In a recent hackathon at the Technion-Israel Institute of
Technology, a team of IBM Research scientists from Haifa won the top prize at the Brain
Inspired Technology for Education (BITE)
Hackathon for their application that screens for early indications of attention
deficit and hyperactivity disorder, or ADHD.
The idea for an ADHD early identification app all started
when researchers at IBM’s research lab in Haifa, Israel were approached by Anat Dahan, a
doctoral student at the Technion’s Virtual-Reality and NeuroCognition Lab. She wanted
their help to develop a fast-acting, low-cost method to help people to
determine if they need further evaluation for a diagnosis of ADHD. Dahan is an
expert in brain-inspired technologies whose idea is supported by a body of
previously published research, noting that individuals with ADHD have a-typical
motoric (muscular movement) characteristics. And her work inspired the IBM scientists to build a
new application that married machine learning with data from wearable devices.
|
Augmented reality implementation |
They set out to build a tool that could accurately classify
people with ADHD based on their arm and hand movements. Available literature
specified only that people with ADHD have different motoric characteristics – but
not which ones. This situation was ideal for the use of supervised machine
learning, which especially useful in situations where we don’t know the exact
model underlying the correspondence between the input (the movement
acceleration measurements) and the output (the classification of individuals as
ADHD positive or negative).
According to
WebMD,
ADHD is a collection of symptoms that may begin in childhood and continue into
adulthood. ADHD as well as ADD symptoms, such as hyperactivity, impulsiveness,
and inattentiveness, can cause problems at home, school, work, or in
relationships. Recent statistics estimate that 4 to 5 percent of U.S. adults
have ADHD, and the
CDC
estimates that 11 percent of American children, ages 4 to 17, have the
attention disorder. Unfortunately, not everyone gets diagnosed or treated for
it.
|
Measuring simple motoric task with
TI Sensortag accelerometer |
The IBM Research project has the potential to help many
sufferers by providing an early indication that will lead individuals to seek
proper medical evaluation.
ADHD Self-Assessment Screening via Natural Hand Movement
Analysis
The IBM researchers designed the self-conducted ADHD
screening test to be a simple repetitive task that anyone can do with their own
smartphone. The app works by asking users to draw a rectangle 10 times, while accelerometers
and sensors in their phone or their wearable tracks the movement. This helped identify
those individuals who have difficulty initiating and maintaining the continuous
motor activity.
For the hackathon, the researchers first established a
baseline. Their 17 test subjects (four of whom self-reported having ADHD)
filled out Harvard University’s
ASRS report form, and the
team also collected their brain activity – which can also indicate ADHD – via
EEG headbands, before letting them use the app.
|
EEG recording as a validation benchmark
(NeuroSteer Headband) |
According to the researchers, the application provided
preliminary evidence for a possible relationship between mental and motoric
characteristics. The data may also be useful in connecting the frequency of the
motoric rhythm people use with the measurement of concentration from the EEG
tasks. And the hackathon judges were pleased that the test results were
available within 25 seconds of the subject drawing the rectangles.
Moving into the future
The IBM team’s work in gesture recognition, anomaly
detection, and machine learning made it possible to collect and analyze the
data, while gaining fascinating insight into both the test and the tested individuals’
abilities for concentration.
Although the project with Dahan is complete, the IBM
researchers believe this type of analytics can help identify different types of
ADHD, not just a single classification of those with difficulties in
concentration. Working with more subjects would make more data available and
offer more significance to their findings. In the future, they hope to work
with other conditions that involve motoric indications, such as Parkinson’s
disease.
IBM Research - Haifa scientists who contributed to this project include: Moran Gavish, Ashraf Haib, Einat
Kermany, Lior Limonad, Ari Volcoff
Labels: ADHD, analytics, augmented reality, IBM Research - Haifa, machine learning, wearables