As a fundamental concept used in human gerontology, frailty refers to the gradual decline in physiological function with aging and the resulting vulnerability to adverse health outcomes. Assessment of frailty and using this for healthcare and resource management have garnered significant interest both to support the elderly and their caregivers. There have been substantial efforts to translate these concepts and outcomes into animal health. In this paper we present the preliminary effort towards a system to characterize functional muscle strength, an important component of frailty in dogs, using wireless and continuous electronic sensors. We have custom developed the front-end hardware with real time wireless data transfer capability with form and function specific for veterinary applications. In this system, frailty is investigated through continuous pulling force measurements collected on a leash-based device. The data can be monitored in real time over a smartphone and are also saved to the cloud for more complete offline analysis after the data collection session. We coupled the sensing system with a camera for synchronous recordings to further analyze the strength performance session footage. The next stages of this research will involve collecting data through a large cohort of dogs and using machine learning to fuse data and provide fragility assessment scores. This is our initial step towards a multimodal machine learning supported sensor system for both quantitative assessment and scoring of a component of frailty in dogs. The vision is to bring these measurements from veterinary environments to subjects homes for a more convenient and regular data collection.
1 Introduction
Frailty refers to a decline in the strength of physiological systems of the body, and is typically considered a qualitative aging effect. In recent years, there has been important work to characterize frailty in dogs, with one study assessing a 5-factor phenotype that included assessments of: weakness, slowness, poor endurance, low physical activity, and shrinking [3]. Frailty in dogs can be analyzed using performance in physical tests, handler questionnaires, and physical examinations. These metrics are evaluated based on an earlier baseline and time until all-cause mortality. As reported in [3] “Frailty status was significantly associated with all-cause mortailty, with median survival times of 10.5 months, 35.4 months, and 42.5 months, respectively for dogs with 3 or more criteria (frail dogs), dogs with 1 or 2 criteria (prefrail dogs), and nonfrail dogs”.
In addition there is interest in the translational potential of studying frailty in non-human animals to improve understanding of aging in humans [8]. There are significant areas of similarity between domestic dogs and humans such as environmental pollutants, lifestyle and nutrition, pathogens, and even their genomes and genetic diseases [2]. As methods for evaluating frailty are being refined, there is a recognized lack of performance test to quantify muscle strength in dogs [3],[4]. In humans, muscle strength, measured via grip strength, is an important measure with no corresponding test in dogs. Use of basic force sensing applications have been reported [3], such as using a dynamometer attached to a leash to capture a peak force during a weight pulling task. However, this area is not well validated. Custom designed force sensing systems have been used for working dog applications such as mushing (sled dog racing or towing) where handlers are interested in better monitoring of canine wellbeing and injury prevention [6]. Modifications of these systems have potential as clinically-expedient functional measures in dogs.
Motivated by the need for a clinically-relevant strength test for dogs, our goal in this effort is to present an electronic and computerized system that fits the clinical use case and provides quantitative and standardized data that may be used for diagnosis or monitoring. Our efforts also envision the incorporation of machine learning to automate the process in the future. This may help move some of the assessment from veterinary environments to subjects’ homes to reduce the financial cost, time burden and potential veterinary office stress in dogs. Beyond frailty, the demonstrated system is expected to appeal to a broader range of health or working applications involving canine biomechanics such as seeing eye dog training, neuromuscular pain assessment, and orthopedic disease, and surgical healing. The design process aims to satisfy the broader need of quantitative and continuous force measurement for canine well-being.
2 Design Process
The system we assembled consists of a tension sensor connected to a wireless microcontroller transceiver built into a leash anchored to a wall for measuring the pulling force of collared or harnessed animals. The integration of electronics into the leash form factor allows for application flexibility. The initial system was designed to collect tension data during experimental sessions in a veterinary office. The continuous data are transmitted over Bluetooth to smart devices while also being recorded internally on an SD card. Data collection is initiated over Bluetooth using the smart device which also displays real time data. Although a specific veterinary use case was targeted in this first step, the custom designed system is also adaptable for other applications with harnessed dogs and other animals. While the standard commercially available force measurement equipment is not typically adapted for wearability or for hands-off operation, we kept this as our most significant design criteria for this system—we aimed for this system to be just as easy to use as a conventional leash.
