Alzheimer’s Disease (AD) and AD-related dementias (ADRD) are a growing public health crisis, affecting about 6 million Americans, and this number is expected to increase to 14 million by 2050. The cost of managing AD is close to $250 billion in the US, and heavily impacts the US healthcare system and disproportionality affects minority populations. Hispanic and African Americans are expected to experience greatest increase in AD/ADRD, with estimated 7 million Hispanics who might suffer from AD/ADRD by 2060. Mild Cognitive Impairment (MCI) is a condition of decline in memory or thinking skills compared to healthy control of same age. Current clinical screening techniques do not provide a mechanism to classify MCI to its two main subgroups: stable MCI (sMCI), which does not progress to AD for at least three years, and progressive MCI (pMCI), which progresses to AD in three years or sooner. Distinguishing biomarkers specific to sMCI from pMCI may be the key to early detection of AD. With advances in biotechnology, numerous multimodal clinical and biological datasets (e.g., neuroimaging, genomic, clinical, and multi-omics) related to AD have been generated, albeit with limited racial diversity. Our preliminary experiments demonstrate that machine learning (ML) models trained using these homogeneous data, when confronted with datasets from multiracial and multiethnic populations exhibit decreased accuracy/sensitivity. We assert that advanced ML models that utilize multimodal imaging and genomic/clinical data are essential for identification and early detection of AD. Our proposal’s overall objective is to design and develop robust, interpretable, and generalizable ML models that can operate on multiple data modalities to distinguish between pMCI and sMCI leading to early prediction of AD and novel biomarkers for AD diagnosis across diverse demographic populations. To attain this objective, we will preprocess datasets from multiple data repositories of AD to generate a large and diverse dataset, and design and develop deep learning (DL) and self-learning models using this harmonized data (when available) to predict diagnosis and progression of AD across diverse demographic populations. Through interpreting these DL models, we aim to discover novel neuroimaging, genomic, and clinical biomarkers. Our robust methods thereby will reduce the health disparities among racial and ethnic minorities, and underserved populations. This new line of investigation is significant since it has the potential to improve on long-stalled effort to increase accuracy and reliability of identifying biomarkers to predict AD while reducing the health disparities. The expected outcome of this work will have a positive impact because these proposed research tasks will lay the groundwork to develop a new class of ML models, and will provide rapid, high-throughput, sensitive, reproducible, and reliable tools for early detection of AD and discovery of biomarkers applicable to individuals from diverse demographic populations.

Methods

Method development for this work is ongoing.