Sun, Yanke; (2025) Development of Quantitative Methods for Movement Sensor Data in Neuroscience Applications. Doctoral thesis (Ph.D), UCL (University College London).

Text Sun_10204985_thesis.pdf Access restricted to UCL open access staff until 1 March 2027. Download (69MB)

Abstract

This thesis examines the potential of using long-term and Real-world Data (RWD) from wearable movement sensors to address challenges in real-world neuroscience applications. A general data collection and analysis pipeline using Inertial Measurement Unit (IMU) sensors and RWD is developed and tested across two neuroscience areas: single-user monitoring of Parkinson’s Disease (PD) and multi-user applications for quantifying social interactions. Both areas require the development of objective methods to overcome limitations of traditional assessment approaches. In the PD application, Machine learning (ML) was applied to long-term sensor data from patients for human activity recognition. Correlation analyses between predicted activity features and clinically assessed disease severity levels revealed significant correlations, demonstrating the feasibility of continuous symptom monitoring and disease progression tracking using a single sensor. A comparative analysis between two sensor locations provided insights into data analysis methods and patient acceptability for long-term monitoring. In social interaction applications, the first dataset captured head accelerations from actors and the audience during live theatre performances. Cross-wavelet analysis techniques successfully measured interpersonal synchrony, a crucial indicator of interaction, across different frequency ranges. Positive correlation was observed between movement synchrony and self-reported emotional engagement scores. The second dataset aims to study social interactions among autistic and neurotypical children in classroom settings with data collected from eight sessions. Various time series similarity analyses were evaluated against video-coded interaction level scores by regression analysis. ML approaches were also employed to classify interaction levels, showing similarity analyses as features outperforming baseline features. Overall, the thesis contributes to developing objective coding and visualisation tools for real-world PD monitoring and social interaction quantification using IMU sensors. These tools can be easily adapted for similar applications involving movement analysis. It also provides valuable insights for improving experiment design and methodologies in real-world neuroscience research from lessons learned in data collection and analysis.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item