FilterNet?: Unlocking Our Pets' Secrets

Our pets are unique. No two of them are the same. There are hundreds of breeds forming a complex matrix of pet types that continually evolve. And this brings with it a set of challenges and questions all pet parents face. How do we select the right care for our pet family when each comes with a set of nuanced needs? How do we give our pets the best chance at the healthiest and happiest life they can live? 

In our pet parenting journey, many of us will face the hard situation when knowing our pet is ill comes a little too late. As they cannot tell us when they feel sick or when they’re in pain, often we discover that something is wrong when their condition becomes serious. During this lapse between their first symptoms and the time when we eventually seek help, their issues can progress, leading to worse health outcomes that drive up the cost of care.

We know that preventive health is key for improving the quality of life for humans and animals alike. This means taking measures to pick up on early signs of disease and taking action to prolong positive health status, delay disease onset and lessen potential impact. So, the question is, how we do enter into a new era of preventative care that empowers us to give our pets the best life possible while keeping the cost of healthcare from ballooning? To accomplish this, we at Kinship are tapping into the immense computational power of machine learning and big data to identify behavioral patterns associated with early signs of disease. 

The minds behind the solution lie in the Pet Insight Project, a pioneering study involving nearly 16 billion minutes of data from more than 11 million days in dogs’ lives to pinpoint behaviors linked to health issues. In collaboration with Whistle Labs, the leading pet location and health wearable, Pet Insight data scientists are pairing data collected from Whistle FIT activity monitors with videos captured by citizen scientist volunteers. We then combine those behavior predictions with participating pets' medical records from Banfield Pet Hospital? and interpret the health stories of hundreds of thousands of dogs. The massive data set we’ve amassed through Pet Insight has already allowed us to observe a wide array of medical conditions, including more than 1,500 dermatological issues and 950 ear infections. More importantly, the unique scale of Pet Insight Project is enabling us to train algorithms to accurately predict health issues with higher confidence. However, to analyze this scale of data as quickly as needed without compromising accuracy required a ground-breaking approach.

Enter the brilliant Pet Insight’s science team. Led by Lead Data Scientist, Rob Chambers and Senior Principal Engineer and Data Scientist, Nate Yoder, this team developed FilterNet?, a flexible deep learning architecture that digests data to classify lifetime events. Additionally, it incorporates ideas from existing models and combines them in a way that’s more accurate and 100 times faster than the standard model. This makes it possible for researchers to process a full day’s worth of data in just a few seconds, where older models would take 20 to 30 minutes – a very expensive amount of computing time.

I like comparing the FilterNet model to viewing a large, very detailed mural up close. The way we used to do this was by standing just inches away, where we could see all the individual brushstrokes in minute detail but couldn’t really see the full picture all at once. FilterNet allows us to take ten steps back where we can still see all the finer details and also see the complete picture to understand what the brushstrokes mean together. For the first time, we have the ability to collect and analyze around-the-clock behavioral data on our pets. This “full-life biomonitoring” approach gives us a 24/7 view of an animal’s well-being, so we not only collect information, but convert it into clinically useful and virtually real-time insights. Pet Insight’s algorithms, in essence give any dog wearing a Whistle device, a “voice” that may allow pet parent and the veterinarian to provide better care than ever before. This is invaluable to pet owners, veterinarians and researchers, and has massive potential beyond veterinary care, as well.

While we are still in the early stages of what FilterNet can accomplish, and there is still more research to be done, our goal is clear – as many happy and healthy years with our pets as possible. I am so proud of the team leading this charge, and the role that Kinship is playing in ushering in this new era. 

If you’re interested in all the technical details of how FilterNet works you can download our initial FilterNet paper here and if you are interested in more information about the transformative power of the Pet Insight Project, please see our website.