Progress 09/01/14 to 08/31/17
Outputs
Target Audience:This project addresses USDA's research in Animal Production and Protection with the new FitGait. The FitGait system noninvasively and objectively assesses the incidence and prevalence of lameness within lactating cattle every day. Cows are identified as lame earlier and more accurately with FitGait than they are using the industry standard methods or techniques. By removing human observers from the process, lameness detection is based on objective, and repeatable parameters. More specifically, this grant is connected to increasing the efficiency of dairy production and improving the health and well-being of dairy herds by enabling the early detection, and therefore treatment, of bovine diseases such as white line disease, sole ulcer, abscess, and other ailments that cause lameness and lead to economic losses. The concept behind FitGait was that automated lameness detection should increase the efficiency and sustainability of the dairy industry by improving the health and well-being of dairy herds. Revenue losses are attributed to diminished milk yields, lowered reproductive efficiency, increased mortality, and involuntary culling. Continuous, accurate detection by FitGait coupled with proper treatment is vital to reducing economic losses due to lameness. The resulting economic benefits to the dairy producer would improve the competitiveness of United States dairy products and bolster the financial security of producers. Arguably, an issue of even greater importance is the concept that proactive intervention by the dairy industry into the welfare issue surrounding pain and debilitation of lameness gain favor with U.S. consumers and expand marketability of milk and dairy bovine products. Even though our current efforts have been restricted to the commercial dairy farms on which we are performing our beta testing, it is our mission to have a global impact on the industry by providing early, objective and accurate lameness detection. Automated lameness detection in dairy cattle will increase the efficiency and stability of the dairy industry by improving the health and welfare of dairy herds. Additionally, each visit by our team to the farms has provided comprehensive medical care and treatment for all the cows examined by our expert veterinarian. Dr. Dyer's services are provided at no cost to the producer and has been imperative to the overall well-being of the herd on Scattered Acres and Green Acres, particularly to the many lame cows found during testing. The FitGait automated lameness detection will provide dairy producers an opportunity to avoid or minimize the costs of up to $400 per incidence of lameness (Greenough and Weaver, 1997), which is one of the three most costly animal health issues on every dairy farm. The results of our grant research have enabled Step Analysis to incorporate larger, more robust functionalities into the FitGait system and provide producers with the vital information in order to make informed decisions, provide proper treatment at the right time, and monitor the health of the herd and individual cows on a daily basis without breaking farm routines. Coupled with non-subjective lameness analysis and a statistical model to detect lameness, our system delivers high accuracy in a commercial dairy farm environment - a pioneering breakthrough in lameness management. While successful development of the FitGait system under this grant was a significant step to improving herd health and management for dairy producers, there are still a number of unaddressed problems that producers face with no effective tools or methods. For example, the enhanced FitGait will be able to provide more detailed information associated with the improving or worsening lameness severity, and to monitor the overall health of a cow by displaying key body weight data. We also continue to research new methods and designs to reduce the cost of manufacturing. Accordingly, there is still a great deal of research and development required before a FitGait product is ready for the commercial market. To support these continued tasks, Step Analysis has recently applied for USDA SBIR Phase I 2018 Proposal Number: 2018-00430, entitled "Developing High Accuracy, Low Cost Lameness Detection andWeight Assessment System for Dairy Bovine." We are fortunate to have the commitment from Lely, a leading manufacturer and innovator of dairy farm equipment in the U.S. and globally, to partner in the research and development of FitGait and contribute to the proposed grant project. While a long term partnership is still in discussion, we have been able to establish many strong relationships among the market leaders and hope to continue our research and development efforts with the support of the USDA and Lely. Lely intends to market FitGait as a component of its total dairy management system. With the support of Lely, FitGait will help farmers around the world to better manage their hoof and body weight problems, improving their bottom line of farm operations. Not only promising to complete the objectives of Phase I, Lely has committed to supporting the project through Phase II and Phase III upon the successful completion of each. In particular, Lely will construct 3-5 systems that will be installed on 3-5 farms sufficient