PhD Thesis Defense: Yunda Liu, Advancing Objective Assessment of Physical and Behavioral Phenotypes Using Wearable and Mobile Technologies
Content
Speaker
Abstract
Neurological disorders and brain injuries are often accompanied by physical and behavioral impairments that significantly limit patients’ ability to perform activities of daily living and maintain social connections. Traditional assessments of these symptoms typically rely on subjective clinical observations, which are constrained by the experience of individual assessors. Although patient-reported outcomes provide a more accessible alternative, they are prone to recall bias and personal interpretation, which can compromise the reliability of the evaluations.
To address these limitations, mobile and wearable technologies are increasingly being deployed outside clinical settings, enabling objective, continuous, and scalable monitoring of patients in their natural environments. This dissertation advances the use of mobile and wearable technologies for assessing physical and behavioral phenotypes in individuals with neurological conditions.
First, ecological momentary assessment (EMA) surveys were administered via mobile devices to stroke survivors, capturing real-time behavioral data in daily life contexts. Language embeddings were derived from survey responses to preserve rich semantic content. Feature selection and supervised machine learning techniques were then applied to detect perceived social isolation among participants.
Second, we explored the potential of wearable sensors to differentiate kinematic characteristics of involuntary movements in Huntington’s disease and Parkinson’s disease with levodopa-induced dyskinesia. We leverage movement decomposition techniques and divide the inertial data of involuntary movements into sub-movements. A combination of unsupervised and supervised machine learning algorithms is employed to automatically select data features extracted from sub-movements and distinguish the two types of involuntary choreic movements.
Finally, we investigate the concurrent validity and reliability of two
movement segmentation approaches widely used to assess the upper-limb motor function of stroke survivors. Acceleration time-series from wrist movements are decomposed into movement segments using each segmentation approach. Reliable features are extracted from the movement segments, and supervised regression models are trained to establish concurrent validity against
existing clinical measures.
Advisor
Ivan Lee