Figure 1:
Figure 1: High level diagram of leash sensor system.
2.1 Hardware
Figure 1 contains a high-level overview of leash-based tension sensor system, indicating the major components. Based on the anticipated use case, we selected a simple off-the-shelf elastic dog leash that would provide some reduction in harsh forces on the dog. The elasticity is not only of benefit to dog comfort, but also smoothes out and mechanically filters impulses in force across the time domain. This is more desirable because it can help to capture sudden movements from the dog resulting from lunging or leaping forward. The electronic components are integrated tightly with the leash to make it easy and safe to use, and maintain the strength and function of the original leash. The force sensor is placed in series with a conventional leash clip and a metal connector is instead sewn into the fabric of the leash. The wires that travel along the system are threaded within the elastic portion of the leash, and the microcontroller, onboard battery and other components are collected towards the handle in a 3D-printed enclosure. Figure 2 is a photo of the portion of the sensing system that clips to the dog’s harness or collar, illustrating the mechanical coupling.
Figure 2:
Figure 2: A photo of the completed assembly of the lower portion (canine side) of the leash sensor system.
The hardware components of the system and their arrangement are depicted in Figure 3. We use a commercially available inline compression and tension load cell sensor (DYMH-103 (100kg) from CALT, SHANGHAI QIYI CO. LTD, China). This particular device fits many of the requirements for a wearable system: it weighs only a few tens of grams, it can cover the entire range of anticipated forces (0-30KG), and has a fully enclosed casing that we can securely attach to our system. The load cell is connected to a 24-Bit Analog-to-Digital Converter (HX711 from Avia Semiconductor, China). This chip is commonly used with load cells, providing ADC conversion, gain, and a serial interface which we use to communicate with our microcontroller. We followed the data-sheet to provide necessary peripherals on the printed circuit board level to activate desired features such a higher sampling rate of 80 Hz.
We use an ‘all-in-one’ Arduino-compatible + Bluetooth Low Energy system-on-board with built-in USB and battery charging (Bluefruit M0 from Adafruit Industries, NY, USA). It is approximately 2.0" x 0.9" x 0.28" and 5.7 grams. It runs on a commercially available microcontroller at 48MHz (ATSAMD21G18 from Microchip Technology Inc, AZ, USA) and has several useful features such as built-in lithium polymer (Li-Po) charging and, a Bluetooth LE module (nRF51822 from Nordic Semiconductor, Norway). It also has a compatible app “BlueFruit Connect” for connecting to the device over Bluetooth using various operating systems including Android or iOS. It is also compatible with Adafruit’s lineup of ‘FeatherWings’ which are stacking expansion boards which allow customized choice of added features. The load cell amplifier feeds into the M0 which handles onboard processing, writing of data to the SD card and Bluetooth communication to mobile smart devices and/or laptops.
We used one of these expansion boards, the “Adafruit Featherwing Adalogger” which includes an SD card support and a real time clock module. This allows us to use on board SD storage as an additional format to record experimental data. In the event that Bluetooth connection is lost during data collection the data will persist on the SD card and the card can be used to transfer the data later. The real time clock module is also useful for timestamping the data points as they are recorded. Instead of marking data based on internal ticks, we can use universal time and make it easier to synchronize the data with footage or other sensor systems during data processing. In our testing we measured that the system consumes on average less than 50mA. We implemented a lightweight 1000mAh lithium polymer battery which means we can be confident in a multiple hour battery life while remaining extremely lightweight as a whole system. Since we use an external LiPo battery, it is very simple to replace the battery with one of larger or smaller capacity based on the user’s needs.
Figure 3:
Figure 3: A block diagram of the electronic components and the associated signal flow.
