisms demonstrated superior results compared to state-of-the-art methods.
• Achieved high accuracy (0.92), AUC (0.96), and F1-score (0.92) in multi-class AD classification.

Conclusions:

• Comorbidity data significantly enhances the accuracy of Alzheimer's Disease classification models.
• GNNs, particularly with attention mechanisms, show great promise for early AD detection and prediction.
• This approach can potentially deepen the understanding of Alzheimer's Disease progression and contributing factors.