Professional Documents
Culture Documents
2012.towards An Automated Detection of Oestrus in Dairy Cattle
2012.towards An Automated Detection of Oestrus in Dairy Cattle
x
ISSN 0936-6768
Review Article
Table 1. Efficiency, specificity and accuracy of oestrus detection by electronic heatmount detectors in dairy cows
Xu et al. 1998 N = 121 (lactating Holstein and Heatmount detector with radiotelemetric 92 ND 100%
Jersey cows calved seasonally) transmission (HeatWatch)
At-Taras and N = 46 (lactating dairy Heatmount detector with radiotelemetric 87 ND ND
Spahr 2001 cows housed in free stall) transmission (HeatWatch) 72 ND ND
N = 53 SO (lactating dairy
cows housed in free stall)
Cavalieri et al. 2003a N = 43 SO (lactating dairy cows) Heatmount detector with radiotelemetric 88 ND 100%
transmission (HeatWatch)
Cavalieri et al. 2003b N = 80 SO (non lactating Heatmount detector with radiotelemetric 94 ND ND
Holstein cows) transmission (HeatWatch)
a
Non-synchronized oestrus otherwise specified by SO for synchronized oestrus.
b
Efficiency (or sensitivity or detection rate) = number of cows in oestrus detected ⁄ number of cows in oestrus · 100.
c
Specificity = number of cows in dioestrus not detected ⁄ number of cows in dioestrus · 100.
d
Accuracy (or positive predictive value) = number of correct oestrus detections ⁄ number of correct + false-positive oestrus detections · 100.
ND, non-determined.
to limit the number of false-positive detections. As a mount accepted by the ewe, the RFID chip is read by
result, the sensitivity of the system may be less than the detector so that the female ID, the date and time of
expected, because up to 40% of mounts per oestrous mounting are registered and transmitted by radiotelem-
period may last <2 s. (Walker et al. 1996). Another limit etry to a computer (Bocquier et al. 2006; Maton et al.
of the HeatWatch system is that the patch containing the 2010). A similar system may be developed for cows but
pressure sensor is sometimes difficult to keep attached to it would require the use of a vasectomized bull or a bull
the cow, especially during moulting. Xu et al. (1998) with a deviated penis. Furthermore, the RFID chip,
reported failures to detect cows in oestrus as a result of currently implanted in one of the two plastic ear-tags of
loss or displacement of monitors. The daily time and cows, would have to be placed in the caudal area to
labour required to check the presence and position of assure the recording at the time of standing heat.
monitors on animals should thus be taken in consider-
ation (Saumande 2000). Last, the HeatWatch system
calls for an initial investment of 6600 USD (4862 euros) Monitoring of Activity
for 50 cows, comprising the base station, the software and Another behavioural change occurring in the cow in
monitors. oestrus is the increase in activity (Roelofs et al. 2010).
Automated systems may also be designed as a Among oestrous behaviours, this sign is difficult to
complement to fully mechanical heatmount detectors detect without technical aids and thus has elicited the
fixed on the rump (such as Kamar Heatmount detec- design of most automated devices currently used in
tors; Kamar products Inc., Zionsville, IN, USA) or tail dairy herds. Three types of devices are currently
paint. Also, an automated camera-software system available: (i) pedometers fixed on one of the legs, which
mounted in the milking parlour has been developed record the number of steps made by the cow per time
and patented by Massey University (New-Zealand unit (e.g. AfiTag, S.A.E. Afikim, Kibbutz Afikim,
Patent 519743, IPC7, G01N33 ⁄ 74), which detects the Israël; Crysta-Heat, Fullwood Ltd, Ellesmere Shrop-
partial or total removal of paint from a strip fixed on the shire, UK); (ii) activity-metres attached around the
sacro-coccygeal area of the cows (Alawneh et al. 2006; neck, which record neck movements in all three dimen-
Williamson et al. 2006). The detection of oestrus in sions (e.g.: Alpro, DeLaval International AB, Tumba,
seasonal-calving dairy cows at pasture using this auto- Sweden; Heat Box, Gènes Diffusion, Douai, France;
mated system resulted in greater sensitivity (85%), Heatime, Créavia, Rennes, France; HeatPhone,
specificity (99.6%) and positive predictive value (88%) Medria, Chateaugiron, France; MooMonitor,
compared with visual observation and tail paint (78%, DairyMaster Co., Kerry, Ireland); and (iii) activity-
98% and 51%, respectively) (Alawneh et al. 2006). metres attached to the leg that comprehensively assess
Automated heat detection could also take advantage the activity of dairy cows by measuring the number of
of the electronic identification of bovine animals by steps and quantifying lying and standing behaviours
Radio Frequency Identification (RFID) chip, available using the 3d-accelerometer technology (e.g. IceTag3D;
(but not compulsory) to all breeders in Europe since IceRobotics Ltd., Edinburgh, UK) (Jonsson et al. 2011).