2.2 Software
The Bluefruit LE Connect app from Adafruit allows for basic communication between Bluetooth Low Energy devices and smartphones. It is available for both Android and iOS operating systems and in our experiments we used this app to monitor data in real time using the plotter or UART functions. The data can then be exported to several formats such as.csv. For all analysis performed, the data stored on the SD card is used preferentially, since it is not affected by BLE disconnects and possible loss of data.
3 Procedures
In order to validate our design and determine technical feasibility, we conducted a pilot study involving two volunteers and their owners. All procedures were reviewed and approved by the NC State University Institutional Animal Care and Use Committee (IACUC) and conducted under the supervision of a board certified veterinary behaviorist. Here we describe procedures in two ways: from the perspective of the participants and from the perspective of the study administrator operating the hardware and software systems.
3.1 Participant Procedures and Ethics
All preliminary tests were conducted in the presence of the volunteering dog owner and veterinary professionals. Dogs were allowed time to acclimate to the environment and persons present while volunteers were briefed on data collection procedure as well as the equipment to be used. Each data collection session only lasted a few minutes and was paused at any moment the dog displayed signs of discomfort or disinterest. Dog owners were welcome to take part in the administering of incentives (assortment of treats and praise) during the session. Otherwise this task was carried out by veterinary professionals. A researcher was tasked with monitoring the harness and equipment throughout the session. If there were any mechanical failures safety concerns, the researcher would be responsible to interrupt the session. Volunteers were also encouraged to stop the session at any point on behalf of themselves or their dog, whether to take breaks or verbally withdraw consent from the session completely.
3.2 Hardware Operation Procedure
The following steps are followed to operate the system during a data collection session:
(1)
Connect the battery to the microcontroller and close the plastic enclosure around the electronics. Make sure it is secured to the leash.
(2)
Open the BluefruitConnect app and connect to the leash system over Bluetooth; From here, either navigate to the UART screen to view the data and later export, or the plotter screen to view incoming data graphically.
(3)
Once connected successfully the data collection will begin; (Optional and recommended: Attach a calibration (known) weight to the leash for validation of scaling. For example a conventional 5lb iron gym weight. Remove weight before continuing with data collection.)
(4)
Connect the leash system to the canine harness
(5)
Set up the camera, start camera recording and carry on with the experiment.
(6)
When completed with the experimental session, follow the power-down instructions below.
(a)
Export data to.csv on the mobile app (Another copy should already be saved automtacially on the leash SD card)
(b)
Disconnect from the leash on the mobile app
(c)
Remove the battery connection to power down the leash completely
(d)
If needed, retrieve the SD card
(7)
Repeat the above procedure for additional dogs/sessions
Figure 4:
Figure 4: An example pull force plot from an entire experimental session.
Figure 5:
Figure 5: A still frame of the combined data and video from an experimental session. Session footage is synchronized with the pull force data stream.
Table 1:
Table 1: Compiled statistics for dogs in preliminary testing
3.3 Preliminary Testing
Preliminary testing with dogs was performed alongside collaborators at the North Carolina State University College of Veterinary Medicine. The subjects were a young female yellow lab and a young female miniature Australian shepherd. Both dogs were generally in good health and presumed non-frail. Approximately 10 tests were performed with each dog, while experimenting with different types of motivations (food, praise, etc.) and various tack (collars, harnesses, etc.). Figure 4 is a plot of data collected during on of the experimental sessions, and illustrates a range of patterns in tension measurements. Figure 5 is a screenshot of tool output that displays synchronized video and tension data during the session.
