This proceeding is part of the 2016 Hill's Global Symposium. If you missed the live-streaming event, click here to check out the sessions for CE credit. You can also click here to view the digital proceedings flipbook.
Disruptive Technologies in Human Medicine
Soon, the world around us will know everything that is going on—our clothing will be taking a steady stream of health and activity data; our cars will know when and where to go and perform the driving for us; and our home will know what room we are in and what activity we are doing. We see some of these devices on the market now, but eventually we will see these devices working together with each other, with others in the population, and with our environment.
We have become dependent on technology. In a study performed by Braun Research, Inc. for Bank of America in 2015, 80% of people reach for their mobile phone first thing in the morning, (often ahead of coffee and their toothbrush!)1, and 44% could not last a day without it.2 Gallup says that 81% of us keep our smartphone with us all day.3
Focusing on medicine, wearables are starting to appear with mobile health sensors, smart fabrics, precision medicine, and medical decision-making, where data is generated and sent to enterprise systems to give a complete picture of the wearer’s health and environment. Twenty percent of people either already own or would consider buying wearable technology. In addition, the top three reasons people want wearables are health-related. 77% want them to assist with exercise; 75% want them for tracking medical information; and 67% want them to assist with better eating habits.4 In addition, “62% of smartphone owners have used their phone in the past year to look up information about a health condition.”5
Do pet owners feel the same way about health data for their companion animals as they do about health data for themselves? The answer is yes. In 2014, according to Grand View Research, the worldwide pet wearable market was valued at 837.6 million dollars, and the compound annual growth rate specifically for medical applications is estimated to grow 20% from 2015-2022. Most of these wearables, such as WUF, FitBark and Whistle, are for GPS tracking or energy-level monitoring, while some, like Voyce and PetPace, track vitals such as temperature and heart rate.6
In February of this year, Merck Animal Health announced a diabetes tracker for pets, allowing the pet owner to set dose reminders and track insulin given.7 While it is not a wearable, it is a mobile means of tracking your pet’s health data.
How Vetrax is different?
Vetrax is a medical device and information platform designed to more quickly alert veterinarians and pet owners to potential health concerns as part of an ongoing monitoring program. On the surface, activity monitoring sounds similar to what Whistle and other companies offer. However, the sensor data is taken at a higher frequency than others on the market. The higher frequency allows a better and more complete understanding of the data.
For example, look at the data given in the figure below. Data for a given behavior might look like the sine wave given in (a). A standard signal coming from standard energy-level monitors might look like (b) where much of the features of the data are missed because the data is taken at a low frequency intervals represented by the red dots. Because Vetrax samples at a much higher frequency, the data signal features are collected and analyzed using all of the characteristics of the signal, as shown by the blue dots in (c).
In addition, currently-available sensors for pets use a single number to determine low, medium or high energy levels. Vetrax is collecting data in three axes. The three axes and higher frequency together allow differentiation of behaviors that would otherwise be missed by Whistle and others. In the figure below, real data collected from dogs with the Vetrax sensor is shown for one axis data (a) and three axis data (b). When comparing single axis data with the three axis data, it is clear that different behaviors are discernable. Some of the visible differences are that the three axes data are farther apart for excreting than they are for drinking; Scratching has a stronger periodicity than walking. With one axis data, these differentiators are not visible.
To optimize algorithms that can differentiate between the various behaviors with the Vetrax data, Georgia Tech is leveraging a unique genetic programming tool with its roots in military algorithm development, Georgia Tech Multiple Objective Programming (GTMOEP). Genetic programming is a bio-inspired approach that allows computers to create algorithms. Traditional genetic programming only supports the use of arithmetic and logical operators on scalar features.
The GTMOEP framework builds upon this traditional programming by also handling feature vectors, allowing the use of signal processing, and machine learning functions as primitives in addition to the more conventional operators. GTMOEP is a novel method for automated, data-driven algorithm creation, capable of outperforming human derived solutions.8
Based on an analysis of frames of collected data, Vetrax is implementing a “rejection chain” algorithm, or series of algorithms (see the figure below). For each definitively identifiable behavior, such as shaking or scratching, the Vertax system will classify the behavior. Using this chain of algorithms, the identified behaviors are picked off, leaving the data to go through the identifying algorithm for the next behavior. This process repeats until all data has been specifically classified. If there is data left after passing through all behavior algorithms, this data will be identified as walk, rest, and run based on the overall energy of the data. This last step of identifying walk, rest, and run behavior from a magnitude of the data is where other companies begin and end their analysis.
Overall, the AGL has collected data on over 500 dogs. Algorithms were generated from datasets resulting from over 1,500 man-hours of detailed annotation work. Behavior Algorithm performance using these datasets have achieved a validated positive predictive value of approximately 79%. These algorithms should produce reliable trends to enhance the veterinarian’s capabilities in managing the health of their patients. Note, the Vetrax system is not a tool for diagnosing diseases. The system supports managing chronic health and wellness programs. The veterinarian will baseline an animal’s quantified behavior before a prescribed intervention (nutritional and/or pharmaceutical) and then observe the animal’s quantified behavior trend after the intervention.
In use, veterinarians prescribe Vetrax for dogs that need to be regularly monitored due to chronic illness or another health condition. The owner affixes the lightweight Vetrax sensor to the dog's collar. The sensor collects detailed data used to classify behaviors the dog is exhibiting—such as basic levels of activity (resting, walking, running) and more advanced activities such as shaking, scratching or drinking—and regularly transmit the information wirelessly to the Vetrax information cloud where the data can be reviewed by the veterinarian to see if the prescribed intervention is working as intended.