2010 (European Commission 2009). The use of the Some of these devices can be attached to a leg or to the
RFID system was found to be successful for studies of neck as well (e.g. Nedap Lactivator; Nedap Agri,
sex behaviour and automated detection of oestrus in Groenlo, the Netherlands; Rescounter II, GEA Farm
sheep (Bocquier et al. 2006; Maton et al. 2010). The Technologies, Düsseldorf, Germany). The data on
design developed and patented (French Patent, Agro.M, activity are registered continuously and transmitted to
INRA-PHASE, Bocquier F., WO ⁄ 2005 ⁄ 065574) is a receiver by radiotelemetry or infrared light at regular
rather simple: the RFID chip is fixed at the base of time intervals (every 5 min for the HeatPhone or every
the tail of the female (ewe), whereas the male (ram) is 2 h for the Rescounter II) or at the entrance of the
equipped with an RFID detector fixed on its back and milking parlour (Heatime; Créavia, France). From this
an antenna fixed between its forelegs. At each standing receiver, the data are automatically forwarded to a
database in a central computer or to a cell phone via the et al. 2003a; Roelofs et al. 2005b; Sakaguchi et al. 2007;
GSM network. The software supplied with each device Hockey et al. 2010; Lovendahl and Chagunda 2010).
compares the activity of each animal with that of a Furthermore, the performance of pedometers for heat
previous reference period (of a various number of days detection may be affected by the environmental condi-
depending on the algorithm) and, in some cases, with tions, as walking activity of dairy cows has been shown
the average activity of the herd. For instance, the to vary according to the type of housing and the herd
DairyPlan system from GEA Farm Technologies management system (Kiddy 1977; Lopez-Gatius et al.
triggers an alert for a cow in oestrus when the activity 2005; Yaniz et al. 2006). On pasture-based systems,
recorded over three consecutive 2-h periods increases by because cows are expected to move different distances
at least two standard deviations over the average from one day to another because of pasture rotation, the
through the 10 preceding days (Hockey et al. 2010). specificity of pedometers may be altered. However, no
The relationships between increased activity, time of clear difference in sensitivity, specificity and accuracy of
ovulation and fertility have been investigated with the a neck activity-meter for oestrus detection could be
help of those devices (Lopez-Gatius et al. 2005; Roelofs observed between two dairy herds differing markedly in
et al. 2005b; Hockey et al. 2010). Ovulation takes place terms of animal environment and calving systems
on average 29–33 h after the onset of increased activity (Hockey et al. 2010). So far, very few studies have
and 17–19 h after the end of increased activity in addressed the effects of environmental conditions on the
lactating Holstein cows (Roelofs et al. 2005b; Hockey performance of cow activity-metres and more data are
et al. 2010). The analysis of more than 5800 heat periods needed. The use of pedometers or activity-metres calls
followed by AI in high-producing dairy cows demon- for an initial investment of approximately 5000–6000
strated that the likelihood of pregnancy at 38–45 days euros for 50 cows. An additional cost for subscription to
post-AI increased with enhanced physical activity, sug- the appropriate software is sometimes required.
gesting a positive correlation between increased activity
during oestrus and fertility (Lopez-Gatius et al. 2005).
