Technical note: Automatic evaluation of infrared thermal images by computerized active shape modeling of bovine udders challenged with Escherichia coli.

Mastitis causes substantial economic losses and animal suffering in the dairy industry. The trend toward larger herd sizes complicates the monitoring of udder health in individual animals. Infrared thermography has successfully been used for early mastitis detection. However, manual thermogram analysis is time consuming and requires a skilled examiner, and automated image processing has not been tested. The aim of this study was to determine whether automatic evaluation of thermograms showed results comparable to those of manual evaluation of thermograms. Five healthy cows underwent an intramammary challenge with Escherichia coli to induce clinical mastitis. Multiple udder thermograms were taken every 2 h for 24 h before and after the challenge, resulting in 4,143 images in total. All images were evaluated using image recognition software (automatically) and a polygon tool (manually) to calculate the average and maximum surface temperatures. Because of the slightly different regions of interest, temperatures ascertained from the thermograms using the automatic method were consistently lower than those ascertained using the manual method. However, average udder surface temperatures evaluated using both methods were strongly correlated (r = 0.98 in the left hindquarter, and r = 0.99 in the right hindquarter) and showed maximum temperature peaks at the same time, 13 and 15 h after intramammary challenge. In the receiver operating characteristic analysis, both methods provided good results for sensitivity and specificity in detecting clinical E. coli-induced mastitis at different threshold values. For automatically evaluated maximum right hindquarter temperature, sensitivity was 93.75% and specificity was 94.96%, and for manually evaluated maximum right hindquarter temperature, sensitivity was 93.75% and specificity was 96.40%. Thus, automatic thermogram evaluation is a promising tool for automated mastitis detection.