An interpretable generative multimodal neuroimaging-genomics framework for decoding Alzheimer's disease

The framework successfully identifies Alzheimer's disease with 92.6% accuracy by combining brain imaging (MRI) and genetic data (SNPs), demonstrating that multimodal approaches integrating different biological data types can improve disease detection compared to single-data approaches.

The system handles incomplete medical data, a common real-world problem in healthcare, by using generative AI to fill in missing brain scans or genetic information—making the approach more practical for clinical settings where patients may not have complete test results.

Explainable AI methods revealed which brain regions matter most, showing that gray matter changes in specific cortical and subcortical areas drive the model's predictions, enabling doctors to understand why the system made a diagnosis rather than treating it as a "black box."

The model can predict which MCI patients will progress to Alzheimer's with 71.1% accuracy, potentially allowing earlier intervention for high-risk patients before symptoms worsen, though this performance is notably lower than disease detection and suggests this application needs further development.

The interpretability analysis aligned with established neuroscience, as the identified brain regions and network impairments matched known Alzheimer's pathology, providing validation that the AI system learned medically meaningful patterns rather than spurious correlations.