The efficiency of oestrus detection using pedometers Video Recording and Automated Image
and activity-metres is generally in excess of 80% but Analysis
varies with the threshold set and the reference period of Secondary behaviours that are more intensively dis-
previous activity that are used to define the increase in played by cows in oestrus are restlessness, mounting,
activity, as well as with the time interval between two being mounted without standing, flehmen, sniffing
successive recordings (De Mol et al. 1997; Nebel et al. the vulva of another cow, resting with the chin on the
2000; At-Taras and Spahr 2001; Firk et al. 2002; back of another cow, licking, rubbing and aggressive-
Cavalieri et al. 2003a; Roelofs et al. 2005b; Sakaguchi ness (Kerbrat and Disenhaus 2004; Roelofs et al. 2005a,
et al. 2007; Galon 2010; Hockey et al. 2010; Lovendahl 2010). When all behavioural traits of oestrus are taken
and Chagunda 2010). The specificity of pedometers and into account, the rate of detection may reach 74–90%
activity-metres is generally between 90% and 100% but with three daily visual observation periods of 30 min or
their accuracy may vary considerably depending on the more (Firk et al. 2002; Roelofs et al. 2010). However, in
devices and on the algorithm used by the software most dairy farms, the workforce is limited and the
(Table 2) (De Mol et al. 1997; Firk et al. 2002; Cavalieri breeder cannot invest such amount of time into
Table 2. Efficiency, specificity and accuracy of oestrus detection by activity-metres attached to the leg or the neck in dairy cows
Activity-meter
References Number of oestrusa (type of animals and herds) attached to Efficiency (%)b Specificityc Accuracyd
De Mol et al. 1997 N = 537 (lactating dairy cows housed in free stall) Leg 81–91e 96–98%e ND
At-Taras and Spahr 2001 N = 46 (lactating dairy cows housed in free stall) Leg 83–87e ND ND
N = 53 SO (lactating dairy cows housed in free stall) Leg 79–91e ND ND
Cavalieri et al. 2003a N = 43 SO (lactating dairy cows) Leg 81 ND 88%
Roelofs et al. 2005b N = 63 (lactating Holstein cows housed in free stall) Leg 79–87e 97–100%e 65–94%e
Sakaguchi et al. 2007 N = 10 (Holstein heifers grazing on pasture) Neck 10–100f ND 20–100%f
N = 12 (Holstein heifers housed in open paddock) Leg 10–100f ND 24–100%f
N = 23 (Holstein heifers housed in tie stall) Neck 26–92f ND 37–100%f
Leg 75–100f ND 36–100%f
Neck 0–91f ND 0–50%f
Leg 13–96f ND 35–100%f
Lovendahl and N = 3674 (lactating Holstein, Jersey and Red Neck 56–84e 96–100%e ND
Chagunda 2010 Dane cows and heifers housed in free stall)
Hockey et al. 2010 N = 141 (lactating dairy, 80% Holstein, cows Neck 79–91e 90–98%e 40–73%e
housed in paddock and calved year-round) Neck 87–94e 90–98%e 41–76%e
N = 135 (lactating Holstein cows grazing
on pasture and calving seasonally)
a
Non-synchronized oestrus otherwise specified by SO for synchronized oestrus.
b
Efficiency (or sensitivity or detection rate) = number of cows in oestrus detected ⁄ number of cows in oestrus · 100.
c
Specificity = number of cows in dioestrus not detected ⁄ number of cows in dioestrus · 100.
d
Accuracy (or positive predictive value) = number of correct oestrus detections ⁄ number of correct + false-positive oestrus detections · 100.
e
Values vary depending on the threshold used in the algorithm to define an increase in activity as an oestrous period.
f
Values vary depending on the threshold and the reference period used in the algorithm to define an increase in activity as an oestrous period.
