Comprehensive Summary
This article describes a multimodal DI-based classification framework that improves early Alzheimer’s disease (AD) detection and predicts whether individuals with mild cognitive impairment (MCI) will progress to Alzheimer’s. Previous models, such as structural MRI, fMRI, and sequencing, don’t have complete multimodal datasets. To mitigate this problem, a cycle generative adversarial network is created to synthesize missing modalities in space. With this, the model can analyze incomplete patient data without sacrificing accuracy. This model was paired with real modalities in a classifier to distinguish among individuals with cognitive normalcy, Alzheimer's disease, and mild cognitive impairment. When evaluated against Alzheimer's disease and cognitively normal individuals, the model achieved an accuracy of 71% and a precision of 0.558. Additionally, it placed heavy weight on structural atrophy in the hippocampus and amygdala that disrupts cognitive control networks and clusters on chromosomes 1,7,11,14,15, and 19, which align with Alzheimer's pathology. The model's generative deep learning framework allowed it to learn biologically plausible patterns, thereby supporting the validity and accuracy of the approach.
Outcomes and Implications
This research is significant because it advances the detection of Alzheimer's disease, improving diagnostic accuracy and patient management. The high accuracy and sensitivity of the model, even with missing imaging or genetic data, make it more aligned with real-world clinical conditions and more applicable to patients than past research and models. Since many patients don’t undergo all imaging modalities or genomic tests, a framework that can reconstruct and utilize missing data to detect disease allows physicians to diagnose more efficiently. By achieving over 70% accuracy, the model identified MCI patients likely to progress to Alzheimer’s, which is very important for future treatment planning, lifestyle interventions, and enrollment in early-stage clinical trials. This helps combine computational modeling with biological understanding to build clinical trust that is essential for real-world adoption. With further validation, the framework can become a meaningful tool to guide individualized risk assessment and enable early detection and prevention, rather than just late-stage management.