Technology in School

How AI and Eye Tracking Could Soon Help Schools Screen for Dyslexia

By Stephen Noonoo     Nov 28, 2017

How AI and Eye Tracking Could Soon Help Schools Screen for Dyslexia

In an era of breakneck change and tech innovation, evaluating dyslexia in young students looks much the same today as it has in the past: A struggling reader’s parents and teachers might sit down, gather information and assess the child on their strengths and weaknesses to determine a diagnosis and appropriate interventions.

Often this is done via paper tests—despite the growing usage of predictive analytics in schools, where there are seemingly as many data dashboards as students in a classroom. All that’s to say, it seems like an industry almost too tempting for deep-pocketed tech investors and an ambitious startup with an eye on using machine learning to trim the fat.

“Today’s methods are quite cumbersome,” explains Frederik Wetterhall, the CEO and co-founder of Lexplore, a company that has devised a dyslexia screening tool that pairs eye tracking cameras with AI and algorithms. “With paper- and pen-based tests, it’s quite hard to read the results and takes a lot of time. [Educators] ask, ‘Who are the kids we think have difficulties?’ and they miss a lot of kids.”

Wetterhall’s company opts for a different approach, screening every student using computers and eye tracking cameras in an effort to find the few that might have dyslexic tendencies. Already, it’s caught the attention of investors, who just this March injected $5.6 million into the company in a funding round led by Gabriel Urwitz, CEO of private equity group Segulah, specifically to help it expand into U.S. schools.

Currently, Lexplore has an established presence in its native Sweden, where it’s used across Stockholm’s municipal education board to help identify kids who may be dyslexic as early as first grade.

“It’s a screening tool, not a diagnostic tool,” clarifies Wetterhall. “The main purpose is to find kids that are struggling with reading early on.”

From Research to Reality

Lexplore claims its technology is new—particularly the algorithm that separates typical from atypical readers. But the concepts it’s based on aren’t. Its tech draws from a deep well of previously-conducted research stretching back decades, which is generally supportive of using a combination of eye tracking and machine learning to screen for dyslexia.

“Eye movements is one of the best ways to index reading ability at an incredibly in-depth level,” says Julie Kirkby, a psychology professor at Bournemouth University in the United Kingdom, who has studied eye tracking and dyslexia for years.

One study from 2015 reported being able to identify dyslexic readers using eye tracking with 80 percent accuracy. That study used different cameras and methods, but Lexplore claims its technology can do even better, achieving 95 percent accuracy in a study conducted by two of the company’s co-founders immediately prior to the launch of the company, originally called Optolexia.

Tools like Lexplore make their analysis by tracing how a reader’s eyes follow words in sequential (or non-sequential) order looking for patterns. Readers at low risk for dyslexia tend to make more progressive movements—from left to right—as they scan words. Dyslexic readers on the other hand make more regressive movements—right to left movements—and do not make regular pauses (or fixations) during reading in the same way non-dyslexic readers do.

A comparison of eye movements between two readers (source: Lexplore)

Claims like Lexplore’s 95 percent accuracy seem a little high but are possible, Kirkby acknowledges, since research has shown that eye trackers can pick up on these reading differences. While researchers are torn on the exact reasons for these differences, they are likely a "product of reading disability rather than a cause," Kirkby says.

For Lexplore, the challenge has been refining this research into something user-friendly and inexpensive enough to make it attractive for schools, particularly ones that screen for dyslexia in traditional ways. According to Wetterhall, his technology is impartial and not subjective the way teachers and paper-based evaluations are. Ultimately, he says, that advantage may help screen at-risk kids faster.

“We have a high balanced sensitivity and specificity,” says Wetterhall. “We are equally good at finding kids with dyslexia and also excluding kids.”

Kirkby, who reviewed the Lexplore research at the request of EdSurge, was most surprised that the study was able to achieve its results using lower-frequency trackers. The data collected might not be suitable for high-quality research of the kind she typically helps conduct, but such innovations could end up making eye-tracking technology more affordable for schools.

Tracking the Market

Lexplore hasn’t made many inroads into U.S. schools, but it has done some preliminary work with a handful of private institutions in the Atlanta area. Among them is the private Galloway School in the tony Buckhead neighborhood, where tuition runs more than $20,000 year. There, the company came to the attention of the school after a suggestion from the parents of a Swedish student. As a result, Galloway became one of the first stateside schools to test drive Lexplore’s technology as part of a research pilot to see how the technology would translate from Swedish to English readers.

