wearable sensors

UW Data Science Seminar with Megan Ebers

Katie Landwehr News, Presentation , ,
Title slide from the UW eScience Data Science seminar that says "Mobile sensing with shallow recurrent decoder networks. Megan R. Ebers"

Steele lab member and postdoctoral scholar, Megan Ebers, was featured in the Winter 2024 UW Data Science Seminar series. You can watch her full presentation on “Mobile sensing with shallow recurrent decoder networks” linked HERE on UW eScience Institute’s YouTube channel.

Abstract: Sensing is a fundamental task for the monitoring, forecasting, and control of complex systems. In many applications, a limited number of sensors are available and must move with the dynamics, such as with wearable technology or ocean monitoring buoys. In these dynamic systems, the sensors’ time history encodes a significant amount of information that can be extracted for critical tasks. We show that by leveraging the time-history of a sparse set of sensors, we can encode global information of the measured high-dimensional system using shallow recurrent decoder networks. This paradigm has important applications for technical challenges in climate modeling, natural disaster evaluation, and personalized health monitoring; we focus especially on how this paradigm has the potential to transform the way we monitor and manage movement-related health outcomes.

Bio: Megan Ebers is a postdoctoral scholar in applied mathematics with UW’s NSF AI Institute in Dynamic Systems. In her PhD research, she developed and applied machine learning methods for dynamics systems to understand and enable human mobility. Her postdoctoral research focuses on data-driven and reduced-order methods for complex systems, so as to continue her work in human-centered research challenges, as well as to extend her research to a broader set of technical challenges, including turbulent flow modeling, natural disaster monitoring, and acoustic object detection.

HA Feldner, C Papazian, KM Peters, CJ Cruetzfeldt, KM Steele (2021) “Clinical Use of Surface Electromyography to Track Acute Upper Extremity Muscle Recovery after Stroke: A Descriptive Case Study of a Single Patient”

Elijah JournalArticle, News, Publications , , , ,

Journal Article in Applied System Innovation:

This work highlights the potential of wearable technologoies to monitor muscle activity changes during stroke recovery in acute clinical settings and their importance for motivation and understanding of progression from the survivor’s point of view: ‘I was hopeful that it would show signs of things that are occurring when I couldn’t physically feel it…if you had other scientific evidence that things were happening, even beyond their notion that it would, it gives you a lot of hope. You just have to be patient, and it’s harder to take when someone tells you, but easier to understand if someone actually shows you’.

Left image depicts arm with pads placed over muscle with right pictures depicting similar image

Aim: Describe the use of wireless sEMG sensors to examine changes in muscle activity during acute and subacute phases of stroke recovery, and understand the participant’s perceptions of sEMG monitoring.

Method: Muscle activity was tracked by five wireless sEMG sensors beginning three days post-stroke and continued through discharge from inpatient rehabilitation. Activity logs were completed each session, and a semi-structured interview occurred at the final session with three- and eight-month follow-up sessions.

Results: The longitudinal monitoring of muscle and movement recovery in the clinic and community was feasible using sEMG sensors. The participant and medical team felt monitoring was unobtrusive, interesting, and motivating for recovery, but desired greater in-session feedback to inform rehabilitation.

Interpretation: This work highlights that barriers in equipment and signal quality still exist, but capitalizing on wearable sensing technology in the clinic holds promise for enabling personalized stroke recovery.

B Nguyen, N Baicoianu, D Howell, KM Peters, KM Steele (2020) “Accuracy and repeatability of smartphone sensors for measuring shank-to-vertical angle” Prosthetics & Orthotics International

Dr. Steele HuskySTEPS, JournalArticle, Orthoses, UbiquitousRehabilitation , , , , , ,

Journal Article in Prosthetics & Orthotics International

Example of how the smartphone app was used for this research. The top images show a black smartphone attached with a running arm band to the side or front of the shank - the two positions tested in this research. The middle figure shows the placement of the reflective markers for 3D motion analysis to evaluate the accuracy of the smartphone measurements. Markers were placed on the lateral epicondyle of the knee, lateral maleolus of the ankle, tibial tuberosity, and the distal tibia. Blacklight was used to mark the position of each marker and hide the position from the clinicians. The bottom panel shows screenshots from the app. The first screen is used to align the device and has arrows at the top and bottom that remind the clinician which anatomical landmarks should be used to align the device while displaying the shank-to-vertical angle in real time. The second screenshot shows an example of the calculated shank-to-vertical angle while someone was walking. The average is shown with a bold black line, with all other trials shown in blue and excluded trials (e.g., when someone was stopping or turning) that deviated more than one standard deviation from other trials are shown in red. There is also text below the graph that provides summary measures, like shank-to-vertical angle in mid stand and cadence (steps/min). The results can be exported as a picture or sent via e-mail using the app.
A) Smartphone positioning on the front or side of the shank. B) Reflective markers on the the tibial tuberosity (TT) – distal tibia (DT) and lateral epicondyle (LE) – lateral malleolus (LM) were used to compare the accuracy of the smartphone to traditional motion capture. UV markings were used to keep placement of these markers constant while blinding clinicians. C) Sample screenshots of the mobile application, including the set-up screen and results automatically produced after a walking trial.

Background

Assessments of human movement are clinically important. However, accurate measurements are often unavailable due to the need for expensive equipment or intensive processing. For orthotists and therapists, shank-to-vertical angle (SVA) is one critical measure used to assess gait and guide prescriptions. Smartphone-based sensors may provide a widely-available platform to expand access to quantitative assessments.

