Seeing red: automated detection of blood in milk

Milk that has changed in colour because of the presence of red blood cells is regarded as abnormal milk. The incidence of visible blood in raw milk is rare but, nevertheless, it should be monitored routinely in any AMS to monitor animal health and to maintain consumer confidence in milk as a pure food. The purpose of this paper is to describe the development of a robust reference method, or "gold standard", and of a non-invasive in-line measurement method for detecting the presence of blood in milk from individual cows or udder quarters.
Our search for a reliable gold standard included colour scanning, atomic flame spectroscopy, particle counting, direct microscopic counts of Red Blood Cells (RBC), commercially available dipsticks, and haemoglobin (Hb) concentration. A standard specification based on a fixed percentage of blood is unreliable because both the RBC and Hb content of blood can vary considerably. A concentration of 0.1% raw bovine blood in milk may contain 5-10 million RBC/mL. As an indication of the appropriate range for a reference method, bovine milk containing 2-4 million RBC/mL has a visible reddish tinge. Milk with 10 million RBC/mL is clearly red.
The output from a practical optical in-line sensor could be reported in terms of either Hb concentration or the RBC/mL of milk. The most accurate and precise output would be in terms of Hb concentration. However, an RBC count is likely to be more acceptable and more easily understood because consumers and farmers are already 'in tune' with visualizing cell counts.
A sensor developed by Sensortec was used in-line to optically measure blood concentrations in milk from each quarter of 28 cows milked in NZ's first AMS. Most of these cows were freshly calved or in early lactation. Each day for a period of 16 days, either composite udder milk samples were collected for comparison by means of a Lely shuttle auto sampler or separate quarter samples were taken from the individual milk tubes during the first minute of milking. All samples were tested within 24 hours to determine the blood concentration of the milk. Based on a theoretical threshold of 10 M RBC / mL, 6 out of 6 samples were correctly identified (a Sensitivity of 100%). Specificity was 99.6%, 473 out of 475 samples being correctly identified with < 10 M RBC / mL.
Subsequently, the performance of the sensor was measured across a range of variables including: haemoglobin density, inter-sensor and inter-sample variation, milk temperature and sensor temperature. The overall precision is shown below.

© Animal Sciences Group - Wageningen UR. Last update: 20-02-2008 10:03. Mail any questions and remarks to: webmaster.asg@wur.nl