Senior Design Projects
ECS193 A/B Winter & Spring 2019
Automated Determination of Human Gait Parameters Using Triaxial Accelerometer Data from a Pediatric Population
| ********** | |
| Name | Erik Henricson ( with Lisa Miller, Daniel Aranki ) |
| Affiliation | UCD Physical Medicine & Rehabilitation, Human Dev. Graduate Group, Berkeley CITRIS Center |
Project's details
| Project title | Automated Determination of Human Gait Parameters Using Triaxial Accelerometer Data from a Pediatric Population |
| Background | Duchenne muscular dystrophy ( DMD ) is a neuromuscular and developmental disorder affecting approximately 1 in 5000 newborn males. Individuals with DMD gradually lose ambulatory ability due to loss of strength and muscle mass, typically in the early-teenaged years. In typically-developing humans, stride length and step cadence increase with increasing level of exercise, and change as a child develops. Typical research and consumer-level step monitoring devices are commonly worn at the wrist or ankle, and are primarily useful for measuring walking and running habits in adult-sized individuals with typical “pendulum” gait patterns. However, few devices have demonstrated the ability to effectively record step counts or distances traveled in small children and in those with atypical gait patters due to neuromuscular disease. We have developed an initial methodology for capturing the shortening of stride length due to muscle wasting beginning at toddler ages ( where benefit from novel genetic and drug therapies may be greatest ) through the loss of ambulation in the early teen years. Low-cost, community-based wearable monitors placed on the trunk instead of the extremities provide a useful tool to measure longitudinal gait changes in boys with DMD. Using machine learning algorithms, it is possible to use data from these readily available sensors to detect individual steps regardless of their gait pattern, to evaluate their average g-forces ( acceleration ) generated with each step while walking at a range of speeds over a known distance in order to quantify individual steps, and to associate g-forces for those steps with observed stride lengths. Using this methodology, we can measure steps and distances traveled during free roaming conditions without the use of GPS, and measure changes in gait in the form of shortening stride length and diminishing variability in step-cadence. Our initial attempts at measuring stride length and distance traveled based on data collected from a low-cost sensor and a computer learning algorithm “trained” to an individual’s specific gait characteristics appear promising ( See attached poster for more information ). |
| Description | Data Set: We collected triaxial accelerometer data using the Mbient Labs Metaware sensor ( $67.00 ) in typically developing youth ( TDY ) and ambulatory boys with DMD from age 18 months to 13 years at the UC Davis Center for Child and Family Studies and the UC Davis Neuromuscular Research Center. Participants underwent age-appropriate functional evaluations of motor ability that included collection of “training” calibration data at various self-selected increasing walking speeds on a 25-meter course, and either the 100-meter walk test or the Six-minute Walk Test ( 6MWT ) as appropriate for age, all while wearing sensors at mid-scapular and lumbar locations. We observed and recorded participant’s step counts and distance traveled for comparison with algorithm-based estimates derived from either the X, Y or Z dimension accelerometer components. Project Description: To continue our work, we will explore potential for applying methods of canine gait evaluation developed by Barthelemy et al ( 2011 ) to our human sensor data. Using those methods and the quaternion sensor signals, we aim to develop a data analysis program which will determine stride frequency, stride length, stride regularity, total power of accelerations, relative power of individual axial accelerations, relative components of total power of accelerations, and relative force index. The resulting gait components will be used in future studies to develop linear discriminant analysis models to evaluate effects of age and disease severity on sensor signal characteristics that could be used to analyze gait data collected in free-roaming community conditions. |
| Deliverable | The resulting program could be stand-alone ( on Mac, PC, iOS or Android platfomrs ) or could be an extension of existing MatLab ML step detection programs . Specifics will include at least: 1. Ability to input raw sensor data files and select specific subsets of sensor data based on time stamps for individual data points. 2. Ability to input participant-specific characteristics such as height, weight and activity type ( ex. walking vs. running ). 2. Ability to use selected subsets of sensor data to calculate stride frequency, stride length, stride regularity, total power of accelerations, relative power of individual axial accelerations, relative components of total power of accelerations, and relative force index. 3. Ability to display calculated parameters on an easily readable "dashboard" output screen. |
| Skill set desirable | Experience with MatLab or similar system for input and analysis of continuously measured triaxial sensor data using machine learning-type algorithms. Basic understanding of biology and mechanics of human movement, including measurement using wearable sensors. |
| Phone number | ********** |
| Client time availability | 30-60 min weekly or more |
| IP requirement | Client wishes to keep IP of the project |
| Attachment | Click here |
| Selected | No |
| Team members | N/A |
| TA | N/A |