monitoring the pregnancy status of dairy cows using mid-infrared ... · example of a mir spectrum...

Monitoring the pregnancy status of dairy cows using mid-infrared spectra of milk collected from milk recording

www.optimir.eu

WITH THE SUPPORT OF

A. Lainé*, L.M. Dale, H. Bel Mabrouk, C. Bastin, and N. Gengler University of Liège, Gembloux Agro-Bio Tech, 2 Passage des Déportés, 5030 Gembloux, Belgium

* [email protected]

IDF/ISO Analytical Week and Symposium OptiMIR Namur, 13-17 April 2015

A pregnancy diagnosis ?

www.partners-in-reproduction.com

Vet’s check Rectal palpation Transrectal ultrasond Echograpy …

AI/mating

Monitoring of the pregnancy along the lactation

2

Calving Calving ?


Using the monthly milk recording data ...

3

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

900 1400 1900 2400 2900 3400 3900

Ab

so

rban

ce

Wavelengths (cm-1)

Example of a MIR spectrum of milk

Mid-infrared spectrum of milk is a fingerprint of the milk composition

Spectra collected and stored routinely

Rapid, inexpensive and non-destructive method which is widely used

Allow for a large scale sampling


... not so easy !

4

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

900 1400 1900 2400 2900 3400 3900

Ab

so

rba

nc

e

Wavelengths (cm-1)


Chemical components of

the milk

Animal status

Pregnant cow or not ?

1. High correlations between some wavenumbers of the spectra

2. Milk composition is influenced by several different effects and the pregnancy status is probably not the main one

3. The pregnancy status is highly linked with the lactation stage


Dealing with drawbacks by modelling them

5

1. High correlations between some wavenumbers of the spectra

2. Milk composition is influenced by several different effects and the pregnancy status is probably not the main one

3. The pregnancy status is highly linked with the lactation stage

Transformation of wavenumbers by PCA

Estimation of various fixed and random effect

for each PCs

Introduction of days in milk in the model using polynomial Legendre


Dealing with drawbacks by modelling them

6

𝑦 = 𝑚𝑖𝑙𝑘𝑖𝑛𝑔 𝑚𝑜𝑚𝑒𝑛𝑡 + 𝑚𝑜𝑛𝑡ℎ 𝑜𝑓 𝑇𝐷 + 𝑝𝑟𝑒𝑔𝑛𝑎𝑛𝑐𝑦 𝑠𝑡𝑎𝑡𝑢𝑠 + 2𝑛𝑑 𝐿𝑒𝑔𝑒𝑛𝑑𝑟𝑒+ 3𝑡ℎ 𝐿𝑒𝑔𝑒𝑛𝑑𝑟𝑒 + 𝑎𝑛𝑖𝑚𝑎𝑙 + 2𝑛𝑑 𝐿𝑒𝑔𝑒𝑛𝑑𝑟𝑒 ∗ 𝑎𝑛𝑖𝑚𝑎𝑙 + 3𝑡ℎ 𝐿𝑒𝑔𝑒𝑛𝑑𝑟𝑒 ∗ 𝑎𝑛𝑖𝑚𝑎𝑙

1st lactations of Holstein cows over the last 3 years of records in the Walloon Region of Belgium 40,990 spectra

𝑝𝑟𝑒𝑔𝑛𝑎𝑛𝑐𝑦 𝑠𝑡𝑎𝑡𝑢𝑠

What is the impact of the pregnancy on the

spectra ?


Significant impact of the pregnancy status on MIR spectra

7

0

2

4

6

8

10

12

14

16

18

20

22

|pregnant|/s.e.

Principal Components explaining 99.9% of the spectral variability

95% of the spectral variability

t value > 2 64% PCs retained

t value > 5 22% PCs retained


Significant impact of the pregnancy status on MIR spectra

8

0

2

4

6

8

10

12

14

16

18

20

22

|pregnant|/s.e.

Principal Components explaining 99.9% of the spectral variability

Whole dataset

Pregnant data from 1 to 120days post AI

Same trends are observed when only pregnant data from 1 to 120 days post-insemination are considered to model the pregnant effect

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

900 1400 1900 2400 2900 3400 3900

Ab

so

rban

ce

Wavelengths (cm-1)



From theoretical concerns to a practical tool for farmers

9

Pregnant ?


Residual spectra as an image of the pregnancy status

10

Observed spectrum = Milk sample on which we want to test the pregnancy

Expected open spectrum = The expected spectrum for the same day in milk if the animal was open


Residual spectra as an image of the pregnancy status

11

Residual spectrum = Observed spectrum – Expected open spectrum

Residual spectra are used to perform discrimination between two groups of classification (Pregnant cow and non-pregnant cow)


Applying a logistic regression on residual spectra

12

2,515 Non-pregnant 5,377 Pregnant Testing dataset

46% 54% 89% 11%

Sensitivity Specificity

Average error rate of classification = 25%

46% of data belonging to

non-pregnant cows were correctly classified

89% of data belonging to

pregnant cows were correctly classified


13

Good perspectives on performances of the tool

Needs for improvement on the detection of non-pregnant cows

Tests on pilot farms are ongoing

Not yet implemented but the strategy is well define

Needs to provide an “useful” information to the farmers


Thanks to a fruitful partnership !

Service Public of Wallonia SPW – DGO3

European Commission (ERDF) through project INTERREG IVb OptiMIR

Author’s contact e-mail:

[email protected]

monitoring the pregnancy status of dairy cows using mid-infrared ... · example of a mir spectrum...

Documents