by George Nott

Facial recognition algorithms promise early autism detection in children

Oct 18, 2017
Artificial IntelligenceCollaboration SoftwareHealthcare Industry

Perth children’s health research institute Telethon Kids has partnered with NEC Australia to explore how algorithms for eye tracking and facial recognition could be used to develop autism detection and diagnosis tools.

The institute will also the technologies’ potential in the early stage identification of Foetal Alcohol Spectrum Disorders.

In both cases, early detection allows health professionals and parents to implement changes and strategies to better manage the conditions. The work also hopes to identify methods to reduce the severity of the condition in children as they grow up.

The partnership will build on work already underway at the institute into how facial features work as indicators of brain development.

“The face is what we think of as a readout for the brain, because the face develops at the same time as the brain develops in pregnancy. Autism isn’t traditionally thought to have a distinctive set of facial features, however we do think there are some kids who might have a subset of features that might indicate genetic differences or prenatal influences,” explains Dr Gail Alvares from Telethon Kids.

“It does sound a lot like phrenology. The difference is we have a substantial amount of biological mechanisms we think actually explain these facial readouts.”

There is also evidence that the way a child visually takes in a face can be an indicator of autism. Eye-tracking technology may allow the researchers to identify the disorder far sooner than currently.

“We’re looking at the use of eye-tracking as a potential predictive marker that might indicate differences in visual attention well before you see actual behavioural differences,” Alvares said.

NEC Australia and the Telethon Kids Institute have agreed to jointly own any intellectual property created under the agreement.

“The Telethon Kids Institute has conductedground-breaking researchinto using facial features for diagnosing autism. We’re at the beginning of our work together but we’re very keen to explore how we can assist and collaborate where possible. For us it has the potential to test and demonstrate completely new applications of our technology in a way that can help improve people’s lives,” said Mike Barber, chief operating officer at NEC Australia.

The institute is currently moving into its new home within Perth Children’s Hospital. The co-location with Western Australia’s only paediatric tertiary hospital will enhance clinical collaboration to translate research into policy and practice.

Telethon Kids’ autism research team investigates the genetic and neurobiological causes of Autism spectrum disorders which affect around 125,000 people in Australia.

Recognising potential

The application of algorithms to encode features of children’s faces for medical analysis is already showing promise.

“We know that there are some kids with autism who have more hyper-masculinised faces, or some kids who might have more asymmetric faces. Traditionally we’ve only been able to assess these by physically measuring kids, actually running around our clinic rooms with a little measuring tape trying to get these measurements, and with a greater degree of error. With 3D cameras we can measure down to the sub millimetre level,” Alvares said.

In August, Telethon Kids and researchers from The University of Western Australia and Princess Margaret Hospital for Children, used 3D photogrammetry to examine whether pre-pubescent boys and girls with an Autism Spectrum Disorder (ASD) displayed more masculine features compared to those without the condition.

Algorithms were used to generate a gender score for the 3D facial images. The gender scores were based on an analysis of 11 facial features such as breadth of a person’s nose, distance between the outer corners of the eyes, upper lip height and width of the mouth and were compared between an autistic group and a control group.

For each sex, increased facial masculinity was observed in the ASD group compared to the control group.

Further analysis – published in Nature – revealed that increased facial masculinity in the group with autism correlated with more social communication difficulties as measured on the Autism Diagnostic Observation Scale.

“Our next step is to understand how specific these findings are to children with autism. If they are specific, then the findings raise the tantalising possibility that facial characteristics may in the future be one feature that helps us identify children with autism earlier than we currently do which is a truly exciting advancement,” said Professor Andrew Whitehouse at the time.

Ethical questions

Despite its huge promise, the use of facial recognition and analysis algorithms to detect traits has proven controversial in some cases.

In September, researchers from Stanford University in the US, developed an algorithm which could correctly distinguish between gay and straight men 81 per cent of the time, and 74 per cent for women.

The study raised some serious ethical questions about the risk of such algorithms.

“Given that companies and governments are increasingly using computer vision algorithms to detect people’s intimate traits, our findings expose a threat to the privacy and safety of gay men and women,” the researchers noted.

Others called the study itself an “abuse of the public’s trust in science and in mathematics”.