Scientific Abstract
Rapid identification and treatment of morbid/injured pigs is essential for swine producers to ensure the health and wellbeing of each individual animal. However, in the modern commercial swine facilities, a single caretaker may have to access the health/wellness of ~4000 pigs/day (500 pigs/hour). This can be an insurmountable task to make individualized diagnoses using the daily “snapshots” obtained through periodic visual observation. Within each phase of production, pigs are exposed to a multitude of stressors that increase rates of illness and/or injury. One of the most challenging times for a piglet, as indicated by increased morbidity and mortality rates, is the pig’s transition to the nursery and the resulting stress of weaning. Stressors experienced during this early phase of production influence the competence of the developing immune system. Such alterations during the early development of the acquired immune system and can render pigs more vulnerable to “subclinical sickness” – morbidity that cannot be identified using traditional single-point human observations due to a lack of visible clinical symptoms. To overcome the limitations of human observations, we utilized an advanced computer vision platform (NUtrack Livestock Monitoring System) to identify changes in the physiological and behavioral changes associated with illness and aggressive/damaging behavior during the nursery and finisher phase. One hundred and ninety-two newly weaned pigs were utilized for this trial in a nursery study and a finisher study. For the nursery phases, pigs were randomly assigned to one of three treatments, a control group, and one of two endotoxin challenge groups. The two endotoxin challenge groups consisted of 100% of the pigs within a pen receiving an endotoxin challenge and the other group consisted of only 50% of the pigs receiving the endotoxin challenge. Following the nursery period, pigs were then assigned to two treatments groups within the finisher phase: mixed population pens or cohort pens. Pigs for the mixed population were randomly selected across the three nursery groups and assigned to a finisher pen. Pigs in the cohort pens were comprised of pigs from the original nursery pens. Results indicated that human observation presented acceptable rates for identification of compromised pigs on day 0 and 1 (0.85 AUC, >70% Sensitivity, >85% Specificity), however by day 2 human observations declined to undesirable rates (<0.57 AUC, <33% Specificity). While NUtrack had superior AUC, true positive, and true negative compared to human observations on d 0, 1, and 3 (>0.98 AUC, >79% Sensitivity, >94% Specificity). When compared to human observation, false positive identification was less frequent (P<0.05) for NUtrack identification. Overall, when compared to human observations, NUtrack had more days with adequate sensitivity and specificity and human observations created more false positives and false negatives. For early alert identification, results indicated that daily total values for each behavior provided greater accuracy (accuracy of 0.915 and 0.989) when compared to changes in daily behaviors (~0.70 and 0.63). As for model, logistic regression model did not serve as an acceptable model for identification of compromised pigs (produced a large number of control pigs classified as immune challenged pigs. B-Spline model provided an acceptable model, when using daily total values. The B-Spline model was not only capable of identifying LPS-challenged pigs (85 of 96 pigs), but also manager identified pigs during the post-challenged period (17 of 44 pigs). Overall results indicate that precision livestock technology has the ability to provide accurate identification of compromised pigs and there is potential for precision livestock technology to provide caretakers with an early-alert system for the identification of compromised pigs.
