Deep learning-based Alzheimer's disease detection using magnetic resonance imaging and gene expression data.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Genomics

Background:

• Alzheimer's disease (AD) presents a global healthcare challenge, necessitating early and accurate diagnosis for effective management.
• Artificial intelligence (AI) offers potential for multi-modality diagnostic models using neuroimaging and genomics, but interpretability remains a challenge.
• Developing interpretable AI models for AD diagnosis is crucial for advancing patient care.

Purpose of the Study:

• To build a comprehensive, interpretable AI-based diagnostic model for Alzheimer's disease (AD).
• To integrate magnetic resonance imaging (MRI) and gene expression data for enhanced AD detection.
• To improve early intervention and personalized treatment strategies for AD patients.

Main Methods:

• Utilized MobileNet V3 and EfficientNet B7 for gene expression feature extraction.
• Developed a hybrid TWIN-Performer model for MRI feature extraction.
• Employed attention-based feature fusion and an ensemble classifier (CatBoost, XGBoost, ERT) for AD identification.
• Integrated Shapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The proposed multi-modality AI model demonstrated superior performance on diverse datasets.
• Achieved consistently high Area Under the Receiver Operating Characteristic (AUROC) scores above 0.85.
• SHAP values enhanced model interpretability, facilitating understanding of diagnostic features.

Conclusions:

• The developed AI model shows significant promise for accurate and interpretable Alzheimer's disease diagnosis.
• Multi-modal data integration and advanced AI techniques can overcome diagnostic challenges.
• Improved interpretability supports earlier interventions and personalized treatment plans for AD.