to collect sufficient data to demonstrate the detection of lameness by the system with the appropriate sensitivity and specificity and statistical significance. This larger dataset will enable Lely and Step Analysis to build and test a robust, statistically significant model for detecting lameness type and severity as well as body weight variations in dairy cows. Lely and Step Analysis hope that positive validation of the accuracy of the prototype systems in Phase II will lead to successful commercial launch in Phase III. Further, Lely's target market for FitGait is dairy farms with a minimum of 150 cows that are milked at least twice daily in a parlor-style milking facility located in United States, Argentina, Australia, Canada, Chile, Czech Republic, Estonia, El Salvador, France, Germany, Holland, Hungary, Israel, Italy, Japan, Mexico, Peru, Poland, Portugal, Puerto Rico, United Kingdom, Ukraine, Saudi Arabia, Slovakia, Spain, South Korea, Switzerland, Taiwan, and Thailand. An owner or decision-maker that has responsibility for the deterrence, detection and treatment for lameness and herd health management on dairy farms with 150 or more cows is the purchase decision-maker. The new FitGait system developed by Step Analysis, with its high level of accuracy, will further expand international markets and open up the cost-conscious U.S. market at a time when animal welfare becomes a growing concern. The information can be displayed to show the temporal progression of lameness by animal, or establish the prevalence of lameness by time, production group, or herd. The need for early and automatic diagnosis of lameness in dairy herds is obvious. We estimate that the current improvement in the accuracy of the FitGait device having a sensitivity and specificity of at least 85% will be sufficient to convince farmers in the U.S. of the economic advantages of FitGait and make the product more appealing. While we have been able to increase the sensitivity and specificity of FitGait to this level, additional features are required before producing a final commercial product. An improved, more accurate FitGait will penetrate the U.S. market and improve the efficiency of the U.S. dairy industry. Changes/Problems:In Phase II, BouMatic decided to no longer pursue the project and Step Analysis redesigned and constructed 3D force sensor modules, known as FitGait, on two commercial dairy farms to collect LMV data of cows. The loss of BouMatic is seen as a gain for Step Analysis. Instead of hiring outside support to develop hardware, electronics, and software, we developed all of these components in-house. Developing the system has allowed us to expand our skillset. We anticipate these new skills to allow us to more easily move into horizontal markets in the future as we now have control over all facets of the system. For example, we have been able to perform specialized equipment design and manufacturing for doctors' stroke experiments and research at Johns Hopkins and Duke University. We also hope these new skills allow us to address the lameness epidemics in other fields gravely affected by the disease, including but not limited to equine, ovine and swine. Further, our work during Phase II enabled Step Analysis to build and test a robust, statistically significant model for detecting lameness in dairy cows using 3D LMVs. We validated the accuracy (sensitivity and specificity values) of the enhanced digital lameness detection system, allowing for the development of an improved FitGait product. Specifically, our work has validated the hypothesis that force measurements in three orthogonal dimensions are sufficient to improve sensitivity and specificity of models that predict lameness to over 85%. These improvements were likely caused by the 3D system's ability to detect load shifts within limbs in the transverse and longitudinal directions that the previous devices did not measure. The FitGait system builds on the proven technical components of the previous systems designed by Step Analysis. After successful operation of our experiment for over a year, we began to notice some small discrepancies in our data output. The source of these discrepancies was not recognizable without extensive testing and checking of all of the software and hardware components of the system. We discovered that the issue was a mechanical failure with the sensor assembly, specifically the limiting members over time were gauging into the sensors causing the system to become misaligned, and, thus, inaccurate data collection. We believe this failure was likely caused by overloading from cows jumping on the platform. Unfortunately, this setback required us to reengineer the entire assembly, but also, fortunately, has provided us with the opportunity to invent a durable design, which will greatly benefit farmers in the long term. Ultimately, because of the machine malfunctions described herein, the prototype on Green Acres could not be rebuilt and we focused our final lameness testing and modeling efforts this past year on Scattered Acres. From a commercialization perspective, we have filed patent applications covering our new inventions, which will add value to our current portfolio. In particular, we filed a patent to cover the improved sensor assembly configurations for the lateral and