Design and Evaluation of Augmented Reality Visualization for Physical Therapy

Physical therapy (PT) is critical for managing motor challenges from injury, aging, stroke, and chronic conditions such as Parkinson's Disease (PD), but current technology does not adequately address crucial challenges. Clinicians are constrained to bi-weekly meetings and lack insight into crucial motor data—measures describing the patient's body in everyday life, such as posture, gait and feet shuffling, and environmental triggers of symptoms—that current approaches cannot capture. In-home video tutorials fail to offer intuitive guidance or adapt difficulty to each person's needs and abilities. Augmented Reality (AR) shows promise to address these challenges by using life-size renderings that mirror in-person clinical observation, and its situated, gamified guidance can support home practice. However, we lack visual analytics for navigating daily movement data and design guidelines for AR motion guidance across varying patient abilities. This dissertation establishes that AR visualization systems designed specifically for the challenges and needs of PT can support both clinical assessment and patient experiences by personalizing PT protocols with (1) efficient, contextualized analysis of large-scale movement data for clinicians, and (2) situated, gamified, and adaptive motion guidance for in-home patient exercise. Following a design research methodology grounded in collaboration with PD clinicians and patients, the work combines interviews, controlled experiments, algorithmic simulation, and system evaluations to develop guidelines for building effective PT systems. The dissertation contributes: (1) task characterizations from clinician and patient interviews that surface current real-world challenges; (2) PD-Insighter, an AR visual analytics system pairing a desktop dashboard with immersive replay, enabling clinicians to efficiently identify motor deficits across large amounts of daily activity; (3) empirical guidelines for AR motion guidance, showing that effective perspective depends on motion visibility and that object encodings increase usability compared to line encodings; and (4) design tradeoffs for multi-parameter dynamic difficulty adjustment in AR exergames, finding that model-based algorithms best balance stability, personalization, and exploration, and that more adaptation does not monotonically improve motivation. Together, these studies inform a roadmap for an integrated AR rehabilitation system spanning the clinic and the home.