{"id":{"repo_id":"calpoly","oai_identifier":"oai:digitalcommons.calpoly.edu:theses-4990"},"canonical_url":"https://search.ndltd.org/etd/calpoly/oai:digitalcommons.calpoly.edu:theses-4990","repository":{"repo_id":"calpoly","name":"Cal Poly","base_url":"https://digitalcommons.calpoly.edu/do/oai/"},"display":{"title":"Practical Multimodal Wearable Sensing for Functional Upper Extremity Primitive Classification With Application to Stroke Rehabilitation","abstract":"Stroke often causes long-term weakness and impaired motor control in the upper extremity (UE), making everyday tasks such as reaching, grasping, and moving objects more difficult. Restoring functional arm use is therefore a central goal of post-stroke rehabilitation. Measuring affected arm use continuously and objectively is important because isolated clinical assessments may not fully capture how the affected arm is used during therapy or daily life. Wearable sensors offer a promising approach for monitoring, but raw sensor signals are difficult to interpret directly. Functional movement primitives address this issue by describing UE behavior as smaller, task-agnostic movement units. This thesis evaluates whether five functional UE primitives can be classified using a wearable system that combines motion and muscle activation data. Fourteen unimpaired participants performed structured functional UE tasks while wearing sensors on the dominant limb. Ground-truth labels for Reach, Reposition, Transport, Stabilize, and Idle functional primitives were extracted from video data. Four model architectures were evaluated: Random Forest, Convolutional Neural Network (CNN), Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM), and Noncausal Temporal Convolutional Network (TCN). Additionally, a temporal decoding pipeline was applied to convert raw model predictions into continuous primitive timelines, enabling summaries of potentially clinically meaningful metrics. The CNN-BiLSTM trained on motion and muscle activation data achieved the strongest performance, with a base classification macro-F1 score of 0.877. After temporal decoding, this model also produced the strongest primitive timelines among the combined-sensor models, with the best temporal overlap and lowest count and duration errors. The best-performing architecture was also trained using motion features alone. In this comparison, the motion-only version slightly outperformed the combined-sensor version, suggesting that muscle activation did not provide an aggregate performance benefit in this dataset. These results suggest that reduced-sensor primitive classification is feasible for prescribed functional tasks in unimpaired participants. By converting predictions into primitive timelines, the system could provide clinician-facing summaries of functional arm use, including how often each primitive occurs, how long each primitive lasts, and the percentage of time spent active versus idle. These summaries may help clinicians better characterize therapy content, compare engagement across sessions, and adjust task selection or therapy dose based on how the affected upper extremity is being used","abstract_html":"Stroke often causes long-term weakness and impaired motor control in the upper extremity (UE), making everyday tasks such as reaching, grasping, and moving objects more difficult. Restoring functional arm use is therefore a central goal of post-stroke rehabilitation. Measuring affected arm use continuously and objectively is important because isolated clinical assessments may not fully capture how the affected arm is used during therapy or daily life. Wearable sensors offer a promising approach for monitoring, but raw sensor signals are difficult to interpret directly. Functional movement primitives address this issue by describing UE behavior as smaller, task-agnostic movement units. This thesis evaluates whether five functional UE primitives can be classified using a wearable system that combines motion and muscle activation data. Fourteen unimpaired participants performed structured functional UE tasks while wearing sensors on the dominant limb. Ground-truth labels for Reach, Reposition, Transport, Stabilize, and Idle functional primitives were extracted from video data. Four model architectures were evaluated: Random Forest, Convolutional Neural Network (CNN), Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM), and Noncausal Temporal Convolutional Network (TCN). Additionally, a temporal decoding pipeline was applied to convert raw model predictions into continuous primitive timelines, enabling summaries of potentially clinically meaningful metrics. The CNN-BiLSTM trained on motion and muscle activation data achieved the strongest performance, with a base classification macro-F1 score of 0.877. After temporal decoding, this model also produced the strongest primitive timelines among the combined-sensor models, with the best temporal overlap and lowest count and duration errors. The best-performing architecture was also trained using motion features alone. In this comparison, the motion-only version slightly outperformed the combined-sensor version, suggesting that muscle activation did not provide an aggregate performance benefit in this dataset. These results suggest that reduced-sensor primitive classification is feasible for prescribed functional tasks in unimpaired participants. By converting predictions into primitive timelines, the system could provide clinician-facing summaries of functional arm use, including how often each primitive occurs, how long each primitive lasts, and the percentage of time spent active versus idle. These summaries may help clinicians better characterize therapy content, compare engagement across sessions, and adjust task selection or therapy dose based on how the affected upper extremity is being used","abstract_has_math":false,"creators":["Weiss, Nicholas"],"institution":null,"degree_name":"MS in Computer Science","degree_level":null,"degree_discipline":"Computer Science","degree_department":null,"school":null,"contributors":["Ria Kanjilal","Computer Science","College of Engineering"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T07:12:58Z","subjects":["Stroke rehabilitation","upper extremity","wearable sensors","inertial measurement units","surface electromyography","functional movement primitives","Biomedical","Signal Processing"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.calpoly.edu/theses/3317","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.calpoly.edu/theses/3317","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://digitalcommons.calpoly.