A multi-modal graph-based framework for Alzheimer's disease detection.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Medical Imaging Analysis

Background:

• Alzheimer's disease (AD) diagnosis is complex, requiring integration of diverse data types like medical scans, genetic information, and cognitive tests.
• Existing methods often struggle with multimodal data integration and handling missing data, limiting diagnostic accuracy.

Purpose of the Study:

• To develop a flexible and scalable Machine Learning (ML) framework for Alzheimer's disease (AD) detection using a compositional, graph-based approach.
• To enable end-to-end deep learning (DL) predictors by representing datasets as nodes and ML models as edges in a directed computational graph.

Main Methods:

• A compositional graph-based ML framework was developed, where datasets are nodes and DL models are edges.
• The framework supports forward and backpropagation for data transformation, model finetuning, and saliency map computation.
• A graph with 11 data nodes and 14 ML model edges was constructed for AD prediction using multimodal data.

Main Results:

• The framework successfully integrated diverse data types, including MRI scans, genetic data, and cognitive tests for AD prediction.
• The modular approach demonstrated robustness in handling distribution shifts and missing modalities.
• The system achieved accurate Alzheimer's disease prediction, showcasing its adaptability and scalability.

Conclusions:

• The proposed compositional graph-based ML framework offers a powerful and adaptive tool for Alzheimer's disease diagnosis.
• This approach effectively integrates multimodal data and handles missing information, advancing medical prediction tasks.
• The framework's modularity and scalability make it suitable for complex medical prediction challenges beyond AD.