| Summary: | Mental and behavioural disorders introduce a significant burden on society, estimated to account for 12% of the global burden of disease, with approximately 450 million suffering from them every day, and only a small number of those getting any treatment. The situation will worsen with time, with unipolar depressive disorders predicted by the World Health Organisation to become the leading cause of disabilities by 2030. Mental disorders affect primarily the mind and the brain, leading to pathological changes in emotions or cognition. Although clinical manifestations of different mental disorders may vary, Liddle et al. suggested five principal symptom dimensions, including reality distortion, disorganisation, psychomotor, mood and anxiety dimensions. For assessment of symptoms in clinical practice, the structured clinical interview, alongside standard questionnaires are used, but in many cases are not providing a reliable and objective diagnostic tool due to the complexity of the assessed phenomena. Activity and motion analysis has the potential to be used as a diagnostic tool for mental disorders. However, to-date, little work has been performed in turning stratification measures of activity into useful symptom markers. The research presented in this thesis has focused on the identification of objective activity and behaviour metrics that could be useful for the analysis of mental health symptoms in the above mentioned dimensions. Particular attention is given to the analysis of objective differences between disorders, as well as identification of clinical episodes of mania and depression in bipolar patients, and deterioration in borderline personality disorder patients. A principled framework is proposed for mHealth monitoring of psychiatric patients, based on measurable changes in behaviour, represented in physical activity time series, collected via mobile and wearable devices. The framework defines methods for direct computational analysis of symptoms in disorganisation and psychomotor dimensions, as well as measures for indirect assessment of mood, using patterns of physical activity, sleep and circadian rhythms. An extensive mHealth software tracking system was constructed, and data collected from over 100 individuals. Using the developed framework, the accuracy of differentiation between healthy controls and bipolar disorder was 67%, healthy controls and borderline personality disorder 70%, and bipolar vs. borderline personality disorder 80%. For identification of clinical states of euthymia, mania and depression the accuracy of differentiation of euthymia and mania was 80%, euthymia and depression 85% and mania and depression 90%, when using leave-one-out cross-validation. For personalised mood models, the mean absolute error of symptom scores estimation was in the range of 1.36 to 3.32 points, this corresponds to the ranges reserved in psychiatric questionnaires for a unique identifiable mood state (4-5 points). Finally, the methods were applied to a new data set (schizophrenia patients and matched controls) and were shown to be 95.3% accurate using leave-one-out cross-validation at classifying the cohort. Both physiological as well as activity features were relevant for classification of this cohort, and so the hypothesis that heart rate added additional predictive power was tested. The combination of HR and locomotor activity features provided almost a 10% increase in classification accuracy above using locomotor features alone, and almost a 17% increase over using heart rate based features alone. The approach of computational behaviour analysis, proposed in this thesis, has the potential for early identification of clinical deterioration in ambulatory patients, and allows for the specification of distinct and measurable behavioural phenotypes, thus enabling better understanding and treatment of mental disorders.
|
|---|