At Galloway, Lexplore representatives guided about 200 kids in grades 1-4 through the brief screening process. Kids read two short texts on a computer as a small, mounted eye tracking camera scanned and recorded their eye movements and uploaded it to the company’s database, along with some basic identifying information such as their name and age. That data was then uploaded to the cloud on a Microsoft Azure-based platform. Early in Lexplore’s development, Microsoft helped the company scale its product and now works with them in a sort of informal partnership, assisting with marketing and lead generation.

According to Wetterhall, the collected data is only used to sharpen its machine learning algorithm, which analyzes each child’s eye movements looking for patterns and abnormalities. “We have the data to help us develop our method further,” Wetterhall says, “but when we store it, it’s not connected with personal info, just a birthdate.”

After the screening, Lexplore made results available to the school and asked for feedback. “The kids were in and out in a very quick amount of time, and they were not stressed at all about it, like a regular test,” says Polly Williams, a principal at Galloway’s primary school. “One of the things I asked for was to be able to see a visual comparison of a child who was struggling versus a regular child, side-by-side. I thought that would have been good to show parents.”

Williams says she was impressed, and thinks the service could be significantly cheaper than current screening methods, although pricing hasn’t been set. But she isn’t sure yet whether her school will continue using it. “I think they’re still working out their business model and really identifying how to position and market themselves,” she says. “I don’t think that’s quite figured out yet.”

Right now, Williams says the company offers a couple of different models: a white glove package where company reps come to the school to conduct screenings; an option where schools purchase the hardware and software along with some support; and a do-it-yourself route that includes only the equipment.

It’s that last model that raises eyebrows for Kirkby, who cautions that educators are not necessarily scientists or researchers and may not have the resources to understand how to make the most of their results.

“If they can afford to dedicate someone to understand that data, I think that could present them with a really interesting way of getting more from their interventions,” she says. “Because then they can design an intervention and test a before-and-after scenario. But the schools would have to invest in that knowledge.”

Technology in School

How AI and Eye Tracking Could Soon Help Schools Screen for Dyslexia

By Stephen Noonoo     Nov 28, 2017

How AI and Eye Tracking Could Soon Help Schools Screen for Dyslexia

In an era of breakneck change and tech innovation, evaluating dyslexia in young students looks much the same today as it has in the past: A struggling reader’s parents and teachers might sit down, gather information and assess the child on their strengths and weaknesses to determine a diagnosis and appropriate interventions.

Often this is done via paper tests—despite the growing usage of predictive analytics in schools, where there are seemingly as many data dashboards as students in a classroom. All that’s to say, it seems like an industry almost too tempting for deep-pocketed tech investors and an ambitious startup with an eye on using machine learning to trim the fat.

“Today’s methods are quite cumbersome,” explains Frederik Wetterhall, the CEO and co-founder of Lexplore, a company that has devised a dyslexia screening tool that pairs eye tracking cameras with AI and algorithms. “With paper- and pen-based tests, it’s quite hard to read the results and takes a lot of time. [Educators] ask, ‘Who are the kids we think have difficulties?’ and they miss a lot of kids.”

Wetterhall’s company opts for a different approach, screening every student using computers and eye tracking cameras in an effort to find the few that might have dyslexic tendencies. Already, it’s caught the attention of investors, who just this March injected $5.6 million into the company in a funding round led by Gabriel Urwitz, CEO of private equity group Segulah, specifically to help it expand into U.S. schools.

Currently, Lexplore has an established presence in its native Sweden, where it’s used across Stockholm’s municipal education board to help identify kids who may be dyslexic as early as first grade.

“It’s a screening tool, not a diagnostic tool,” clarifies Wetterhall. “The main purpose is to find kids that are struggling with reading early on.”

From Research to Reality

Lexplore claims its technology is new—particularly the algorithm that separates typical from atypical readers. But the concepts it’s based on aren’t. Its tech draws from a deep well of previously-conducted research stretching back decades, which is generally supportive of using a combination of eye tracking and machine learning to screen for dyslexia.

“Eye movements is one of the best ways to index reading ability at an incredibly in-depth level,” says Julie Kirkby, a psychology professor at Bournemouth University in the United Kingdom, who has studied eye tracking and dyslexia for years.

