Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Classifying Diseases Affecting Gait with Body Acceleration-Based Machine Learning Models
This Ph.D. dissertation introduces a comprehensive framework designed to harness acceleration data as a uniquely valuable tool for early disease classification, specifically focusing on gait-related diseases. In the modern healthcare landscape, timely and accurate classification of such diseases is paramount, as it can significantly impact treatment outcomes and patient quality of life. As a compelling case study, we conducted a meticulous experiment to identify individuals afflicted with peripheral artery disease (PAD) and classify them from those without PAD. Our framework leverages acceleration data extracted from strategically placed anatomical reflective markers, including locations such as the sacrum, to train sophisticated classification models. Additionally, we incorporate a wearable accelerometer positioned at the waist for validation purposes. Reflective marker data have a long-standing history in the realm of gait analysis and have been integral to studies evaluating and monitoring human gait for decades. However, they typically require access to specialized laboratories for data collection. In contrast, wearable accelerometers offer the unique advantage of enabling classification in real-world, non-laboratory settings. They empower us to bridge the gap between controlled research environments and the practical challenges of classifying gait-related diseases. Our findings demonstrate the effectiveness of this novel framework. Models trained using raw marker data obtained from the sacrum exhibit an impressive accuracy rate of 92% when classifying PAD patients from non-PAD controls. However, we observed a drop in accuracy to 28% when utilizing data from the wearable accelerometer at the waist to validate the model. To address this challenge, we enhanced our model by employing advanced features extracted from the acceleration data rather than using the raw acceleration signals alone. This modification proved fruitful, as the marker-based model's accuracy decreased only moderately from 86% to 60% when validated with data from the wearable accelerometer. In conclusion, this dissertation establishes a promising avenue for early disease classification, particularly in the context of gait-related diseases. By leveraging acceleration data and innovative modeling techniques, we advance our ability to identify and address health challenges more effectively and efficiently in diverse real-world scenarios. Integrating wearable technology into medical classification marks a significant step towards enhancing healthcare accessibility and patient outcomes.
Biostatistics|Biomedical engineering|Artificial intelligence
Takallou, Mohammad Ali, "Classifying Diseases Affecting Gait with Body Acceleration-Based Machine Learning Models" (2023). ETD collection for University of Nebraska-Lincoln. AAI30811310.