- This event has passed.
Thesis Defence: Neuropsychology’s Machine Assistant: Predicting Functional Outcomes with Machine Learning
November 18, 2022 at 12:30 pm - 4:30 pm
Graham Armstrong, supervised by Dr. Brian O’Connor, will defend their thesis titled “Neuropsychology’s Machine Assistant: Predicting Functional Outcomes with Machine Learning” in partial fulfillment of the requirements for the degree of Master of Arts in Psychology.
An abstract for Graham’s thesis is included below.
Defences are open to all members of the campus community as well as the general public.
Registration is not required for in-person defences.
Stroke affects 50,000 Canadians every year and millions of people worldwide. patients with stroke frequently experience a variety of impairments including prolonged cognitive deficits. Neuropsychological assessment is the most effective way of measuring the nature and magnitude of cognitive impairments. One of the main functions of neuropsychological assessment in a rehabilitation setting is to inform clinical decision-making regarding patient treatment trajectories. Recommendations provided by a neuropsychologist depend on the utility of the neuropsychological battery for predicting the patient’s levels of daily functioning. However, this research has lacked the necessary specificity to capture the comprehensive association between cognition and functionalabilities. Further, domain-specific score practices, regularly employed by clinicians, can lead to substantial levels of misclassification. The goal of the present study was to evaluate the clinical utility of machine learning linear regression, a comprehensive neuropsychological battery, individual neuropsychological predictors and an adjunctive clinical decision-making algorithm in a rehabilitation setting. Data was taken from 167 neuropsychological assessments that were administered to patients with stroke during their stay at the Kelowna General Hospital rehabilitation department. The results indicate that a broad neuropsychological assessment accounts for a significant level of post stroke daily functioning scores. Machine learning is a more powerful tool for identifying individual cognitive predictors of stroke than traditional OLS methods. Machine learning did not provide incremental improvement over OLS methods in terms of multivariate effect sizes. The adjunctive clinical decision-making algorithm did not perform to a sufficient level to provide meaningful input to clinical decision-making in this setting.