5 min read
The future of pet care? How Vet AI is training its AI to diagnose symptoms in animals
Founded in 2017 by tech entrepreneur Paul Hallett, experienced vet Robert Dawson and vet dermatologist Sarah Warren, the early-stage company is looking to bring vet care in line with some of the groundbreaking innovations seen in the human health sector.
Vet AI recently partnered with pet insurer Animal Friends to offer its customers free access to its smartphone app. Called Joii, it lets owners check their animal’s symptoms and receive instant advice from qualified vets on treatment through free online video consultations, which would usually cost £20 per call. Joii will advise users to visit a vet clinic if necessary, or it may decide that no action needs to be taken.
The company has collaborated with a number of partners both internally and externally to develop its offering. They range from Leeds-based user experience and behavioural research consultancy SimpleUsability to the Knowledge Transfer Partnership and multiple Leeds University departments – including its School of Computing, School of Mathematics and School of Electrical Engineering.
Placing any amount of trust in an app would prove a seismic shift for the estimated 12 million households that collectively own 52 million pets in the UK. In a 2015 poll by opinion researcher ICM, 94% of 2,000 respondents said that they trust vets more than their GPs and dentists.
However, it could also alleviate any fears of being judged; according to research conducted by Legal & General earlier this year, 1 in 4 (or 27%) of UK dog owners (equivalent to 2.4 million people) surveyed said they worried that their vet would personally judge how well they looked after their animals.
Vet AI’s team of vets, data scientists and product developers are confident that they can challenge the status quo using their proprietary tech to overhaul the industry. But how does it work?
At a recent event held at Leeds University’s Nexus tech hub, where Vet AI’s team is based, the company’s Chief Data Scientist Trevor Hardcastle shed light on how Joii combines AI, computer vision and machine learning in an attempt to diagnose dermatological and inflammatory conditions in animals.
It has been designed as a suite of tools for making vets’ lives easier. “We don’t want to go in guns blazing, issuing diagnoses left, right and centre using an AI intelligence model,” says Trevor. “We think they still need to be signed off by a human being.”
Joii begins by mimicking the process of a human clinical examination. “The first thing that any medical professional does when a patient walks in the room is look at them, so that’s the first thing we wanted to automate,” says Trevor.
As this works by diagnosing problems which are externally visible, users are required to upload a picture of their pet’s ailment or injury to be examined by Joii’s AI, which taps into existing datasets (consisting of images and supporting textual information) to highlight areas that need further attention.
Trevor provides an example of an injured dog’s paw to demonstrate how Joii identified interdigital dermatitis (inflammation of the paw) from a photo. Stats show that vets are most frequently visited by owners whose dogs have injured themselves after stepping on something, and this has resulted in a rich database of images of damaged paws that can be used to train the company’s AI.
The human touch
In training the AI to identify a range of dermatological ailments, the company undergoes what it calls “supervised learning”. This sees it bring in human vets, such as Vet AI’s cofounder and veterinary dermatologist Sarah Warren, to programme into the AI extra information that – alongside an image – would help it identify an ailment. In the example of the dog with the injured paw, that would be a list of underlying diseases that cause lesions.
“Contextual information is really important when diagnosing a patient – it’s not just what you see, it’s what you know about that patient that informs your decision as to what the likely diagnosis is,” says Trevor.
The company isn’t stopping at still photos. By analysing a video of an animal walking, Joii’s AI can also determine whether or not it has a condition such as arthritis – including how bad it is and which joints are affected – all within 10 seconds.
Of course, Vet AI has considered the implications of its AI making a misdiagnosis. When attempting a diagnosis, Joii’s AI adds a weighting to every potential illness, which has been fine-tuned by a veterinary professional, that could be affecting the animal. If the AI is confident enough that issuing a misdiagnosis would not result in disaster for the animal, it will issue a diagnosis. If it’s not confident enough, the pet owner is advised to see a vet.
If Vet AI has developed a solution that can earn the trust of pet owners across the country, if not the world, its team is facing an unprecedented and exciting opportunity – one that animals will appreciate too.