One study from 2015 reported being able to identify dyslexic readers using eye tracking with 80 percent accuracy. That study used different cameras and methods, but Lexplore claims its technology can do even better, achieving 95 percent accuracy in a study conducted by two of the company’s co-founders immediately prior to the launch of the company, originally called Optolexia.

Tools like Lexplore make their analysis by tracing how a reader’s eyes follow words in sequential (or non-sequential) order looking for patterns. Readers at low risk for dyslexia tend to make more progressive movements—from left to right—as they scan words. Dyslexic readers on the other hand make more regressive movements—right to left movements—and do not make regular pauses (or fixations) during reading in the same way non-dyslexic readers do.

A comparison of eye movements between two readers (source: Lexplore)

Claims like Lexplore’s 95 percent accuracy seem a little high but are possible, Kirkby acknowledges, since research has shown that eye trackers can pick up on these reading differences. While researchers are torn on the exact reasons for these differences, they are likely a "product of reading disability rather than a cause," Kirkby says.

For Lexplore, the challenge has been refining this research into something user-friendly and inexpensive enough to make it attractive for schools, particularly ones that screen for dyslexia in traditional ways. According to Wetterhall, his technology is impartial and not subjective the way teachers and paper-based evaluations are. Ultimately, he says, that advantage may help screen at-risk kids faster.

“We have a high balanced sensitivity and specificity,” says Wetterhall. “We are equally good at finding kids with dyslexia and also excluding kids.”

Kirkby, who reviewed the Lexplore research at the request of EdSurge, was most surprised that the study was able to achieve its results using lower-frequency trackers. The data collected might not be suitable for high-quality research of the kind she typically helps conduct, but such innovations could end up making eye-tracking technology more affordable for schools.

Tracking the Market

Lexplore hasn’t made many inroads into U.S. schools, but it has done some preliminary work with a handful of private institutions in the Atlanta area. Among them is the private Galloway School in the tony Buckhead neighborhood, where tuition runs more than $20,000 year. There, the company came to the attention of the school after a suggestion from the parents of a Swedish student. As a result, Galloway became one of the first stateside schools to test drive Lexplore’s technology as part of a research pilot to see how the technology would translate from Swedish to English readers.

At Galloway, Lexplore representatives guided about 200 kids in grades 1-4 through the brief screening process. Kids read two short texts on a computer as a small, mounted eye tracking camera scanned and recorded their eye movements and uploaded it to the company’s database, along with some basic identifying information such as their name and age. That data was then uploaded to the cloud on a Microsoft Azure-based platform. Early in Lexplore’s development, Microsoft helped the company scale its product and now works with them in a sort of informal partnership, assisting with marketing and lead generation.

According to Wetterhall, the collected data is only used to sharpen its machine learning algorithm, which analyzes each child’s eye movements looking for patterns and abnormalities. “We have the data to help us develop our method further,” Wetterhall says, “but when we store it, it’s not connected with personal info, just a birthdate.”

After the screening, Lexplore made results available to the school and asked for feedback. “The kids were in and out in a very quick amount of time, and they were not stressed at all about it, like a regular test,” says Polly Williams, a principal at Galloway’s primary school. “One of the things I asked for was to be able to see a visual comparison of a child who was struggling versus a regular child, side-by-side. I thought that would have been good to show parents.”

Williams says she was impressed, and thinks the service could be significantly cheaper than current screening methods, although pricing hasn’t been set. But she isn’t sure yet whether her school will continue using it. “I think they’re still working out their business model and really identifying how to position and market themselves,” she says. “I don’t think that’s quite figured out yet.”

Right now, Williams says the company offers a couple of different models: a white glove package where company reps come to the school to conduct screenings; an option where schools purchase the hardware and software along with some support; and a do-it-yourself route that includes only the equipment.

It’s that last model that raises eyebrows for Kirkby, who cautions that educators are not necessarily scientists or researchers and may not have the resources to understand how to make the most of their results.

“If they can afford to dedicate someone to understand that data, I think that could present them with a really interesting way of getting more from their interventions,” she says. “Because then they can design an intervention and test a before-and-after scenario. But the schools would have to invest in that knowledge.”

STAY UP TO DATE ON EDTECH
News, research, and opportunities - sent weekly.
STAY UP TO DATE ON EDTECH
News, research, and opportunities - sent weekly.