oestrus detection. Nevertheless, new technologies may vaginal temperature of 21 induced oestrus in 12 non-
simplify and shorten oestrus observation. All oestrous lactating Holstein cows and recorded a mean tempera-
behaviours, in addition to standing heat, can be video- ture increase of 0.48C (ranging from 0.40 to 3.22C) at
recorded for animals housed in free-stall or tie-stall the time of the LH peak. The same authors developed a
barns. A camera-software system recently designed in model based on vaginal temperature to estimate the time
France is based on the continuous recording of animals, of the LH surge. The LH peak could be predicted with a
including at night using infrared cameras, but only video 6-h approximation in 76% (16 ⁄ 21) of cows in oestrus
scenes when animals are moving or interacting are being monitored (Fisher et al. 2008). Because the interval
examined (Hétreau 2008; Hétreau et al. 2010). The between the LH peak and ovulation amounts to 29 h on
camera-software generates icons every 1, 10 or 60 min, average (Saumande and Humblot 2005), the prediction
depending on the detection threshold selected, and all of the LH peak with a 6-h approximation made it
sequences in which animals are inactive or little active possible to inseminate cows prior to ovulation. However,
are eliminated. Thus, only video sequences of interest these results require confirmation on a larger number of
are being viewed in either normal or accelerated mode. cows. Clearly, a variation in outside temperature (which
In an experimental herd of 80 lactating dairy cows was taken into account in the model described previ-
housed in free-stall barns and calved year-round, the ously), disease-related hyperthermia and ⁄ or some
camera-software system alone made it possible to detect systemic or local inflammation may alter heat detection
81% of cows in oestrus (N = 168) confirmed by based on body or milk temperature measurement (Firk
frequent milk progesterone assays (Hétreau et al. et al. 2002; Fisher et al. 2008) and thus increase the
2010). The reading of the image data required approx- number of false positives. Incorporating the last oestrus
imately 20 min for 24 h of recording in the first year of data in the algorithm may diminish the number of false-
the trial but could be reduced to 5 min per day in the positive signals. Altogether, the automated thermome-
third year when the detection threshold was lowered, ters currently used for calving detection still require
without significantly affecting the detection efficiency. In validation for use in oestrus detection.
this trial, heat detection by video recording provided an
efficiency equivalent to that of direct visual observation
(81% and 82%, respectively), but the time required was Automated Milk Progesterone Measurement
only 12.5–50% of that taken by classical visual detection In the non-pregnant cow, from day 18 after ovulation, a
(Hétreau et al. 2010). Nevertheless, as with visual obser- release of prostaglandin F2a by the endometrium leads
vation, video recording and analysis detect only heat to the lysis of the corpus luteum and consequently to a
periods with obvious behavioural changes. Moreover, drop in the blood concentration of progesterone (P4)
because of the low resolution of images recorded by the preceding a new ovulation (Thatcher et al. 2003). This
wide-angle cameras (initially developed for surveillance drop in P4 concentration may also be observed in milk,
of parking areas and airports), the reading of the ear-tag because the concentration in milk is closely correlated
number and thus, the identification of animals, may be with that in plasma (Roelofs et al. 2006). In milk, the P4
impossible. The only trial reported to date took place in a concentration declines to <5 ng ⁄ ml on average 80 h
herd with bi-colour breeds (Abondance, Montbéliarde before ovulation and <2 ng ⁄ ml on average 71 h prior to
and Holstein), which certainly made identification easier ovulation but with large inter-individual variations
that it would be in cows with a single colour coat. The new (standard deviation of 11 and 17 h, respectively,
types of surveillance cameras provide a higher resolution n = 20) (Roelofs et al. 2006). Non-automated P4 mea-
and make it possible to zoom within the image covering a surement in milk or blood for oestrus detection is not
rather wide area, but the cost is of course higher. The currently used in dairy cattle because of various limita-
installation of a video surveillance system in a farm calls tions, among which the serial collection, labelling and
for an investment approximately 4000 euros for two storage of samples and the relatively high cost of P4 assays.
cameras, the number of required cameras depending on A fully automated system of P4 assay in milk using
the area and design of the stall barns. A specific camera- dry-stick technology has been marketed in Denmark
software system for heat detection in cattle is expected to since 2009 and is becoming available in the rest of
be marketed in France in the near future and, thus, can be Europe since the end of 2010 (Herd Navigator; Lattec
evaluated in a larger number of herds. I ⁄ S, Hillerød, Denmark). The Herd Navigator (HN)
system can be combined with a DeLaval milking robot
or parlour and offers monitoring of reproduction,
Monitoring of Body Temperature mammary inflammation and metabolism by frequent
Several automated systems of body temperature mea- measurements in milk of P4, lactate dehydrogenase
surement with radiotelemetric transmission are currently (LDH), urea and b-hydroxybutyrate (BHB), respec-
marketed for calving detection and health monitoring in tively (Asmussen 2010). For purposes of oestrous cycle
cows. Some devices are ingested as a bolus and measure monitoring, the system collects several millilitres of
the temperature in the reticulum (ex: Thermo-bolus; milk, measures P4 and provides a curve of P4 concen-
Médria), whereas others are placed in the vagina (e.g. tration over the last few days. The system adjusts the
Vel’Phone; Médria; Radco; Verdor NV, Oelegem, frequency of assays, depending on the post-calving
Belgium). In cows, the body temperature declines period and on the stage of the cycle, to an average of 6–7
slightly approximately 2 days before oestrus and then progesterone assays per oestrous cycle. From the
increases at the time of the LH peak (Firk et al. 2002; smoothed P4 curve, the algorithm developed in the
Fisher et al. 2008). Fisher et al. (2008) measured the HN system classifies the cows into three categories:
post-partum anoestrus, oestrous cycling and potentially additional assays provided by the HN system (urea,
pregnant (Friggens and Chagunda 2005; Friggens et al. BHB, LDH) make it also possible to identify metabolic
2008). For oestrous cycling cows, an alert is triggered by disorders or mastitis that may be responsible for poor
the software as soon as the P4 value drops below reproductive performance.