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.calpoly.edu%3Atheses-4990","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Ria Kanjilal","Computer Science","College of Engineering"]},{"key":"dc:creator","label":"Author","values":["Weiss, Nicholas"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2026-06-11T07:00:00Z"]},{"key":"dc:date.created","label":"Dc Date Created","values":["2026-06-01T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science"]},{"key":"thesis:degree_name","label":"Degree Name","values":["MS in Computer Science"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Stroke rehabilitation","upper extremity","wearable sensors","inertial measurement units","surface electromyography","functional movement primitives","Biomedical","Signal Processing"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.calpoly.edu/theses/3317"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Stroke often causes long-term weakness and impaired motor control in the upper extremity (UE), making everyday tasks such as reaching, grasping, and moving objects more difficult. Restoring functional arm use is therefore a central goal of post-stroke rehabilitation. Measuring affected arm use continuously and objectively is important because isolated clinical assessments may not fully capture how the affected arm is used during therapy or daily life. Wearable sensors offer a promising approach for monitoring, but raw sensor signals are difficult to interpret directly. Functional movement primitives address this issue by describing UE behavior as smaller, task-agnostic movement units. This thesis evaluates whether five functional UE primitives can be classified using a wearable system that combines motion and muscle activation data. Fourteen unimpaired participants performed structured functional UE tasks while wearing sensors on the dominant limb. Ground-truth labels for Reach, Reposition, Transport, Stabilize, and Idle functional primitives were extracted from video data. Four model architectures were evaluated: Random Forest, Convolutional Neural Network (CNN), Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM), and Noncausal Temporal Convolutional Network (TCN). Additionally, a temporal decoding pipeline was applied to convert raw model predictions into continuous primitive timelines, enabling summaries of potentially clinically meaningful metrics. The CNN-BiLSTM trained on motion and muscle activation data achieved the strongest performance, with a base classification macro-F1 score of 0.877. After temporal decoding, this model also produced the strongest primitive timelines among the combined-sensor models, with the best temporal overlap and lowest count and duration errors. The best-performing architecture was also trained using motion features alone. In this comparison, the motion-only version slightly outperformed the combined-sensor version, suggesting that muscle activation did not provide an aggregate performance benefit in this dataset. These results suggest that reduced-sensor primitive classification is feasible for prescribed functional tasks in unimpaired participants. By converting predictions into primitive timelines, the system could provide clinician-facing summaries of functional arm use, including how often each primitive occurs, how long each primitive lasts, and the percentage of time spent active versus idle. These summaries may help clinicians better characterize therapy content, compare engagement across sessions, and adjust task selection or therapy dose based on how the affected upper extremity is being used"]},{"key":"dc:title","label":"Title","values":["Practical Multimodal Wearable Sensing for Functional Upper Extremity Primitive Classification With Application to Stroke Rehabilitation"]}]}],"canonical_facts":{"dc:contributor":["Ria Kanjilal","Computer Science","College of Engineering"],"dc:creator":["Weiss, Nicholas"],"dc:date.available":["2026-06-11T07:00:00Z"],"dc:date.created":["2026-06-01T07:00:00Z"],"dc:description.abstract":["Stroke often causes long-term weakness and impaired motor control in the upper extremity (UE), making everyday tasks such as reaching, grasping, and moving objects more difficult. Restoring functional arm use is therefore a central goal of post-stroke rehabilitation. Measuring affected arm use continuously and objectively is important because isolated clinical assessments may not fully capture how the affected arm is used during therapy or daily life. Wearable sensors offer a promising approach for monitoring, but raw sensor signals are difficult to interpret directly. Functional movement primitives address this issue by describing UE behavior as smaller, task-agnostic movement units. This thesis evaluates whether five functional UE primitives can be classified using a wearable system that combines motion and muscle activation data. Fourteen unimpaired participants performed structured functional UE tasks while wearing sensors on the dominant limb. Ground-truth labels for Reach, Reposition, Transport, Stabilize, and Idle functional primitives were extracted from video data. Four model architectures were evaluated: Random Forest, Convolutional Neural Network (CNN), Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM), and Noncausal Temporal Convolutional Network (TCN). Additionally, a temporal decoding pipeline was applied to convert raw model predictions into continuous primitive timelines, enabling summaries of potentially clinically meaningful metrics. The CNN-BiLSTM trained on motion and muscle activation data achieved the strongest performance, with a base classification macro-F1 score of 0.877. After temporal decoding, this model also produced the strongest primitive timelines among the combined-sensor models, with the best temporal overlap and lowest count and duration errors. The best-performing architecture was also trained using motion features alone. In this comparison, the motion-only version slightly outperformed the combined-sensor version, suggesting that muscle activation did not provide an aggregate performance benefit in this dataset. These results suggest that reduced-sensor primitive classification is feasible for prescribed functional tasks in unimpaired participants. By converting predictions into primitive timelines, the system could provide clinician-facing summaries of functional arm use, including how often each primitive occurs, how long each primitive lasts, and the percentage of time spent active versus idle. These summaries may help clinicians better characterize therapy content, compare engagement across sessions, and adjust task selection or therapy dose based on how the affected upper extremity is being used"],"dc:identifier":["https://digitalcommons.calpoly.edu/theses/3317"],"dc:subject":["Stroke rehabilitation","upper extremity","wearable sensors","inertial measurement units","surface electromyography","functional movement primitives","Biomedical","Signal Processing"],"dc:title":["Practical Multimodal Wearable Sensing for Functional Upper Extremity Primitive Classification With Application to Stroke Rehabilitation"],"thesis:degree_discipline":["Computer Science"],"thesis:degree_name":["MS in Computer Science"]},"updated_at":"2026-11-19T07:12:58Z"}