Smartphone camera can be used to diagnose concussion

13 September 2017
The app - PupilScreen - can detect changes in a pupil’s response to light using a smartphone’s video camera and deep learning tools to quantify changes imperceptible to the human eye. This pupillary light reflex has long been used to assess whether a patient has severe traumatic brain injury. Recent research finds it can be useful in detecting milder concussions — opening up a new avenue for screening.

The team of UW computer scientists, electrical engineers and medical researchers has demonstrated that PupilScreen can be used to detect instances of significant traumatic brain injury.  A broader clinical study this fall will have PupilScreen tested in real life by coaches, emergency medical technicians, doctors and others.  They will help gather more data on which pupillary response characteristics are most helpful in determining ambiguous cases of concussion. The researchers hope to release a commercially available version of PupilScreen within two years.

Half of yearly concussions go unnoticed
The U.S. Centers for Disease Control and Prevention estimates about half of the 3.8 million concussions per year in the U.S. from recreational sports injuries alone still go undiagnosed, with risk for future head injury and permanent cognitive deficits.

“Having an objective measure that a coach or parent or anyone on the sidelines of a game could use to screen for concussion would truly be a game-changer,” said Shwetak Patel, the Washington Research Foundation Endowed Professor of Computer Science & Engineering and of Electrical Engineering at the UW. “Right now the best screening protocols we have are still subjective. A player who really wants to get back on the field can find ways to game the system.”

PupilScreen can assess a patient’s pupillary light reflex almost as well as a pupilometer, an expensive and rarely used machine found only in hospitals. It uses the smartphone’s flash to stimulate the patient’s eyes and the video camera to record a three-second video.

The video is then processed using deep learning algorithms that can determine which pixels belong to the pupil in each video frame and measure the changes in pupil size across those frames. In a small pilot study that combined 48 results from patients with traumatic brain injury and from healthy people, clinicians were able to diagnose the brain injuries with almost perfect accuracy using the app’s output alone.

No surefire way yet to diagnose concussion
So far there is no surefire way to diagnose concussion — even in the emergency room, said co-author Dr. Lynn McGrath, a resident physician in UW Medicine’s Department of Neurological Surgery. Doctors usually run tests to rule out worst cases like a brain bleed or skull fracture. After more serious head injuries are excluded, a diagnosis of concussion can be made.

Medical professionals have long used the pupillary light reflex — usually in the form of a penlight test where they shine a light into a patient’s eyes — to assess severe forms of brain injury. But a growing body of medical research has recently found that more subtle changes in pupil response can be useful in detecting milder concussions.
“PupilScreen aims to fill that gap by giving us the first capability to measure an objective biomarker of concussion in the field,” McGrath said. “After further testing, we think this device will empower everyone from Little League coaches to NFL doctors to emergency department physicians to rapidly detect and triage head injury.”

Computer can quantify subtle changes
One of the challenges in developing PupilScreen involved training the machine learning tools to distinguish between the eye’s pupil and iris, which involved annotating roughly 4,000  images of eyes by hand. A computer has the advantage of being able to quantify subtle changes in the pupillary light reflex that the human eye cannot perceive.

“Instead of designing an algorithm to solve the specific problem of measuring pupil response, we moved this to a machine learning approach — collecting a lot of data and writing an algorithm that allowed the computer to learn for itself,” said co-author Jacob Baudin,  a UW medical student and doctoral student in physiology and biophysics.

The PupilScreen researchers are currently working to identify partners interested in conducting additional field studies of the app, which they expect to begin in October. The project was funded by the National Science Foundation, the Washington Research Foundation and Amazon Catalyst.