Title: " Machine Learning for Smartphone-based Monitoring and Treatment of Unipolar and Bipolar Disorders"
Supervisors:
Principal supervisor: Professor Ole Winther, DTU compute
Co supervisor: Professor Jakob Bardram, DTU Compute
Examiners:
(Chairman) Professor Sune Lehmann Jørgensen, DTU Compute
Professor Cecilia Mascolo, University of Cambridge, UK
Associate Professor Aki Vehtari, Aalto University, Finland
Chairperson at defence:
Associate Professor Per Bækgaard
Abstract:
|
Bipolar disorder is a common mental illness characterized by unusual changes in mood and energy and is regarded as one of the most important causes of disability worldwide. By replacing traditional paper-based self-assessments with a smartphone-based solution for disease monitoring and treatment, users can unobtrusively collect and monitor their own data. Modern smartphones additionally enable pervasive collection of sensor data that can track a wide range of human behaviors relevant to managing mental illness. Automated analysis of smartphone data can potentially be used to detect early warning signs and predict disease outcomes, which can facilitate early intervention and thus potentially mitigate severity of affective episodes and prevent costly hospitalizations.
This PhD thesis seeks to establish methods and algorithms for analysis of behavioral smartphone data from patients with bipolar disorder. By applying statistical methods and machine learning, we have successfully been able to predict symptoms of depression and mania from self-assessment histories and assess individual risk of relapse. Additionally, we applied simple features of objective smartphone data representing physical and social activity to distinguish between patients with bipolar disorder and healthy individuals.
Based on the current research, we are confident that the use of smartphones and algorithms has the potential to transform treatment of bipolar disorder and possibly similar illnesses. To accomplish this goal, it is crucial to develop technology with high usability supported by algorithms able to produce accurate, interpretable and actionable results.
|