Click image for larger version

Name:	untitled.png
Views:	463
Size:	9.5 KB
ID:	48821

Brand new Leeds-based animal healthcare research and development startup, Vet-AI, has ambitious plans to overhaul the veterinary industry for the better using artificial intelligence. I’m lucky enough to have become its chief data scientist during its inception, and my task in the next few hundred words is to explain how we’ll do it.

Vets are highly trained, intellectually versatile, compassionate, knowledgeable, overworked, and rare. The UK vet industry is a patchwork of big corporations and smaller private practices, and practice management systems containing patient records are bespoke to each. Clinical notes are fragmented across practices, have little common structure, and are therefore resistant to standardisation. Consequently, all of this expert insight is tucked away in disconnected pockets only accessible through expensive face-to-face appointments.

Vet-AI has ambitions to build nothing less than the biggest cloud database of domestic pet data on the planet by providing a global gateway to easy, accessible and affordable veterinary care. We are beginning this journey with our flagship app, Joii, due for release in summer 2019. Joii will facilitate long-distance video consultations with vets, allowing anybody with a smartphone to access veterinary care. It is being built by vets, and we are conducting clinical trials under the close scrutiny of the Royal College of Veterinary Surgeons (RCVS) to ensure that remote services can be delivered safely. It will also give vets the option to work remotely, improving life-work balance while still receiving a proper salary. Pet owners will benefit from more convenient and less expensive access to vets, paving the way to greater overall animal welfare.

Unifying pet data has deeper scientific motivations, too, consistent with our longer-term ambitions. Healthcare data is what some scientists refer to as “large p, small n”. Vets consider lots of parameters, p, when assessing a patient, such as age, gender, breed, rabies vaccination status, whether it lives near deer and livestock or not, and so on. This means that in order to reliably learn the statistical trends in the data, the number of patients in the database, n, needs to be huge. Essentially, we are striving to massively increase n.

Specifically, it dramatically increases our ability to learn the structure of statistical dependencies throughout the population. This unlocks new pathways to scientific discovery in animal healthcare. Some of these relationships are obvious. Certain breeds, for example, are lighter than others, so there is a statistical dependency between breed and weight, because knowing a dog’s breed influences your belief about what its weight might be.

Some dependencies are much more subtle, or even downright startling. Scientists at Stanford University famously demonstrated in 2017 that sexuality can be detected from a picture of a human’s face using a deep convolutional neural network. This came as a huge surprise. And that’s the whole point. By bringing together rich sources of data, we will be creating a new ecosystem of scientific discovery with countless avenues for data scientists to explore and discover new and sometimes unexpected connections between categorical, numerical, image, audio, video and motion data, leading ultimately, we hope, to novel and anticipatory treatments and intervention strategies.

The crux of AI – the bit that makes machines appear intelligent – is achieved through a process called statistical generalisation. This means building a mathematical model that learns from existing data in such a way that it is a good fit to new data that has not yet arrived. A toddler’s brain demonstrates generalisation: show them two pictures of giraffes, and they can then identify new giraffes – which look slightly different – that they have not yet seen. Increasing the volume and comprehensiveness of pet data will allow us to develop and train cutting-edge AI models that generalise to the trends and patterns in the pet population, including those pets not registered in the app. Our ultimate goal is to use AI in this way to bring veterinary care into the era of predictive and preventative medicine.