Stroke affects millions of people each year and although modern medicine has improved chances of survival after stroke, it has not yet been able to affect a change in repairing damaged neural tissue leaving one to two-thirds of survivors with chronic disability in their affected upper-extremity; specifically, hemiparesis, hypertonicity, loss of coordination, and spasticity. These impairments require survivors to adapt and compensate for the loss of function in their arm, wrist, and hand. These impairments also discourage many survivors and cause them to use their arm less and less leading to atrophy and limitations in range of motion.One phenomenon which limits in part both the rehabilitation process, as well as the efficacy of mechanical intervention (exoskeletons or wearable assistive devices), is the abnormal synergy, in which greater proximal effort of the arm leads to increased tone and contraction in a patterned way throughout the arm, wrist, and hand. Lifting the arm against gravity often causes unintentional co-activation of elbow, wrist, and finger flexors. By reducing effort at the shoulder, this phenomenon is reduced thus enabling greater control and range of motion at these joints. Without arm support, individuals with stroke and any equipment they utilize must overcome these increased abnormal joint torques. With the advancement of exoskeletons, a powered device that supports humeral elevation is foreseeable but with that comes the requirement for a control system to control it. This dissertation explores the possibility of using pattern recognition, a type of machine learning, to control vertical support at the shoulder after stroke thus reducing the effort required and consequently reducing the severity of presentation of the abnormal synergy. It is hoped that this work will contribute towards the realization of a control paradigm that will aid in the rehabilitation and assistance of those surviving stroke. These chapters progress from a purely isometric single direction shoulder task (Chapter 2) to a quasi-static, quasi-dynamic dual-task (Chapters 3 and 4), to finally real-time control of vertical support using machine learning (Chapter 5). The nature of this work limits any strong conclusive statements but ultimately has shown the promise and efficacy of such a control system. Specifically, Chapter 5 shows that myoelectric-based pattern recognition control enables survivors of stroke to control both vertical support force and vertical position well enough to place their arm in a target window and improve their forward reach ability by increasing joint excursion at the elbow and the shoulder. Chapter 4 questions the presence of abnormal synergy within shoulder joint degrees of freedom. It concludes that a natural synergy exists between adduction/internal rotation based on normal muscle biomechanics that constrains shoulder movements “outside of synergy” such as adduction/external rotation or abduction/internal rotation. This is a deviation from the commonly accepted hypothesis that abduction or adduction drive causes patterned and obligatory torque coupling with external rotation and internal rotation respectively. Ultimately these results have advanced our understanding of control of the shoulder after stroke and have demonstrated the feasibility and efficacy of using a myoelectric-based control system to control vertical support in order to increase function. Further work is required to refine and optimize these techniques and possibly develop automated decision-making systems to aid in rehabilitation.

Creator

• Kopke, Joseph

DOI

• https://doi.org/10.21985/n2-8exq-yb59

Subject

• Robotics

Language

• en

Alternate Identifier

• http://dissertations.umi.com/northwestern:15547
• etdadmin_upload_817626

Keyword

• Stroke
• Pattern Recognition
• Hemiparesis
• Abnormal Synergy
• Rehabilitation
• Machine Learning

Date created

• 2021-01-01

Resource type

• Dissertation

Rights statement

• In Copyright