Automated Mastitis Detection
for Dairy Farms
Amanda Sterrett & Jeffrey Bewley
University of Kentucky
Dairy Systems Management
But how we monitor it is different…
Take advantage of simplicity
 Because cows are routine-oriented, we can monitor
their behavior and examine differences in:
 Eating time / DMI
 Standing / Lying time
 Rumination time
 Location within barn
Physiological monitoring
 Body temperature
 Ear, milk, reticulorumen, udder, vagina
 Milk composition
 SCC
 Fat, lactose, protein, LDH, etc.
 Electrical conductivity
Potential Benefits
Early
Mastitis
Detection
Early
Treatment
Improved
Treatment
Outcome
Less
Production
Loss
Improved
Prevention
Program
Less
Economic
Loss
Improved
Animal WellBeing
Accuracy and Precision
Sensitivity and Specificity
Sensitivity (true positive rate): alert with an observed
mastitis case
𝑡𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 =
𝑡𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝑓𝑎𝑙𝑠𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑠
Specificity (true negative rate): no alert with no mastitis
𝑡𝑟𝑢𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑠
𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑖𝑡𝑦 =
𝑡𝑟𝑢𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑠 + 𝑓𝑎𝑙𝑠𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
Electrical Conductivity
 Ion concentration of milk changes during mastitis
 Inexpensive and simple equipment
 Wide range of sensitivity and specificity reported
 Affected by sample time, milk viscosity, temperature, and
sensor calibration
 Most useful when combined with other data
Automated CMT or WMT
• CellSense (New Zealand)
• r = 0.76 with Fossomatic SCC
Alert based on
EC
Alert based on InLine SCC
Alert based on
EC and SCC
Alert
time
period
Observation
period
Sensitivity
False
alert
rate
Sensitivity
False
alert
rate
Sensitivity
False
alert
rate
96
48
80
4.7
83.3
2.9
80
1.2
48
24
80
7.8
83.3
3.7
80
2.1
Somatic Cell Count
• In-line detection of cell count, milk
temperature, and electrical
conductivity
• Uses ATP luminescence as an
indicator of the number of somatic
cells
• Sensor connected to the milk hose
below the milking claw
• Reagent cassette attached below
display
Spectroscopy
• Visible, near-infrared, mid-infrared, or radio frequency
• Indirect identification through changes in milk composition
• AfiLab uses near infrared
– Fat, protein, lactose, SCC, and MUN
LDH Threshold (µmol min-1|-1) Sensitivity Specificity
4.3
95.2
92.0
6.5
72.6
98.5
http://blog.
modernmec
hanix.com/r
obot-cowmoos-andgives-milk/
Temperature Limitations
• Not all cases of mastitis result in a
temperature response
• Best location to collect temperature?
• Noise from other physiological impacts
Udder Thermography
• Udder temperature closely related to rectal temperature
• No early detection in LPS challenge (Hovinen et al., 2008)
• Potential use in dry cows
Before Infection
After Infection
Hovinen et al., 2008
Accelerometers
• Measures lying time and activity/motion
index
• Well researched and applied to many areas
• Lying is a high priority behavior
• May change lying time around mastitis
• May decrease activity around mastitis
• Lying time decreased by 73 minutes on the
day of challenge (P < 0.01, Cyples et al., 2012)
Rumination Behavior
• Cows with mastitis may ruminate less
• r = 0.93 for automated rumination with live
observations in cows (Schirmann et al., 2009)
Animal Position
• Real Time Location System
• Cows may stay in same spot
longer around mastitis
Multi-parameter Sensors
• Combination monitors may find a better market than
those sensors only targeted at one parameter:
– Temperature
– Activity
– Rumination
– Feeding Time
• Multivariate analyses
Considerations
 Economics
 Positive return on investment
 Producer satisfaction
 What data is useful?
 Reading frequency
 What do we do with the data?
 Culture, monitor, treat, ignore?
Conclusions
 Using technologies for mastitis
monitoring is newer than using
them for estrus detection
 Algorithms are not yet
perfected
 Continued research is needed,
particularly in naturally
occurring mastitis
Questions?
Amanda Sterrett
408 WP Garrigus Building
Lexington, KY 40546
412-558-2075
amanda.sterrett@uky.edu
Dr. Jeffrey Bewley
407 WP Garrigus Building
Lexington, KY 40546
859-699-2998
jbewley@uky.edu