Collectively we came across the following observations and predictions: 1) The larger dog of similar health had significantly higher peak forces, therefore, forces will likely need to be normalized by weight in future sessions to glean variations due to frailty; 2) Collars that are not designed for pulling application tend to cause choking and gagging when dogs fully exert themselves against the leash, therefore we will equip dogs in future sessions with harnesses like the x-back style [7], designed to relocate harness loads to the front shoulder instead of the neck; 3) Dogs would sometimes get ‘turned around’ by the force of the leash, and this would often cause twisting of the harness, therefore, we modified the attachment point of the leash to the wall to make the height adjustable. This should reduce twisting by making sure the wall attachment point remains at least slightly above their center of mass. In addition, the x-back harnesses have leash attachment points further back by the hind quarters. All conventional harnesses and collars we initially used attach near the neck which likely contributed to dogs becoming turned around by the elastic leash; 4) Dogs would occasionally slip on the untextured lab floor. For more consistent results we placed a traction mat on the floor and saw no further slipping. This slipping may be exacerbated by the youthful and non-frail dogs in our initial testing but the traction mats will be carried into further testing to safeguard against this possibility; 5) Dogs may adopt a loading and lunging behavior in response to the resistance to reaching the motivating factor. It is important that we use elastic material in our leash to prevent dog injury and also prevent equipment destruction and eventual loss of data.
4 Data Analysis
The data system saves subsequent experiments as separate.csv files. The real time clock module of the system provides time stamping to each data point. This is to aid in later data processing and synchronization with other devices and camera recording footage. Currently, the two data streams are synchronized manually using visual cues.
A MATLAB (from Mathworks, MA, USA) script is used to process the pull force data of the session and can produce an animated line graph that scrolls with accurate speed and timing. The processing can also be performed programmatically in batches similarly to the data collection capabilities of the device. This results in a streamlined system where the veterinarian or operator of the device can in single steps collect and later process batches of data. The script is also designed to stitch together camera recording footage of the data collection with the animated force plot, as seen in Figure 5. This is a multimedia representation of a session and aims to offer a quantitative and more comprehensive record of the session than is otherwise possible.
Using this combined tool, we were able to discern four common patterns of behavior as depicted in Figure 6: a) a sustained pull, b) loading, c) lunging, and d) relaxed. Often times loading and lunging were interspersed, creating cyclical patterns of tension as can be observed in the tension plots of Figure 6 (B) & (C). It is an open question as to whether these behaviors can be detected and measured from tension data alone, or if additional sensing (e.g., an intertial measurement unit) or computer vision may be required.
In addition the following statistics were calculated for each dog based on the force data collected in the preliminary testing and compiled in Table 1: peak pulling force, average force, standard deviation, variance, and total session length. The statistics were calculated based on the compilation of all recorded sessions for that dog. Some initial sessions were not included for the female Labrador because of changes to collars, harness and other equipment such as the floor surface, in the interest of comfort and safety that interrupted the data collection. Based on the preliminary data analysis we can see that in the young and assumed able bodied dogs, there is a strong relation between the magnitude of pulling force (both peak and average) and the body mass of the dog. This is a relationship that we expected and designed the system to handle the force ranges of various strength and sized dogs. Another interesting result is that when we compensate for the body mass of the dog, the force/body-mass appears very similar. The sample size due to the preliminary nature of our testing is limited so these serve as some interesting features to analyze and potentially validate in future work.
Figure 6:
Figure 6: Several still frames of combined output tool illustrating several common canine actions and their associated effect on tension data: A) sustained pulling, B) loading, C) lunging, and D) relaxed.
5 Discussion and Future Work
This device will be used in an ongoing clinical trial at North Carolina State University College of Veterinary Medicine focused on tracking frailty and improving healthspan and well-being in dogs [1]. The study has received ethical approval through the NC State IACUC and enrolls volunteer dogs at multiple points throughout the year. In this randomized, controlled trial, enrolled dogs will be allocated to receive physical rehabilitation (intervention group) or continue their standard exercise (non-intervention group) and followed for six months. Dogs will have study visits at Month 0, Month 3, and Month 6 for evaluation of physical, neurologic, and orthopedic health, as well as assessments of frailty, gait, postural stability, and strength using our leash sensor system. Comparisons will be made between the intervention and non-intervention groups, and the predictive power of each measure in the dogs’ outcome will be assessed.
