Utilizing Machine learning and AI for early detection, and identification of Alzheimer's Disease and Related Dementias

Objectives of the proposed research:
Alzheimer’s disease and its related dementias (ADRD) are a growing public health crisis with no known cure. ADRD diagnosis remains challenging due to its inherent heterogeneity, variability of early symptoms, overlap and possibility of combined etiologies. The overarching objective of this project to develop novel interpretable deep learning methods to integrate multimodal neuroimaging, genomic and clinical datasets collected from demographically diverse cohorts for early detection of ADRD.
Study design:
Our central hypothesis is that an integrative deep learning model that operate on both MRI data, and NGS data will be an ideal solution for isolating ADRD biomarkers that can lead to early diagnosis. We aim to integrate multi-modal datasets including genomic, clinical, demographic, neuroimaging from diverse populations in studies such as ADNI, ADSP, and HEBLA and develop a multimodal fusion-based approach for optimized and accurate classification of various presentations of ADRD. Further, we will design and develop an interpretability-framework that would help explain the decisions made by deep-learning models leading to knowledge discovery for neuroscientists, and transparent analysis for clinicians.
Analysis plan:
We plan to utilize genetic variants as features in our deep learning model. To this end, we will perform GWAS to find SNPs that associate with disease diagnosis and endophenotypes such as neuroimaging, biospecimen measures, and cognitive performance. We will utilize genetic data to perform a network propagation study to discovery AD-associated genes.

Alzheimer’s disease and related dementias (ADRD) are a growing health crisis, with current global management costs over $350 billion and cases expected to grow to 14 million globally by 2050. Emerging evidence of advantages of early detection of the disorder, and the current limitation of the clinical practices suggest that more quantitative methods are needed to identify, and early diagnose ADRD. In this project, we aim to develop interpretable deep learning models to integrate genetic, imaging, and clinical datasets for diagnosis and early prediction of ADRD.