AI-enabled assessment of gait patterns via smartphone

• Autor(a) principal: Silvestre, Ricardo Miguel Fortunato
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/160965
• topic: decision support systems; gait; dual-task; smartphone; sensors; machine learning; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Musculoskeletal and neurological diseases greatly impact the quality of life of those afflicted. Symptoms typically worsen over time and thus early detection is critical. Gait alterations are a common early warning sign but tests to detect gait alterations require expensive clinical equipment and experienced physical-health workers. Thus, there has recently been great interest in finding lower cost alternatives to data collection and gait alteration detection. We believe that the ubiquity of smartphones and their ability to record motion data reflect an opportunity for democratizing data collection equipment. We found that a particularly smartphone-friendly clinical approach is the dual-task method. Dual-task simultaneously places a patient under both a cognitive and a motor load. The additional cognitive task causes affected individuals to exhibit decreased gait quality allowing for detection of the underlying disease. In this dissertation, we develop dualgAIt, a system for the recording and analysis of a patient’s gait using low-cost smartphones. Data acquisition is achieved through an Android application that leverages only common sensors such as the accelerometer, the microphone and the gyroscope. This data is recorded in a database and processed by a desktop module, which displays visualisations of spatial and temporal gait features in a dashboard for a clinician. Additionally, we train supervised classifiers on these features to provide further decision support through an initial guess presented in the dashboard. We implemented and evaluated gait features from prior work and propose trial distance, a novel feature based on dynamic time warping which has shown good clinical potential as it is correlated with the cognitive dual-task cost. We validate the system through two sets of experiments consisting of single- and dual-task trials, evaluating both the classifiers and the predictive capacity of the features, and show that the system is able to provide effective clinical decision support. Our work shows that it is possible to reduce the cost clinical diagnosis and improve decision support through the use of data-driven technology, while also providing insights into the most relevant gait features.