FACIAL BEHAVIOUR ANALYSIS FOR CLINICAL APPLICATIONS

Abstract

This thesis explores whether facial feature tracking and facial behaviour analysis can be used to monitor the severity of disease. Some chronic and progressive diseases, such as Parkinson's Disease and Schizophrenia, are associated with impaired facial expressions. Since there is no cure for the disease, current treatments aim to reduce the severity of the symptoms in order to improve patients' quality of life. Clinicians use measures of a patient's facial behaviour (such as expressivity) when assessing their current state, but this is usually done infrequently, and is not practical for daily monitoring. In diseases such as Parkinson's the severity of symptoms vary from one day to the next, so an evaluation on any particular day may not be representative of their longer term status. These limitations have attracted many researchers to investigate the feasibility of developing automatic methods for measuring facial expressivity which could be used daily. However, reported findings are inconsistent and sometimes contradictory. In this thesis, we investigate whether measures can be derived from the results of facial feature tracking which correlate with disease severity, with the aim of providing more information to help clinical assessments. We examine data from a study of patients with Parkinson's disease in which each participant was recorded whilst making a variety of different expressions in response to prompts from a computer program. By tracking the face through each video sequence we were able to monitor their facial movements. We explored a variety of different parameters measuring behaviour (such as the intensity or duration of each expression) in order to identify whether (a) there was a measurable difference between behaviour of people with Parkinson's compared to controls and (b) whether such parameters were correlated with a "Quality-of-Life" (QoL) score estimated from daily questionnaires. We analysed facial movements with respect to a 27- and 51-point model. We found that there are differences in facial behaviour between controls and those with the disease; and that for some individuals there are some parameters which correlate with the QoL scores. Based on our experiments, we found several impaired and correlated facial expressions with QoL scores in both 27- and 51-point models. % First, in correlation assessment with QoL scores, we found that mimicked expressions of anger were the most consistent across all models, while in facial expression impairments, happy and mimicked-happy were the most consistent in both models. % The former, suggests that mimicked expressions of anger are more suited for the assessment of PD QoL, while happy and mimicked-happy expressions are more suited for highlighting facial expression impairments in PD.

Date of Award	31 Dec 2019
Original language	English
Awarding Institution	The University of Manchester
Supervisor	David Morris (Supervisor) & Timothy Cootes (Supervisor)