Objectives

Assess accuracy and repeatability of smartphone-based measurement of SVA compared to marker-based 3D motion analysis.

Method

Four licensed clinicians (two physical therapists and two orthotists) measured SVA during gait with a smartphone attached to the anterior or lateral shank surface of unimpaired adults.  We compared SVA calculated from the smartphone’s inertial measurement unit to marker-based measurements. Each clinician completed three sessions/day on two days with each participant to assess repeatability.

Results

Average absolute differences in SVA measured with a smartphone versus marker-based 3D motion analysis during gait were 0.67 ± 0.25° and 4.89 ± 0.72°, with anterior or lateral smartphone positions, respectively. The inter- and intra-day repeatability of SVA were within 2° for both smartphone positions.

Conclusions

Smartphone sensors can be used to measure SVA with high accuracy and repeatability during unimpaired gait, providing a widely-available tool for quantitative gait assessments.

Try it out!

The app for monitoring shank-to-vertical angle is available for you to download and use on either Android or iOS smartphone. Please complete THIS SURVEY which will then send you an e-mail with instructions for installation and use. This app is not an FDA approved medical device and should be used appropriately.

YC Pan, B Goodwin, E Sabelhaus, KM Peters, KF Bjornson, KLD Pham, WO Walker, KM Steele (2020) “Feasibility of using acceleration-derived jerk to quantify bimanual arm use” Journal of NeuroEngineering and Rehabilitation

Dr. Steele JournalArticle, UbiquitousRehabilitation , , , , ,

Journal Article in Journal of NeuroEngineering & Rehabilitation

Two plots illustrating jerk ratio results. The plot on the left shows the probability distribution from one child with cerebral palsy before, during, and after constraint induced movement therapy. Before therapy, the probability distribution is shifted to the left of the center line, indicating that the individual relies much more on their non-paretic hand during daily life. During therapy, when their non-paretic hand is in a cast, the curve shifts to the right of the center line. This indicates they are using their paretic hand much more - which makes sense, since the other hand is in a cast. Unfortunately, after the cast is removed at the end of therapy, the curve is nearly identical to the curve before treatment, suggesting that after this intensive therapy the child did not use their paretic hand more during daily life. The figure on the right shows the summary metric from this plot, called jerk ratio 50 - which is just the 50% value of the probability density function - for all 5 children with cerebral palsy before, during, and after therapy. All the children have JR50 greater than 0.5 before therapy, which means they use their non-paretic hand more during daily life. During therapy, these values drop to 0.2 - 0.5, indicating that they use their paretic hand much more during CIMT. However, after therapy the JR50 values for all five participants return to close to their baseline value before therapy.
(Left) Example of jerk ratio distribution for one child with cerebral palsy before, during, and after constraint induced movement therapy. (Right) Summary metric of jerk ratio (jerk ratio-50) for all five children with cerebral palsy.

Background

Accelerometers have become common for evaluating the efficacy of rehabilitation for patients with neurologic disorders. For example, metrics like use ratio (UR) and magnitude ratio (MR) have been shown to differentiate movement patterns of children with cerebral palsy (CP) compared to typically-developing (TD) peers. However, these metrics are calculated from “activity counts” – a measure based on proprietary algorithms that approximate movement duration and intensity from raw accelerometer data. Algorithms used to calculate activity counts vary between devices, limiting comparisons of clinical and research results. The goal of this research was to develop complementary metrics based on raw accelerometer data to analyze arm movement after neurologic injury.

Method

We calculated jerk, the derivative of acceleration, to evaluate arm movement from accelerometer data. To complement current measures, we calculated jerk ratio (JR) as the relative jerk magnitude of the dominant (non-paretic) and non-dominant (paretic) arms.  We evaluated the JR distribution between arms and calculated the 50th percentile of the JR distribution (JR50). To evaluate these metrics, we analyzed bimanual accelerometry data for five children with hemiplegic CP who underwent Constraint-Induced Movement Therapy (CIMT) and five typically developing (TD) children. We compared JR between the CP and TD cohorts, and to activity count metrics.

Results

The JR50 differentiated between the CP and TD cohorts (CP = 0.578±0.041 before CIMT, TD = 0.506±0.026), demonstrating increased reliance on the non-dominant arm for the CP cohort. Jerk metrics also quantified changes in arm use during and after therapy (e.g., JR50 = 0.378±0.125 during CIMT, 0.591 ± 0.057 after CIMT). The JR was strongly correlated with UR and MR (r = -0.92, 0.89) for the CP cohort. For the TD cohort, JR50 was repeatable across three data collection periods with an average similarity of 0.945±0.015.

Conclusions

Acceleration-derived jerk captured differences in motion between TD and CP cohorts and correlated with activity count metrics. The code for calculating and plotting JR is open-source and available for others to use and build upon. By identifying device-independent metrics that can quantify arm movement in daily life, we hope to facilitate collaboration for rehabilitation research using wearable technologies.

Code

The algorithm for calculating jerk ratio, as well as user-friendly code to produce plots similar to the figure above are provided open-source as Python 3.6 code as a Python Jupyter Notebook within Google Colab. With this resource, research groups can use existing or newly created data from accelerometers to analyze jerk ratio as a complementary metric to existing measures, enabling comparison between research studies or centers that may rely on different sensors and activity count algorithms.