4 ng ⁄ ml (Friggens and Chagunda 2005). In case of an
oestrus alert, the algorithm also provides the probability
of success of a prospective AI (between 0% and 100%) Conclusion
based on the duration of the previous luteal phase and In conclusion, several types of automated heat detection
the kinetics of the decrease in P4 concentration (Friggens devices for dairy cattle were marketed over the last few
et al. 2008; Asmussen 2010). When tested for purposes years, mostly heatmount detectors and activity-metres.
of validation in an experimental herd, the in-line milk P4 Other automated systems, such as camera-software and
measurements detected 99% of confirmed oestrus (for body temperature measurement systems, still require
which an AI resulted in a pregnancy; N = 121) and further testing in the field in dairy herds but may soon
93% of ratified oestrus (for which the P4 profile matched become available. In the future, biosensors integrated
that of confirmed oestrus, N = 679) with a specificity of with in-line milking systems and combinations of several
94% (Friggens et al. 2008). An average heat detection systems will probably develop as potent tools for heat
rate of 95% was later confirmed in commercial herds in detection, but also for global health monitoring in dairy
Denmark (Asmussen 2010). However, the HN system cows. Investment in these automation systems may
calls for a major initial investment of 40 000 euros for a become a source of profit provided the recorded data are
herd of 120 dairy cows and an additional average of 50 properly analysed and managed by breeders and veter-
euros per cow-year for assay reagents. The break-even inarians.
price for implementation of an in-line milk P4 measure-
ment in a dairy herd of 120 dairy cows previously using
Acknowledgements
visual oestrus detection has been estimated to range
Authors are grateful to Marc Chodkiewicz and Karine Reynaud for
from 3 to 81 euros per cow-year depending on the earlier reviewing the manuscript.
reproductive performance and on the implementation
type (Ostergaard et al. 2005). Beyond heat detection, the
in-line milk P4 assay may generate a substantial Conflict of interest
improvement in reproduction monitoring because it None of the authors have any conflict of interest to declare.
allows the identification of cows with post-partum
anoestrus, those with few or no behavioural changes at
oestrus, with luteal cysts and persistent corpora lutea or Author contributions
with early or late embryonic mortality after AI. The SCM and MSD have equally contributed to this review.
tion in dairy cows from activity and lying Roelofs J, Van Eerdenburg FJ, Soede NM, Thatcher WW, Guzeloglu A, Meikle A,
data using on-line individual models. Kemp B, 2005a: Various behavioral signs Kamimura S, Bilby T, Kowalski AA,
Comput Electron Agr 76, 6–13. of estrus and their relationship with time Badinga L, Pershing R, Bartolome J,
Kerbrat S, Disenhaus C, 2004: A proposi- of ovulation in dairy cattle. Theriogenol- Santos JE, 2003: Regulation of embryo
tion for an updated behavioural charac- ogy 63, 1366–1377. survival in cattle. Reprod Suppl 61, 253–
terisation of the oestrus period in dairy Roelofs JB, van Eerdenburg FJ, Soede NM, 266.
cows. Appl Anim Behav Sci 87, 223–238. Kemp B, 2005b: Pedometer readings for Thatcher WW, Bilby T, Bartolome J, Silves-
Kiddy CA, 1977: Variation in physical estrous detection and as predictor for time tre F, Staples CR, Santos JE, 2006:
activity as an indication of estrus in dairy of ovulation in dairy cattle. Theriogenol- Strategies for improving fertility in the
cows. J Dairy Sci 60, 235–243. ogy 64, 1690–1703. modern dairy cow. Theriogenology 65,
Lopez-Gatius F, Santolaria P, Mundet I, Roelofs JB, Van Eerdenburg FJ, Hazeleger 30–44.