longitudinal directions as well as other new advantages. The new assemblies now allows for the sensing platform to be suspended in all three directions improving, among other benefits, the durability and measuring accuracy of the system. Further, we quickly learned of the load cell configuration's unreliability when employed on large scaled dairy operations. As found in Phase I testing, Phase II experienced numerous hardware failures occurring with the load cells and the mounts aligned to measure longitudinal (acceleratory-braking) forces. Despite upgrading to load cells rating for 2,500 lb and adding lock nuts to the bolts fastening the mounts to the plate and frame, these failures became more frequent and apparent when the prototypes were tested on larger scale dairy operations over an extended period of time in Phase II. These failures inspired us to innovate and discover a new technique for lameness detection without the use of load cells. Patent protection for these new features is also currently pending. Accordingly, after a system failure caused by the load cells, Step Analysis took the opportunity to reconstruct the prototype as a dual-system consisting of both load cells and accelerometers. Step Analysis has already completed an important initial task of proving that accelerations generated by a traversing cow on the system can distinguish between sound and lame cows with high accuracy. We also validated our hypothesis that the linear correlation between force and acceleration should facilitate progression of past successful work with load cells in order to incorporate accelerometers into our system. As detailed herein, the preliminary results appear promising, particularly for developing a new FitGait system that solely incorporates accelerometers, which offers not only much greatly reliability but also are tenfold cheaper. In particular, the cost of a standard load cell is $310 compared to about $100 per accelerometer, and the current design requires 14 load cells and merely 4 accelerometers. This is a potential savings in manufacturing costs of almost $4,000 per device. Accordingly, it is a primary objective of our future research and development to reduce the cost of the machine and improve its overall reliability by incorporating accelerometers into our system. We plan to build a new FitGait system that minimizes the number of load cells and relies more on accelerometers as we continue its development, which offers not only much greatly reliability but also are tenfold cheaper. While the ability achieved in our research and development to distinguish between lame and sound cows was significant progress for accurate automated lameness detection, there are still a number of improvements necessary to provide dairy producers with robust information to make informed and timely decisions for the proper care of their herd. Preliminary evaluation of the newly developed FitGait system demonstrates tremendous improvements above the projected goals in lameness detection capabilities for Phases I and II. Despite the need for additional improvements, the confirmed sustainability of the high sensitivity and specificity of the current system will place the accuracy of automated lameness detection substantially higher than the 45-50% level of accuracy achieved by most producers. However, it is imperative for producers not only to identify lameness and the infected limb as done by previous FitGait modules, but also determine the type and severity of the disease and track its progression or recovery over time. Further, producers currently have no effective tool to assess the body weight variations of individual cows on a daily basis with high accuracy and minimal expense. The future of the FitGait system will integrate with a farm's existing systems to provide the producer with a new metrics of real-time, automated, objective lameness detection and body weight variations. By combining the capabilities of lameness detection and body weight measurements into a single product, the next generation of FitGait will revolutionize the dairy industry's ability to make educated and effective decisions on the treatment and care of herds without breaking daily routines and at minimal investment. What opportunities for training and professional development has the project provided?The team has been fortunate to have the assistance of many students during the course of the project. In particular, Dr. Neerchal's team included Qing Ji who is studying for his Ph.D. in statistics at UMBC. Dr. Dyer's team included undergraduate students--Bridget Aylward and Shannon Heintz--from University of Delaware, majoring animal sciences. Dr. Gröhn's team at Cornell University consisted of two post-doctorate students--Julia A. Hertl and Karun Kaniyamattam--who focus their research on veterinary epidemiology, economic modelingin dairy cattle. Further, during Phase II, Jason Dunthorn from Step Analysis's team completed his masters in mechanical engineering at UMBC based on his work with this project. And, Dan Tasch from Step Analysis started and finished a master's degree in business administration from Georgetown during Phase II. All of the students have received a great deal of training in their respective fields with their work on this project. Also, they all have been able to