AN INTERPRETABLE GENERATIVE MULTIMODAL NEUROIMAGING-GENOMICS FRAMEWORK FOR DECODING ALZHEIMER'S DISEASE.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) is the leading cause of dementia, with Mild Cognitive Impairment (MCI) representing a transitional stage.
• Accurate detection of AD and prediction of MCI progression are crucial for timely intervention.
• Multimodal data integration and advanced computational methods are needed to address the complexity of AD and MCI.

Purpose of the Study:

• To develop and validate a multimodal deep learning framework for classifying Alzheimer's disease (AD) patients versus healthy controls.
• To detect individuals with Mild Cognitive Impairment (MCI) who are likely to convert to AD.
• To identify structural and functional brain modulations and genetic factors associated with AD and MCI conversion using an interpretable AI approach.

Main Methods:

• A multimodal deep learning framework integrating MRI and Single Nucleotide Polymorphisms (SNPs) was proposed.
• A generative module using Cycle Generative Adversarial Networks (CycleGANs) was employed for imputing missing data in multimodal datasets.
• Explainable AI (XAI) methods were utilized for feature relevance extraction and enhancing model interpretability.

Main Results:

• The framework achieved high accuracy in AD detection (0.926 ± 0.02) and MCI conversion prediction (0.711 ± 0.01).
• Interpretability analysis highlighted gray matter alterations in brain regions associated with AD.
• Impairments in resting-state networks and genetic associations with endocytosis, amyloid-beta, and cholesterol pathways were identified.

Conclusions:

• The proposed integrative and interpretable deep learning approach demonstrates significant potential for AD detection and MCI prediction.
• The study provides valuable biological insights into the mechanisms underlying AD progression.
• This methodology offers a promising avenue for advancing neurodegenerative disease research and clinical applications.