Yaniz JL, 2005: Walking activity at estrus W, Soede NM, Kemp B, 2006: Relation- Van Eerdenburg FJ, Loeffler HS, van Vliet
and subsequent fertility in dairy cows. ship between progesterone concentrations JH, 1996: Detection of oestrus in dairy
Theriogenology 63, 1419–1429. in milk and blood and time of ovulation cows: a new approach to an old problem.
Lovendahl P, Chagunda MG, 2010: On the in dairy cattle. Anim Reprod Sci 91, 337– Vet Q 18, 52–54.
use of physical activity monitoring for 343. Walker WL, Nebel RL, McGilliard ML,
estrus detection in dairy cows. J Dairy Sci Roelofs J, Lopez-Gatius F, Hunter RH, van 1996: Time of ovulation relative to
93, 249–259. Eerdenburg FJ, Hanzen C, 2010: When is mounting activity in dairy cattle. J Dairy
Lucy MC, 2007: Fertility in high-producing a cow in estrus? Clinical and practical Sci 79, 1555–1561.
dairy cows: reasons for decline and correc- aspects Theriogenology 74, 327–344. Williamson NA, Alawneh J, Bailey D, Butler
tive strategies for sustainable improvement. Sakaguchi M, 2011: Practical aspects of the K, 2006: Electronic heat detection. Pro-
Soc Reprod Fertil Suppl 64, 237–254. fertility of dairy cattle. J Reprod Dev 57, ceeding of 2006 South Island Dairy Event.
Maton C, Bocquier F, Debus N, Bodin L, 17–33. Xu ZZ, McKnight DJ, Vishwanath R, Pitt
Teyssier J, 2010: Suivi automatisé des Sakaguchi M, Fujiki R, Yabuuchi K, Ta- CJ, Burton LJ, 1998: Estrus detection
chaleurs et différence de saisonnalité entre kahashi Y, Aoki M, 2007: Reliability of using radiotelemetry or visual observation
brebis Texel et Mérinos dans un envi- estrous detection in Holstein heifers using and tail painting for dairy cows on
ronnement méditerranéen. Rencontres a radiotelemetric pedometer located on pasture. J Dairy Sci 81, 2890–2896.
Recherches Ruminants 17, 125–128. the neck or legs under different rearing Yaniz JL, Santolaria P, Giribet A, Lopez-
Nebel RL, Dransfield MG, Jobst SM, Bame conditions. J Reprod Dev 53, 819–828. Gatius F, 2006: Factors affecting walking
JH, 2000: Automated electronic systems Saumande J, 2000: La détection électronique activity at estrus during postpartum per-
for the detection of oestrus and timing of des chevauchements pour la détection des iod and subsequent fertility in dairy cows.
AI in cattle. Anim Reprod Sci 60–61, 713– vaches en chaleur: possibilités et limites. Theriogenology 66, 1943–1950.
723. Rev Méd Vét 151, 1011–1020.
Ostergaard S, Friggens NC, Chagunda MG, Saumande J, 2002: Electronic detection of
2005: Technical and economic effects of oestrus in postpartum dairy cows: Submitted: 7 Jul 2011; Accepted: 22 Nov 2011
an inline progesterone indicator in a dairy efficiency and accuracy of the DEC
herd estimated by stochastic simulation. (showheat) system. Livest Prod Sci 77, Author’s address (for correspondence):
Theriogenology 64, 819–843. 265–271. M Saint-Dizier, AgroParisTech, UFR Géné-
Pecsok SR, McGilliard ML, Nebel RL, Saumande J, Humblot P, 2005: The vari-
tique Elevage Reproduction, 16 rue Claude
1994: Conception rates. 1. Derivation ability in the interval between estrus and
Bernard, 75231 Paris Cedex 05, France.
and estimates for effects of estrus detec- ovulation in cattle and its determinants. E-mail: stdizier@agroparistech.fr
tion on cow profitability. J Dairy Sci 77, Anim Reprod Sci 85, 171–182.
3008–3015.