utilize the work from this project into academic opportunities in their own coursework. In addition, this project and its data were able to serve as a real world client for purposes of a student research project for the Research Experiences for Undergrads Site: Interdisciplinary Program in High Performance Computing at University of Maryland, Baltimore County. Under the guidance of Dr. Neerchal, the five students from various institutions formed a team to prepare a preliminary analysis of our Scattered Acres data during the summer of 2017. Our participation in this program allowed us to engage young academics excited about statistics and potential innovative applications of their interests. How have the results been disseminated to communities of interest?An article was published in the Journal of Dairy Research (2015) entitled "Predictive Models of Lameness in Dairy Cows Achieve High Sensitivity and Specificity with Force Measurements in Three Dimensions," which details the results of Phase I of our USDA project. In addition, "Predictive Models of Lameness in Dairy Cows Achieve High Sensitivity and Specificity with Force Measurements in Three Dimensions" was accepted for oral presentation at the 2015 Conference Lameness in Ruminant in Valdivia, Chile. While we were unable to take advantage of this opportunity to share the progress of our USDA project at that time with the international community of academics, practitioners and hoof trimmers, we plan to apply for future publication and presentation of the results from this grant project. Step Analysis has also made significant efforts to achieve its commercialization goals, by expanding its network in the dairy market and attending the World Dairy Expo in Madison, Wisconsin in the past two years. During our visits, we had the opportunity to meet with all facets of the industry and share with them our most current results. We plan to share these successful and exciting final results from Phase II with all of them. Hopefully with the help of these potential strategic partners and customers, we will be able to execute our commercialization goals and provide U.S. dairy producers and farmers around the world with the vital hoof health information they need to treat and care for their herds. What do you plan to do during the next reporting period to accomplish the goals?
Nothing Reported
Impacts
What was accomplished under these goals?
All of the primary technical objectives to meet our goal of accurate, automated detection of lameness have been successfully completed in Phase II. Objective 1-Construction and testing of the hardware and software was completed and installed by Step Analysis on two different commercial dairy farms. Software to interface with the force-sensing module was completed by Step Analysis. Step Analysis contracted installation of the two devices with the assistance a licensed BouMatic dealer. The two commercial dairy farms were Scattered Acres in Reinholds, Pennsylvania and Green Acres, in Lewes, Delaware. The prototypes were built with load cells oriented to report movement in vertical, lateral, and longitudinal directions. GRFs retrieved from FitGait converted into LMVs for each direction to create a cow's LMV by time series step signature. LMV generation was paired with the clinical scores generated during the farm visits. Software to automatically match log files to clinical data has been developed and implemented. Also, we were able to also develop software to automatically distinguish between and recognize the front and hind limbs. Objective 2-We examined over 1200 cows (1,262 total) across all levels of lameness (1-5) during the entire course of Phase II. During our farm visits, randomly selected cows were examined for lameness by an experienced expert. Cows were assigned an overall lameness score based upon these measures. Cows with overall lameness scores 1 and 2 were considered sound while cows with lameness scores 3-5 were considered lame. GRFs recorded by FitGait on the day of the inspection and for the two days previous to the inspection were retrieved for use in the modeling procedures. Over the entirety of the project, we have been able to collect and perform clinical examinations, including lesion type and severity, for each limb of over 1,200 cows. Specifically, 1,173 unique cows from both Scattered Acres (1,053 cows) and Green Acres (120 cows) and 89 additional cows' clinical reports were removed for reasons such as front limb lameness, blocked hoofs or repeated cows, combining for a total of 1,262 examined cows. In total, 21.06% of the 1,173 examined cows were found to be lame. With Scattered Acres and Green Acres exhibiting overall lameness of 19.09% and 38.33%, respectively, for the herd over the testing period. While we achieved our goal of examining 1,262 cows, the databases for our final best lameness modeling only included data from 2017, because the 2015-16 data collected prior to the prototype redesign was not combinable with the new machine data. Objective 3-Although a training database of 800 cows was populated by Step Analysis, only the 549 cows from our visits at Scattered Acres in May to July 2017 could be used to develop a model accurately predicting the probability of lameness from inputs of clinical data and ground reactions forces. Specifically, the