We are also in the process of incorporating inertial measurement units to the system as well as adapting it to perform left-right differential pulling force measurements, based on research showing unequal and biased forces on harnessed dogs [5]. Such a recording of force location and direction, with potential improvements in force resolution, could be useful in various other applications. In the raising of guide dogs, efforts are made to match handlers with dogs who share their preferences to how strong their on-leash/handle behavior is, or even directional preferences. If canine force sensing systems were made available, this characteristic could be quantified and used to streamline the process. Continuous force measurements can also be used to understand neuromechanics of aging and underlying pain processes as well as assessment of strength recovery following surgery. While these initial measurements are intended to be performed in controlled veterinary environments, the lower cost and user-friendly nature of the presented system would enable its use at home as well. When combined with machine learning based assessment and scoring, this system has a potential to help owners of aging dogs to track this important aspect of their dog’s frailty over time.
6 Conclusion
We presented the initial efforts towards quantitative assessment of strength in dogs. This initial step focused on constructing and assembling the hardware infrastructure using force sensors to characterize pull-forces. The combined camera and force measurements enable future incorporation of machine learning to automate the assessment. Beyond assessment of this important component of frailty, the system can also be used for other health and working dog applications. Finally, if this measure is comparable in predictive power to grip strength in humans, it will further the translational potential for the study of frailty in dogs.
Acknowledgments
The authors acknowledge the support from the United States National Science Foundation through the grant numbers EEC 1160483 (ASSIST Center), IIS 2037328 and EF 2319389. We share our deepest appreciation for all volunteering animals and handlers who participated in our experimental data collection.
Jessica M. Hoffman, Kate E. Creevy, Alexander Franks, Dan G. O’Neill, and Daniel E.L. Promislow. 2018. The companion dog as a model for human aging and mortality. Aging Cell 17 (6 2018). Issue 3.
Romane Lemaréchal, Sara Hoummady, Inès Barthélémy, Claude Muller, Julie Hua, Caroline Gilbert, and Loïc Desquilbet. 2023. Canine Model of Human Frailty: Adaptation of a Frailty Phenotype in Older Dogs. Journals of Gerontology - Series A Biological Sciences and Medical Sciences 78 (8 2023), 1355–1363. Issue 8.
Rachel L. Melvin, Audrey Ruple, Elizabeth B. Pearson, Natasha J. Olby, Annette L. Fitzpatrick, and Kate E. Creevy. 2023. A review of frailty instruments in human medicine and proposal of a frailty instrument for dogs. Frontiers in Veterinary Science 10 (2023).
Charles Ramey, Arianna Mastali, Cole Anderson, William Stull, Thad Starner, and Melody Jackson. 2022. WAG’D: Towards a Wearable Activity and Gait Detection Monitor for Sled Dogs. ACM International Conference Proceeding Series (12 2022).
Hao Yu Shih, Fillipe Georgiou, Robert A. Curtis, Mandy B.A. Paterson, and Clive J.C. Phillips. 2020. Behavioural Evaluation of a Leash Tension Meter Which Measures Pull Direction and Force during Human-Dog on-Leash Walks. Animals : an open access journal from MDPI 10 (8 2020), 1–10. Issue 8.
Jeremy Walston, Evan C. Hadley, Luigi Ferrucci, Jack M. Guralnik, Anne B. Newman, Stephanie A. Studenski, William B. Ershler, Tamara Harris, and Linda P. Fried. 2006. Research Agenda for Frailty in Older Adults: Toward a Better Understanding of Physiology and Etiology: Summary from the American Geriatrics Society/National Institute on Aging Research Conference on Frailty in Older Adults. Journal of the American Geriatrics Society 54 (6 2006), 991–1001. Issue 6.
A point-of-care system for continuous health monitoring should be wearable, easy to use, and affordable to promote patient independence and facilitate acceptance of new home healthcare technology. Reconfigurability, interoperability, and scalability are ...
PETRA '08: Proceedings of the 1st international conference on PErvasive Technologies Related to Assistive Environments
e-Health has clearly started to become an important issue for implementation, operational deployment of services and a promising market for industry. The need for concentration of the information society technologies on the future so-called convergence ...
The COVID-19 pandemic has considerably shifted the focus of scientific research, speeding up the process of digitizing medical monitoring. Wearable technology is already widely used in medical research, as it has the potential to monitor the user’...
Colt Nichols, 
Best Paper