Evaluating kinematic metrics: Towards sensor-based upper limb assessment for home-based stroke rehabilitation

Abstract

Stroke is a major cause of disability worldwide, with two-thirds of survivors experiencing upper limb (UL) impairments that limit their ability to perform daily activities. Effective UL rehabilitation is crucial for maximizing recovery and improving quality of life. To guide therapy selection and track patient progress, UL function assessment tests are commonly used. However, these assessment tests are subjective, lack detail, and are time-consuming. Moreover, the increasing stroke burden and healthcare workforce shortages highlight the need for innovative rehabilitation approaches, including more home-based rehabilitation solutions. Wearable sensor technology, particularly inertial measurement units (IMUs), offers a promising solution for objective, detailed UL assessment, enabling at-home monitoring. In response, the Towards@HomeRehab project, led by the Erasmus Medical Center, was established to develop and evaluate an IMU-based upper limb assessment tool for home use.

This thesis, as part of the Towards@HomeRehab project, had two main objectives: 1) to explore clinicians’ perspectives on an IMU-based assessment tool that uses a standardized drinking task to evaluate upper limb function, and 2) to examine the technical feasibility of extracting kinematic metrics from IMU data collected during the drinking task.

Chapter 2 presents a qualitative study on clinicians’ perspectives regarding the potential IMU-based assessment tool. Three different rehabilitation clinicians were interviewed, all recognizing the tool’s value in objectively and consistently tracking patient progress over time. However, they emphasized the important requirement for the tool to generate a simple, concise overview of clinically relevant movement deviations compared to normal performance. Determining the most relevant output measures remained challenging, as interpreting numerical data requires a shift from their traditional reliance on visual movement assessment.

Chapter 3 presents a pilot study with sixteen healthy subjects examining the technical feasibility of using QSense IMUs to extract kinematic metrics from a standardized drinking task. The results demonstrate that it is technically feasible to derive movement time and joint angle metrics from the IMU data collected during drinking tasks in healthy subjects. The analysis showed excellent relative consistency (ICC ≥ 0.9, SEM < 5% of the mean) and successfully captured different movement patterns, as demonstrated by significant differences between normal and stroke-mimicked drinking task data (p < 0.05). However, a preliminary validation test revealed notable errors in shoulder angle estimations (MAE > 10°).

Chapter 4 summarizes final conclusions, future perspectives, and recommendations. We have demonstrated the potential of an IMU-based upper limb assessment tool for post-stroke rehabilitation. However, the presented data processing methods require refinement to improve metric accuracy and should be extended with additional metrics. Future research should focus on the clinical validation of the extracted metrics, test-retest reliability, and responsiveness of the metrics to functional changes in stroke patients. Additionally, a simple and concise method of presenting the tool’s outcomes should be developed in collaboration with clinicians to ensure alignment with clinical needs, preferences, and workflows.