final database contained 760 observations (runs) comprised of 655 runs among the 253 sound cows and 105 runs among the 44 lame cows. We confirmed our previous success by developing a model that performed with transformation of 1 degree and 4 knots resulting in 91% sensitivity, 91% specificity, and 95% AUC. It must be noted that for future commercial implementation of the lameness detection model, further conservative statistical evaluation of the model's prediction ability may be needed. To test the consistency of the diagnostic model, a model was built based on the same data from May 2017 to July 2017 (297 cows and 760 records). From this data, 13 lame cows selected with 120 before and after trimming records, and 34 sound cows with 521 before and after trimming records. Our analysis shows that a majority of the probabilities predicting lameness after trimming are consistently less than the predictions before for sound cows, while the majority of the probabilities predicting lameness after trimming for lame cows are consistently higher. As with the training database, the data available to be employed in our validation goals only included our visits at Scattered Acres. We used the data collected from May to July of 2017 (760 records from 297 cows) as training data and the data from collect in August 2017 (175 records from 91 cows) as a testing set. Validation required comparing the model's prediction to the veterinarian's clinical diagnosis. The prediction is on the individual cow level by taking the maximum probability of being lame as the prediction. The AUC from the testing set is 94%. At threshold 0.25, the specificity and sensitivity reached 85% and 90% respectively. Further, at threshold 0.2, the specificity and sensitivity achieved 81% and 100%, respectively. And, while we attempted to utilize the data collected in 2015 and 2016 for purposes of validation, these efforts only affirmed that the data required appropriate parsing for proper analysis. Objective 4--With the assistance of Dr. Grohn, a trial run using dynamic programming was utilized to estimate the economic losses due to lameness in dairy cattle. We continue to develop an economic losses model that can provide producers with the real-time costs associated with lameness in an individual cow and the whole herd that is incorporated into the FitGait system. The entire cow database of 1,173 unique cows will be utilized in the economic losses model development and with the collection of future farm data from Scattered Acres and Green Acres further analysis will be performed. The optimal profit per cow (average across all cow characteristics) per year under assumed expenses and revenues was $426.05 when the incidence of lameness was assumed to be 0 (ideal situation) in the herd. The profit per cow per year reduced to $395.22, when total incidence of 19.70% for lameness was used in the model, out of which 12.2% was severe cases and 7.5% was moderate cases. The model recommended that 95.9% of total cases of lameness is to be treated under the scenario which assumed both severe and moderate lameness in the herd. The average cost per case of lameness was $156.50 in an average herd with 19.7% incidence of lameness. Out of the total cost of $180.82 in the case when just severe cases of lameness are assumed, $90.08 (49.8%) was due to milk loss, 30.17 (16.7%) was due to fertility loss and 60.57 (33.5%) was due to treatment associated costs. The retention payoff (optimal economic return by retaining a cow with specific characteristics) was obtained from the dynamic program. The RPO's in the case of a 2nd lactation cow, which is open and has average milk production. RPO's for the scenarios a) No lameness (healthy cow), b) the cow has moderate lameness, and c) the cow has severe lameness are shown. The RPO's of healthy cow ranged from $1,274 (first 3 days of calving) to -90.21 (17 months after calving). The model recommended to cull the healthy cow after being open for 11 months. In the case of severe lameness, the model recommended to treat and inseminate the cow till 9 months after calving, following which the optimal decision was to cull the cow. If the lameness was moderate, the cow got to stay in the herd till 10 months after calving, following which the recommendation was to replace the cow (RPO's are negative). The costs of severe and moderate lameness at the individual cow level, specific to cow characteristics were calculated based on the parameters available through literature review. Once the parameter estimates from the 1,173 cows are available we will repeat this process to estimate the lameness costs specific to Scattered Acres and Green Acres farms. In addition to the estimations done in this report, we can do different sensitivity analysis after accounting for the savings in the a) milk loss, b) fertility decline and c) culling risk, provided FitGait can detect lameness earlier than the traditional once a month hoof trimmer VLS analysis.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2015
Citation:
Dunthorn, J., Dyer, R., Neerchal, N., McHenry, J.S., Rajkondawar, P.G., Steingraber, G., Tasch, U. Predictive models of lameness in dairy cows achieve high sensitivity and specificity with force measurements in three dimensions. Journal of Dairy Research (2015) 82 391-399
|