2021 Virtual Undergraduate Research Symposium
Task Balanced Multimodal Feature Selection to Identify Biomarkers Associated With Alzheimer’s
PROJECT NUMBER: 36 | AUTHOR: Braedon O’Callaghan, Computer Science
MENTOR: Hua Wang, Computer Science
Alzheimer’s disease (AD) remains very poorly understood despite posing significant financial and social burdens on our society. It can be very difficult in many cases to diagnose patients with AD especially in the earlier stages of the condition. Our research aims to aid in the diagnosis of patients with AD using statistical learning techniques adapted to handle multi-modal medical data. The data used for our experiments was provided by ADNI and consists primarily of genetic, brain-imaging, and cognitive score data. Traditional learning techniques might miss out on patterns which depend on the type of data being analyzed. To help this method better analyze data with multiple modalities we introduced a novel objective function and optimization algorithm. The objective function, which I focus on in this presentation, is at the core of any statistical learning technique, and in this case utilizes a system of balanced multimodal feature selection. The method is balanced because it simultaneously trains a support vector machine (which learns from the data) and performs feature selection (which reduces the number of features considered), allowing for the identification of important biomarkers relevant to AD. Using this technique our experiments validated existing biomarkers and identified several new biomarkers which hadn’t previously been correlated with Alzheimer’s disease.
Braedon is a current Junior at Mines pursuing a computer science major and a math minor. Since his sophomore year, Braedon has been working with the MInDS@Mines research group with a primary focus on helping to diagnose Alzheimer’s using machine learning techniques. Outside of research and classes, Braedon likes to play jazz piano and rock climb.