Machine Learning Approach for Early Lactation Mastitis Diagnosis Using Total and Differential Somatic Cell Counts

Area of Science:

• Veterinary Medicine
• Animal Science
• Data Science

Background:

• Dairy production faces increasing herd sizes and milk yields, necessitating data-driven efficiency improvements.
• New technologies generate vast data, requiring advanced methods for information extraction.
• Intramammary infection (IMI) in dairy cows impacts production and requires accurate diagnostics.

Purpose of the Study:

• To evaluate the accuracy of machine learning (ML) models using somatic cell count (SCC) and neutrophils + lymphocytes count/mL (PLCC) for identifying cows with intramammary infection (IMI) caused by major pathogens (MajPs).
• To compare the diagnostic performance of real-time PCR (qPCR) and conventional bacteriology for detecting MajPs in dairy cows post-calving.

Main Methods:

• A preliminary study involving 424 cows and 1696 quarter milk samples.
• Utilized machine learning (ML) algorithms to analyze SCC and PLCC data.
• Identified major pathogens (S. aureus, S. agalactiae, S. uberis, S. dysgalactiae) using qPCR and conventional bacteriology.

Main Results:

• qPCR detected a higher prevalence of MajPs compared to conventional bacteriology.
• Cellular markers (SCC, PLCC) showed high accuracy and specificity in diagnosing MajP IMI when using quarter milk samples.
• ML algorithms demonstrated similar performance regardless of the specific algorithm used.
• Both methods showed comparable negative result prevalence (~71-72%).

Conclusions:

• SCC and PLCC, analyzed with ML, show promise for identifying healthy cows or quarters, particularly due to high specificity.
• Confirmation of "non-negative" results via follow-up testing within 7-15 days is crucial for accurate diagnosis.
• Further research is needed to improve the overall diagnostic accuracy for IMI in dairy herds.