“Reality is a sound, you have to tune in to it… You have to work at it. Pay attention to it. And learn to hear what it’s saying to you.”
Continuous sound monitoring systems have been shown to effectively detect clinical episodes of respiratory disease2-5. However, by their nature and by the nature of sound dissipation with increasing distance, microphones used in such systems would all be expected to have distance-related limits of sound detection. The purpose of this project was to evaluate the optimal placement and configuration of a continuous sound monitoring system in large airspace buildings in the United States containing growing pigs to enable both a high sensitivity for detection and establishing directionality of clinical respiratory episodes.
Materials and Methods:
Respiratory Distress Monitors (SOMO+ Respiratory Distress Monitor, SoundTalks NV, Leuven, Belgium) were obtained and installed in three large commercial wean-to-finish facilities designed to house 1200 to 2400 pigs per airspace. Three different farm sites / production systems were enrolled in the project. Pigs were placed into these site facilities per normal practice. Five devices were installed in each of the two 1200 head buildings, spaced equidistant from each other along the center alleyway (Figure 1). In the 2400 head building, 11 devices were installed, with four devices over the middle of the pens on each side of the building spaced equidistant from each other and three in the central alleyway spaced equidistant from each other (Figure 2).
Once installed, the SOMO+ devices (Figure 3) continuously monitored temperature using two sensors and humidity using one sensor. Also, each device had one connected microphone continuously recording sound. An algorithm was applied to the continuous stream of sound and classified specific sound events as coughs. The sound events classified as coughs were then counted, with the counts uploaded to a cloud database with a web user interface. A mobile app was used to monitor the SOMO+ devices remotely from a smart device (e.g., smart phone or tablet). An algorithm-based respiratory distress index (RDI) was continuously generated from the recorded sound files and cough counts, and was accessible via the web and app interfaces for monitoring and evaluation of various dynamic visualization tools, including summary tables and charts.
RDI’s were continuously monitored and alerts were automatically sent to pre-determined personnel when a significant rise in RDI was detected by the system’s algorithm. When an RDI alert was generated, diagnostic samples were collected via cotton rope-origin oral fluid sampling and tested by PCR for PRRS, IAV-S, Mycoplasma hyopneumoniae, PCV2 and parainfluenza. RDI episodes were aligned with their corresponding diagnostic results using event creation and tracking tools available via the web site interface. The resulting aggregate cough patterns were then characterized and categorized according to the diagnostic results. Other personnel observations representing “notable events” were recorded for each barn airspace and site using tools available via the web site interface.
A correlation analysis was conducted to estimate the optimal sound (cough) detection range for each microphone. The assumptions used for this analysis were:
- Each microphone detects coughs inside a circle of radius R
- Radius R is equal for all microphones
- The circle, defined by R, around each microphone represents a “hard” boundary, i.e., coughs inside of the circle are reliably detected and coughs outside of the circle are not reliably detected
- Pigs were (relatively) uniformly distributed inside the circle covered by each microphone
The correlation of detected coughs for each pairing of two different microphones was calculated based on the overlapping (intersecting) area of the circles with radius R around each microphone. The distance between each pair of microphones was estimated from the barn layout. The cross-correlation was measured for each pair of microphones with overlapping circles, and the computed correlations were plotted as a function of distance between microphones in each pair. Given that the calculated cross-correlations yield finite values (even though cough events detected by two different microphones may not all be correlated), a correlation baseline was determined. Following this, the measured correlation was fit to the correlation predicted by the hard-bounded model (Figure 4).
From the plots representing the fit of the measured correlations to the predicted model, estimated of microphone coverage zone (circles) were made for the 2400 and 1200 head barns. From these results, the optimal number and placement configuration of microphones (zones) were estimated for the 2400 and 1200 head barn types.
Where the device microphone was the center of a circle, the estimated optimal diameter for best detection of cough was determined to be approximately 18.3 meters in the 1200 head barn and 20.4 meters in the 2400 head barn (60 ft and 67 ft, respectively).
For optimal sound coverage in the 1200 head buildings the optimal number of devices was determined to be three to four per room or airspace, i.e., six to eight per two x 1200 head barn (Figure 5), and for the 2400 head building the optimal number of devices was determined to be six to eight (Figure 6).
Discussion and Conclusions:
Each device represents an 18.3 to 20.4 meter (60 to 67 ft) sound detection “zone”. Inherent differences in the acoustical characteristics of each of the two barn types are likely at least partly responsible for the range in optimal sound detection zones between the barns.
The sensitivity for the detection of and judging the directionality of cough events is then a function of the square meters covered by the “zones” out of the total possible square meters of animal space in a barn. This dynamic is highly analogous to the impact of sample size and sample selection where sampling pens of animals within airspaces (rooms) of barns and sites – i.e., the hard-bounded “zone” sampled when using a single rope to collect oral fluids constitutes one pen (sometimes two pens where a single rope is split and hung in both of two pens). Thus, fewer microphones (zones) would be expected to result in decreased cough detection sensitivity and reduce the ability to determine directionality of cough events.
Accurate and minute measurement seems to the non-scientific imagination, a less lofty and dignified work than looking for something new. But nearly all the grandest discoveries of science have been but the rewards of accurate measurement and patient long-continued labour in the minute sifting of numerical results.
- Carson A; The Autobiography of Red: Vintage; 1998; 149pp
- Genzow M, Duran CO et al.; Course of cough in two batches of fattening pigs with different respiratory pathogen exposure; Proceedings of the 23rd IPVS Congress; Cancun, Mexico; June 8-11, 2014; p212
- Genzow M, Duran CO et al.; Monitoring of a commercial fattening herd by means of the Pig Cough Monitor and oral fluid diagnostics; Proceedings of the 23rd IPVS Congress; Cancun, Mexico; June 8-11, 2014; p205
- Genzow M, Duran CO et al.; Novel approach for monitoring respiratory diseases in fattening pigs; Proceedings of the 2014 Allen D. Leman Swine Conference; September 15-16, 2014; p28
- Polson D, Playter S et al.; Precision Livestock Farming (PLF) for pig health and production: Sound as a diagnostic sample; Seminar #8 – The 4-Dimensional Revolution in Food Animal Health and Production: The synthesis of diagnostics, devices, digital platforms and data analytics; Proceedings of the 49th Annual Meeting of the American Association of Swine Veterinarians; March 4, 2018; pp21-24
- Kelvin, W; Presidential inaugural address to the General Meeting of the British Association, Edinburgh; Report of the Forty-First Meeting of the British Association for the Advancement of Science; 1872