Modern breeding farms rely on consistent flow of pigs through the production system. Reduced fertility can seriously disrupt animal production, flow and profit. These problems can originate from an array of reproductive failures and inefficiencies that can be additive. Preventing or quickly addressing these problems would require the capability to identify animals and their reproductive status. This would aid in management and flow of animals in the breeding herd scheduled for puberty induction, estrous detection, artificial insemination, pregnancy diagnosis, movement, farrowing, treatment intervention, or culling.


From a management standpoint, reproductive status for females in the breeding herd is based on daily detection of estrus, day of breeding, pregnancy diagnosis, and observation of farrowing. Yet reproductive failures or inefficiencies in each of these areas occur often and are additive to cost of production. This leads to inefficient utilization of space and resources, reduced animal production and congested flow of animals through the barns. Estrous detection is one of the most important and labor-intensive procedures used on the farm each day. It is a process where 1-2 humans must move a mature boar in front of the sow and allow them 1-15 minutes of contact (depending upon type of housing) before one person applies physical stimulation on the back of the female to elicit a standing response. However, estrus behavior is subject to considerable variation and sometimes leads to uncertain classification by humans. Pregnancy status after breeding can be confirmed as negative if the female returns to estrus 3 weeks later while confirmation of pregnancy is possible at 4 weeks with ultrasound. Day of farrowing is predicted based on the average gestation length of the pig (114-115 days after breeding), but is also unpredictable and can vary by 8 days in a single group of sows bred on the same day (d 112-119). The spread in gestation length has consequences for farms using all-in-all-out technology to limit disease, as it: 1) does not allow younger piglets to nurse for an optimal amount of days before the group is weaned; 2) uses the high cost farrowing crates for holding sows starting on day 110 before farrowing; and 3) limits precision scheduling for labor in the farrowing barn. The inability to diagnose whether a female is in estrus, cycling, pregnant, or when she will farrow, severely limits any precision in management, intervention, treatment, and removal.

We have developed a tool for accurately diagnosing current reproductive status as well as predicting the timing and likelihood of the next reproductive event. The diagnostic device uses a handheld intravaginal sensor with a smartphone interface. The electrical impedance spectroscope (EIS) device has been calibrated for use in sows and when inserted, can complete the scan within 15 seconds. The EIS device is inserted into the vagina to transduce electrical signals from the probe to the hardware circuit board to measure impedance between 10-400 ohms across a frequency range of 1,000-29,000 Hz. Diagnostic value is based on the impedance of the vaginal tissue in response to the endocrine activity of the ovary. The EIS data is collected and uploaded from a custom-built iOS app running on an iPod, to LabCore, a custom-built and industry-tested web-platform for data storage and analysis. The impedance data is labeled with production data and physiologic events for individual animals. Data analysis, prediction models, and machine learning models have been trained to generate the best models to predict and/or diagnose estrus, early pregnancy, and days to farrowing. Our EIS data indicates the electrical impedance (Ω) signature changes with frequency (Hz), stage of reproduction, and days to a reproductive event. For estrus, the data were unbalanced for frequency of anestrus and days to estrus. Nevertheless, estrus prediction had an r2= 0.98 at 24 hours before or on the day of estrus (RSTUDIO) and with a machine learning model (PYTHON) providing an overall accuracy of 90%. For early pregnancy diagnosis at day 18-20, the data for failures were also limited (6%) and presented an unbalanced data set for models. In this case, prediction was similar for days 18-20 and with an r2= 0.80 (RSTUDIO) while the machine learning model (PYTHON) included additional non-pregnant sows (inclusion of non-pregnant data from weaned sows) which increased the overall accuracy to 96%. For day of farrowing, variation in days was limited to days of measure (d 112-116) and days of farrowing (113-116), but still allowed prediction of farrowing within 0-24 h before the event with an r2= 0.98 (RSTUDIO) and at 0.89 at 48 h before farrowing. The machine learning model (PYTHON) provided an lower accuracy at 0-24 h (r2= 0.81) but a similar prediction accuracy 48 h before farrowing (0.89). Collectively, the data indicate EIS can accurately predict the occurrence of estrus one day before or on the day of estrus. With a balanced data set, EIS can predict day 30 pregnancy status as early as day 19. Lastly, EIS is highly predictive within 24 h before farrowing. More balanced and variable data sets are needed to move the technology towards generating the optimized machine learning prediction equations for installation into the system for immediate decision capability.

To contribute to improving the reproductive management of the breeding herd, in association the challenges that arise with the organization and location of greater numbers of sows and gilts in dif-ferent locations and stages of reproduction, we have developed the barn-mapping tool. Rapid iden-tification of animals, their location, and production status, would allow for improved organization and management. This would minimize hours of searching the farm for animals and labor used in re-locating females to other locations based on office records and stall cards. As a management-focused component of LabCore, we have developed a SmartBarn module as a Web-based plugin designed to integrate the animal with its location within the barn, its reproductive status, and employee tasks for animal or facility management. SmartBarn also features a spatial support to visualize, group, and map the location of animals and their reproductive status within a barn. The database structure is capable of capturing, querying, and reporting the geolocation of an animal (specific position within a crate or a pen, within a specific barn, or on a specific farm), with a display system. The program features animals identified by barn location, animal number, breeding group, reproductive status, and, importantly, to generate work orders for individuals or groups, or sets of stalls or pens within the facilities. The system also has a calendar feature to register reproductive events and tasks for any system classification with notifications generated. The SmartBarn system is novel in that it can generate barn mapping, integrate reproductive status, and generate work orders to aid in animal flow and breeding barn management.