in association with
AI for sight: The technology transforming infant eye disorder diagnosis
Vision is one of the most important tools for early childhood development, making the diagnosis and treatment of visual impairment in babies essential to their long-term health and wellbeing. But diagnosis in pre-verbal infants is a challenging task. Lucy Ingham hears from Victoria Pueyo, pediatric ophthalmologist and co-founder of DIVE Medical, about how her company’s TrackAI project is taking on the challenge with AI.
or babies, the sense of sight is one of the most important tools in their development toolbox. It provides key stimuli that help with many aspects of their development, building a foundation for their growth towards adulthood.
“We need vision to figure out how life works; to interact with others and even to understand feelings,” explains Victoria Pueyo, pediatric ophthalmologist and co-founder of DIVE Medical, during a talk at Huawei Connect 2019, a conference held in Shanghai, China.
For babies struggling with visual impairments, a lack of effective sight can be devastating.
“Visually impairment will interfere with general development and social life. And in certain settings, up to 60% of the children becoming blind will die during the following two years,” explains Pueyo.
“And here comes the most striking fact for me: 70% of these children could have absolutely normal vision if the problem had been detected and treated early enough.”
But detecting visual impairments in babies can be an immense challenge, particularly given they are unable to communicate effectively.
“The problem is that congenital or slowly progressing visual disorders may go unnoticed in very young children, since they are used to it and they are able to adapt to almost everything in life,” she says.
Parents may easily not notice if a baby’s sight is not developing as it should – in fact, problems are also easily missed by non-specialist medical practitioners.
“With the current tools only highly experienced ophthalmologists can assess visual function in very young children,” she explains.
“We show the children different types of objects or pictures and we observe the way they react. And based on the behaviours, we estimate if they are seeing properly or not.
“This is far from ideal and as a result of that, many pathologies escape undetected until it's too late.”
Using technology to improve visual screening
This problem is an issue being targeted by DIVE Medical, by developing tools that helps medical professionals that achieve more accurate diagnoses.
DIVE’s flagship product and namesake, Device for an Integral Visual Examination (DIVE) is combatting this using eye-tracking technology, which works on the basis that a child with normal vision will follow images in a different manner to those with healthy vision.
“We know that children with visual problems may look in a completely different way than children with normal vision. We can sit a child in front of a screen and present him with different visual stimuli,” explains Pueyo.
The DIVE eye-tracking system creates a map of how a child follows the visual stimuli, which medical professionals can use to determine whether the child’s gaze is within normal range, or if it is indicative of a particular visual problem developing. And if a child does have visual problems, the map “may look completely different” from a healthy child’s.
The company has produced two versions of its product, one for ophthalmologists and one of paediatricians. The former can test for colour perception, sensitivity to contrast, visual acuity, smooth pursuit and fixation, all of which can indicate particular vision problems. This enables accurate diagnoses to be made.
The version for paediatricians, meanwhile, is designed to provide rapid screening for referral to a specialist, and so just provides texts for fixation, smooth pursuit and visual acuity.
However, in both cases the products are designed to be primarily used in pre-established medical settings, which is why the company has also developed TrackAI.
TrackAI: democratising access to vision treatment
A partnership between DIVE and mobile device manufacturer Huawei, TrackAI brings the eye-tracking capabilities of DIVE to a portable setting, allowing it to be used in areas without access to robust medcial facilities or high levels of medical expertise.
“Visual screening programmes are the best tool to decrease the rate of undetected visual problems in children and therefore to decrease the rate of later visual impairment, not only in childhood but also in adult life,” says Pueyo.
“So we need to empower non-trained people from the community to identify these children who may have risk of having visual problems and to refer them to a specialist able to diagnose and treat them properly.”
In order to widen the use of DIVE to areas in the developing world and other regions without significant medical facilities, the TrackAI project need to create“an easy-to-use, portable and affordable device able to detect these visual problems as soon as possible, ideally when they are only babies”.
The solution takes the form of a Huawei P30 smartphone that is paired with a DIVE device, which together replicate the DIVE setup in medical facilities, but in a manner that is far more portable and affordable to run.
“The doctor will use the P30 to control the whole system and also to manage data collection, which will be done with the DIVE device,” explains Pueyo.
However, the company has also integrated further capabilities that puts much of the analysis of the results into the hands of AI, enabling it to be used by people who are not trained medical professionals.
“DIVE's device presents several visual stimuli on a high-resolution screen from Huawei's Matebook E laptop, and captures the gaze of the children using an eye tracker,” she says.
“But analysing gaze information is a really challenging task. So starting from the DIVE platform and from the gaze analysing software, we are now incorporating artificial intelligence thanks to Huawei's HiAI 2.0 platform.”
This enables the results of the eye-tracking tests to be automatically analysed, even producing a likely diagnosis in many cases.
“We are now able to find patterns in the way the children look at the pictures and to relate them to visual development and to even to different pathologies,” says Pueyo.
“Running the artificial intelligence model in the P30 enables, in real-time, an accurate and reliable screening of the visual functions, and this combination unleashes the potential not only to measure visual function in a really accurate way, but also to estimate the probability for the patients to have certain visual pathologies.”
Bringing TrackAI to the world
While the technology is making significant strides, there is still some way ahead before TrackAI can be rolled out around the world.
One of the most important issues is the collection of training data: the information used to build the AI and ensure that its conclusions are accurate.
“Developing a system based on artificial intelligence requires information from thousands and thousands of patients,” explains Pueyo.
This is an ongoing effort, involving ophthalmologists at a host of facilities across five different countries around the world, with data being collected both on infants with healthy vision and those with a wide range of conditions.
“All this information gathered during the data collection is being used to train the neural network,” she says. “We have already examined more than 1,500 children of different ages and ethnicities.”
This has been enough to build a whole system prototype that is currently undergoing testing, which Pueyo hopes will lead to a final product that can be proliferated in a host of different medical settings.
“The unprecedented capabilities of the Huawei P30 enables DIVE's device to reach out and to be useful for many people across the world, from an ophthalmologist in a hospital in the UK to a family doctor in a remote place in Mexico,” she says.
It will also allow the technology to be brought to areas without any form of medical facilities.
“Now that we are close to having a functioning and validated system, able to run an accurate visual screening in very young children from six months of age, we are already thinking about the next step,” she says.
“And this is collaborating with different NGOs and social entities to bring this technology to many places excluded from healthcare. We are already working on it and I think we will be able to start some of these projects during next year 2020.
“We are facing a high-impact challenge, and the results of this project will indeed change the lives of many children around the world.”
Share this article