AIB-6 The Use of Auditory Data and the Role of Machine Learning as a Diagnostic Tool for Early-Onset Alzheimer’s Disease
Abstract
Alzheimer's Disease (AD) is a chronic neurodegenerative disorder characterized by irreversible cognitive decline and memory loss. Symptoms of AD can develop years before clear clinical manifestation, leading many patients to start treatment later. Early detection leading to early treatment is crucial to slow the disease's progression and improve patient outcomes. Current diagnostic treatments are often time-consuming and typically can only be performed years after the disease's onset. A way to measure this deficiency is by detecting specific patterns in speech, such as repetition of words, duration of words, vocal frequencies, and articulation, since language impairment is a common symptom of AD.
Machine learning (ML) techniques have emerged as promising non-invasive tools by using vocal pattern detection to provide a more accurate diagnosis for early-onset Alzheimer's Disease. ML algorithms offer the ability to analyze large and complex datasets capable of integrating multiple sources of information, such as clinical, genetic, neuroimaging, and auditory data. ML shows great promise in the early detection of AD. However, there are some challenges, including the need for extensive and diverse datasets, knowledge of how to interpret different ML models, and integration of multiple data sources into global databases to be shared with multiple investigative teams and clinicians.
The overall project focuses on auditory data from the DementiaBank repository (https://dementia.talkbank.org/access), which was used to train the ML algorithm during some preliminary analysis. Our focus was the preprocessing stage to ensure that the file formats could be transformed into a format the algorithm could use. The ML model used was able to identify patterns, extract meaningful features, and then apply them to predict future outcomes. Through the use of ML and with future collaboration from medical professionals, we expect to have a trained program with at least 80% accuracy. With our current results, we plan to integrate other data files, including imaging, into the pool of datasets to ensure a more robust system for accurate and precise diagnosis.
Keywords
Alzheimer's Disease; Artificial Intelligence; Machine Learning
AIB-6 The Use of Auditory Data and the Role of Machine Learning as a Diagnostic Tool for Early-Onset Alzheimer’s Disease
University Readiness Center Greatroom
Alzheimer's Disease (AD) is a chronic neurodegenerative disorder characterized by irreversible cognitive decline and memory loss. Symptoms of AD can develop years before clear clinical manifestation, leading many patients to start treatment later. Early detection leading to early treatment is crucial to slow the disease's progression and improve patient outcomes. Current diagnostic treatments are often time-consuming and typically can only be performed years after the disease's onset. A way to measure this deficiency is by detecting specific patterns in speech, such as repetition of words, duration of words, vocal frequencies, and articulation, since language impairment is a common symptom of AD.
Machine learning (ML) techniques have emerged as promising non-invasive tools by using vocal pattern detection to provide a more accurate diagnosis for early-onset Alzheimer's Disease. ML algorithms offer the ability to analyze large and complex datasets capable of integrating multiple sources of information, such as clinical, genetic, neuroimaging, and auditory data. ML shows great promise in the early detection of AD. However, there are some challenges, including the need for extensive and diverse datasets, knowledge of how to interpret different ML models, and integration of multiple data sources into global databases to be shared with multiple investigative teams and clinicians.
The overall project focuses on auditory data from the DementiaBank repository (https://dementia.talkbank.org/access), which was used to train the ML algorithm during some preliminary analysis. Our focus was the preprocessing stage to ensure that the file formats could be transformed into a format the algorithm could use. The ML model used was able to identify patterns, extract meaningful features, and then apply them to predict future outcomes. Through the use of ML and with future collaboration from medical professionals, we expect to have a trained program with at least 80% accuracy. With our current results, we plan to integrate other data files, including imaging, into the pool of datasets to ensure a more robust system for accurate and precise diagnosis.