Research In Action

Research In Action

CHOP head sensor data research
Roadmap for Protocols Using Head Impact Sensors
March 11, 2020
Share  

My recently published paper in the American Journal of Sports Medicine, titled ‘Video Confirmation of Head Impact Sensor Data from High School Soccer Players’, is part of a large study at Children's Hospital of Philadelphia (CHOP) which aims to advance pediatric concussion assessment, diagnosis, and treatment. In this blog post, I will explain the role of head impact sensors in this work and how to get research-quality data from the devices.

For the biomechanical arm of this study, we needed to collect in vivo head impact biomechanics data on youth athletes. We partnered with the Shipley School, where their athletes were already wearing Triax SIM-G headband-mounted impact sensors during sports competition.

Typically for a sensor study, the sensor is selected before partnering with sporting teams. However, this was a unique case with the added benefit that the Shipley athletes were already accustomed to wearing the sensors, and the Athletic Department had a protocol in place to distribute, collect, and charge the sensors, as well as wash the headbands.

Insights for Field Use of Head Impact Sensors

The sensor had previously been evaluated for accuracy in the laboratory, but a field evaluation had yet to be performed. Some previous studies on other sensors had suggested that video confirmation is necessary to remove false positives; therefore, we decided to film every game and develop a rigorous video analysis protocol to review each sensor event.

During this time-intensive video analysis, we found that the number of sensor events recorded during verified game times was different from those recorded during scheduled game times. We also found that many sensor events were from players on the bench.

Once we had cleaned the data, we classified sensor events into three distinct categories: 

  1. Impact Events - Events that occur when head contact can be identified, such as when a soccer player heads the ball.
  2. Trivial Events - Events that occur when the sensor experiences movement that can be observed, but is not a head contact of interest, such as when a player adjusts his or her headband.
  3. Sensor-Recoded Events - Events for which there is no observable head contact or sensor motion, which we defined as nonevents.

Without video-confirmation, approximately one-in-five events recorded by the sensor are actual head impacts. This finding supports the previous literature that video confirmation is necessary to remove false positives.

Some sensors incorporate processing algorithms designed to remove false positives, which are typically developed in a laboratory setting without being evaluated in the field. The Triax SIM-G has one of these such algorithms, which is able to be deactivated. When compared to our video analysis, we found that the algorithm incorrectly classified a third of all sensor events. Therefore, we recommend that sensor processing algorithms be avoided until they have been evaluated in the field.

Stay Tuned For My Next Blog

In the past year, our research team switched to the use of mouth guard head impact sensors for our in vivo data collection. In my next blog post, I will share the benefits and challenges of using mouth guards in our research.

Read a press release on this study