Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session A

Poster A6 - An Audio Based Approach to Early Alzheimer’s Detection Using Artificial Intelligence

Thursday, October 24, 2024

10:00 AM - 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Justin Keowen (he/him/his)

Student/Machine Learning Undergraduate Researcher
Louisiana State University - Alexandria Department of Mathematics and Computer Science
Baton Rouge, Louisiana, United States

Last Author(s)

Ahmed Shaffie, PhD (he/him/his)

Assistant Professor of Computer Science
Louisiana State University - Alexandria Department of Mathematics and Computer Science, United States

Introduction: As of 2024, 6.9 million Americans aged 65 and older are diagnosed with Alzheimer’s disease (AD), accounting for 10.9% of this age group, or one in nine individuals. Without including unpaid care, the cost of care for AD and other dementias is projected to reach $360 billion this year, with Medicare and Medicaid covering 64% ($231 billion) of this cost. As the elderly population grows, the number of people with AD is expected to rise. Currently, AD diagnosis involves physical and neurological tests, standardized neuropsychological assessments, laboratory biomarker tests, and imaging techniques such as MRI, CT, and PET scans. These assessments typically begin after the patient becomes aware of AD concerns and consults a specialist, by which time significant cognitive decline and brain grey matter degradation may have occurred. Therefore, there is a need for a system that can detect these concerns earlier, enabling timely intervention. We have developed a computer-aided diagnosis (CAD) system that detects AD by analyzing audio recordings.

Materials and

Methods: The dataset used is the ADReSSo 2020 audio recordings, which includes approximately 108 recordings of individuals being interviewed about the Cookie Theft Image. Approximately half of the recordings are from healthy controls, while the other half are from individuals diagnosed with AD. AD affects speech production, which can be quantified through acoustic features. Our system focuses on these features. The audio files were normalized for length and amplitude, and approximately 6,300 features were extracted. These features were ranked by relevance using a MultiSurf relief algorithm, and it was experimentally determined that 700 features were optimal for this classifier and feature set. The dataset was divided into an 80-20 split, with 80% used for training an XGBClassifier and 20% for testing.

Results, Conclusions, and Discussions: This system was tasked with classifying if the person in a recording has Alzheimer’s or not using the extracted ranked feature set and the specified classifier with optimal parameters. Through this process, it was found to have an accuracy of 83.8%, a sensitivity of 90.9%, and a specificity of 70.8%.
These results show the feasibility and efficacy of such a CAD system. A system such as this could in the future potentially be a first line of detection for Alzheimer’s disease diagnostics. Through regular analysis of patient voice recordings those in AD risk groups could be directed to seek medical help over AD concerns sooner thus possibly leading to better outcomes and a reduction in the overall care costs of this patient group. Moreover, envisioning further integration, this approach could seamlessly integrate with smart assistants, enhancing accessibility and continuous monitoring in home environments.
A CAD system to diagnose Alzheimer’s disease through audio recordings using acoustic features was developed. This system demonstrates the feasibility of such a system as well as potential for increased system efficacy. The CAD system can help detect Alzheimer’s in patients without doing more invasive, expensive, or time-consuming tests.

Acknowledgements (Optional):