Publication Type
PhD Dissertation
Version
publishedVersion
Publication Date
7-2025
Abstract
Dementia is a neurodegenerative disease with a prevalence rate expected to triple by 2050, posing a significant challenge for health services. To impede the increasing prevalence, medical professionals and scientists are actively investigating technology to detect cognitive decline at a reversible stage known as Mild Cognitive Impairment (MCI). Digital biomarker technology is an emerging pragmatic approach to permit objective, ecologically valid, and long-term continuous measurement of cognitive health status, rendering it as one of the promising technologies for early MCI detection. Despite its potential, it is nontrivial to encode, extract and combine predictive information from these digital biomarker technologies; advanced machine learning (ML) techniques are necessitated to handle the large volume of noisy data measurements as a consequence of long-term continuous monitoring in a naturalistic setting such as the spatiotemporal data, coupled with limited human sample sizes mainly due to ethical and financial reasons.
In this dissertation, the overall machine learning task for digital biomarker technology to detect MCI is organized into three research problems to be successively tackled. First, the knowledge gaps in the understanding of the state-of-the-art (SOTA) ML techniques employed for digital biomarker technologies to detect cognitive decline were addressed by conducting a systematic review and meta-analysis of relevant studies. Application Programming Interface (API)-assisted literature search was employed on major academic research databases including IEEE-Xplore to extract more than 60,000 articles for analysis. Next, the technical challenge of tackling the prevalent small subject sample size real-world data problem in digital biomarker studies was tackled via investigating a unique ML technique, namely the predictive self-organizing fuzzy neural network architecture based on the fusion Adaptive Resonance Theory (fusion ART). A unique Singapore cross-sectional study with 49 subjects measured using multiple sensors over two months was employed as the real-world data for the evaluation of MCI detection efficacy. Lastly, the problem of encoding, extracting and integrating multiple long-term, noisy continuous data, particularly from in-home spatiotemporal data, was studied. Specifically, the Episodic Memory Adaptive Resonance Theory (EM-ART) and SpatioTemporal Episodic Memory (STEM), which are variants of fusion ART, were employed to model movement trajectory and spatial time-series data of in-home room trips, respectively. A contrastive layer and a three-channel fusion ART model were sequentially stacked above the spatiotemporal model layer, providing the multimodal contrastive spatiotemporal machine learning model for maximizing predictive signals to detect MCI. A longitudinal real-world dataset collected from high-frequency passive infrared motion sensors paired with annual neuropsychological assessments over one year was subsequently employed to validate the predictive efficacy of the novel multimodal contrastive spatiotemporal machine learning model.
In summary, this thesis presents a review of the SOTA ML techniques and their predictive accuracy to detect geriatric disease using digital biomarkers, followed by the development and validation of a suite of novel ML techniques, particularly involving the fusion ART, that can push the boundary of predictive accuracy for in-home detection of MCI. Together, the findings from this dissertation pave new opportunities for more advanced ML architecture to be explored on digital biomarker technologies for a variety of geriatric disease detection in an ecologically valid environment. For instance, the Generalized Heterogenous Fusion ART (GHF-ART), a more sophisticated multimodal fusion technique possessing the capability to integrate heterogeneous types of data, can be particularly suited to be further developed and employed on the data collected from the in-home digital biomarker technologies, a field where a multitude of sensor technologies are increasingly explored for detection of different geriatric diseases such as MCI, frailty and cognitive frailty.
Keywords
Machine Learning, Self-Organizing Neural Network, Adaptive Resonance Theory, Mild Cognitive Impairment, Digital Biomarker
Degree Awarded
PhD in Computer Science
Discipline
Artificial Intelligence and Robotics | Health Information Technology
Supervisor(s)
TAN, Ah Hwee
First Page
1
Last Page
152
Publisher
Singapore Management University
City or Country
Singapore
Citation
THE, Seng Khoon.
Machine learning for digital biomarker-based detection of cognitive decline. (2025). 1-152.
Available at: https://ink.library.smu.edu.sg/etd_coll/793
Copyright Owner and License
Author
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.