Explainable AI-based Alzheimer's prediction and management using multimodal data.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Dementia, including Alzheimer's disease, is a leading cause of death and disability globally.
• Current machine learning models for Alzheimer's diagnosis lack trust due to their black-box nature and often rely solely on neuroimaging data.
• There is a need for improved, explainable diagnostic tools for Alzheimer's disease.

Purpose of the Study:

• To propose a novel, explainable Alzheimer's disease prediction model using a multimodal dataset.
• To address the limitations of existing models by incorporating clinical, MRI segmentation, and psychological data.
• To advance the understanding of multimodal five-class classification for Alzheimer's disease.

Main Methods:

• Utilized nine popular machine learning models, including Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM), for five-class classification.
• Performed data-level fusion of clinical, MRI segmentation, and psychological data.
• Employed SHapley Additive exPlanation (SHAP) for model explainability.

Main Results:

• The Random Forest classifier achieved a 10-fold cross-validation accuracy of 98.81% for classifying Alzheimer's disease, cognitively normal, non-Alzheimer's dementia, uncertain dementia, and others.
• Explainable AI (XAI) using SHAP provided insights into the prediction reasoning.
• The study is the first to present a multimodal five-class classification of Alzheimer's disease using the OASIS-3 dataset.

Conclusions:

• The developed explainable AI model demonstrates high accuracy and reliability in Alzheimer's disease prediction.
• Multimodal data fusion significantly enhances diagnostic capabilities for Alzheimer's disease.
• A novel Alzheimer's patient management architecture was proposed, offering potential for improved patient care.