Use of infrared thermography and machine learning to predict the infectious etiology of subclinical bovine mastitis in a robotic milking system

Mastitis is a major challenge for dairy production, and infrared thermography (IRT) is a promising non-invasive tool for monitoring udder health. This study evaluated whether thermal information obtained by IRT, combined with animal, production, and environmental variables, could be used in supervised machine learning models to predict subclinical mastitis (SCM) and its most probable infectious etiology in pasture-raised cows managed under a voluntary milking system. Eighty-nine lactating cows from a single herd underwent thermographic evaluation over six months, and milk samples from 530 mammary quarters were analyzed, of which 61 (11.5%) were positive for SCM. Mean (TMED), minimum (TCOLD), and maximum (THOT) udder surface temperatures, somatic cell count (SCC), milk composition, animal characteristics, and environmental data were used to develop a multilevel binary logistic model and a multinomial logistic model. The binary model, including SCC and TCOLD, showed an apparent area under the receiver operating characteristic (ROC) curve (AUC) of 0.96. However, using a ten-fold internal cross-validation, AUC decreased to 0.81 and sensitivity decreased from 0.90 to 0.33. The thermographic variables improved model fit modestly but significantly, whereas SCC accounted for most of the predictive information. In the multinomial model, apparent AUC values ranged from 0.78 to 0.99 across microorganism categories, but sensitivity was low for coagulase-negative staphylococci (0.03) and Staphylococcus aureus (0.30) and decreased further in leave-one-animal-out internal cross-validation. IRT-derived variables may be a complementary component of multivariable predictive models, but the validation of the present models in independent herds and under different management conditions is required previous its use for SCM diagnosis.

Authors

Institutions

Publication Details

Journal
Tropical Animal Health and Production
Published
2026-11-16
DOI
https://doi.org/10.1007/s11250-026-05475-y
Primary Topic
Milk Quality and Mastitis in Dairy Cows
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Use of infrared thermography and machine learning to predict the infectious etiology of subclinical bovine mastitis in a robotic milking system

Fernando David Caracuschanski, Alexandre Rossetto Garcia, Marcos Veiga dos Santos, Larissa Cristina Brassolatti et al.
4 links
Tropical Animal Health and Production
Milk Quality and Mastitis in Dairy Cows
article

Use of infrared thermography and machine learning to predict the infectious etiology of subclinical bovine mastitis in a robotic milking system

Fernando David Caracuschanski, Alexandre Rossetto Garcia, Marcos Veiga dos Santos, Larissa Cristina Brassolatti, Luiz Francisco Zafalon, Simone Cristina Méo Niciura, Christine Elisabeth Grudzinski, Raul Costa Mascarenhas Santana, Maria Laura da Silva, Teresa Cristina Alves
article en

Abstract

Mastitis is a major challenge for dairy production, and infrared thermography (IRT) is a promising non-invasive tool for monitoring udder health. This study evaluated whether thermal information obtained by IRT, combined with animal, production, and environmental variables, could be used in supervised machine learning models to predict subclinical mastitis (SCM) and its most probable infectious etiology in pasture-raised cows managed under a voluntary milking system. Eighty-nine lactating cows from a single herd underwent thermographic evaluation over six months, and milk samples from 530 mammary quarters were analyzed, of which 61 (11.5%) were positive for SCM. Mean (TMED), minimum (TCOLD), and maximum (THOT) udder surface temperatures, somatic cell count (SCC), milk composition, animal characteristics, and environmental data were used to develop a multilevel binary logistic model and a multinomial logistic model. The binary model, including SCC and TCOLD, showed an apparent area under the receiver operating characteristic (ROC) curve (AUC) of 0.96. However, using a ten-fold internal cross-validation, AUC decreased to 0.81 and sensitivity decreased from 0.90 to 0.33. The thermographic variables improved model fit modestly but significantly, whereas SCC accounted for most of the predictive information. In the multinomial model, apparent AUC values ranged from 0.78 to 0.99 across microorganism categories, but sensitivity was low for coagulase-negative staphylococci (0.03) and Staphylococcus aureus (0.30) and decreased further in leave-one-animal-out internal cross-validation. IRT-derived variables may be a complementary component of multivariable predictive models, but the validation of the present models in independent herds and under different management conditions is required previous its use for SCM diagnosis.

Tropical Animal Health and ProductionVol. 58(9)
Universidade de São Paulo (BR), Brazilian Agricultural Research Corporation (BR), Núcleo de Pesquisas Aplicadas (Brazil) (BR), Embrapa Pecuária Sudeste (BR), Universidade Estadual Paulista (Unesp) (BR)
Empresa Brasileira de Pesquisa Agropecuária
Zero hunger, Good health and well-being
Milk Quality and Mastitis in Dairy Cows
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon