A study on the use of thermal imaging as a diagnostic tool for the detection of digital dermatitis in dairy cattle

• Published in: Journal of dairy science 2021-09, Vol.104 (9), p.10194-10202
• Main Authors: Anagnostopoulos, A., Barden, M., Tulloch, J., Williams, K., Griffiths, B., Bedford, C., Rudd, M., Psifidi, A., Banos, G., Oikonomou, G.
• Format: Article
• Language: English
• Subjects: digital dermatitis; infrared thermography; lameness; M-scoring
• ISSN: 0022-0302; 1525-3198
• DOI: 10.3168/jds.2021-20178

Our aims were to (1) determine how interdigital skin temperature (IST), measured using infrared thermography, was associated with different stages of digital dermatitis (DD) lesions and (2) develop and validate models that can use IST measurements to identify cows with an active DD lesion. Between March 2019 and March 2020, infrared thermographic images of hind feet were taken from 2,334 Holstein cows across 4 farms. We recorded the maximum temperature reading from infrared thermographic images of the interdigital skin between the heel bulbs on the hind feet. Pregnant animals were enrolled approximately 1 to 2 mo precalving, reassessed 1 wk after calving, and again at approximately 50 to 100 d postpartum. At these time points, IST and the clinical stage of DD (M-stage scoring system: M1–M4.1) were recorded in addition to other data such as the ambient environmental temperature, height, body condition score, parity, and the presence of other foot lesions. A mixed effect linear regression model with IST as the dependent variable was fitted. Interdigital skin temperature was associated with DD lesions; compared to healthy feet, IST was highest in feet with M2 lesions, followed by M1 and M4.1 lesions. Subsequently, the capacity of IST measurements to detect the presence or absence of an active DD lesion (M1, M2, or M4.1) was explored by fitting logistic regression models, which were tested using 10-fold validation. A mixed effect logistic regression model with the presence of active DD as the dependent variable was fitted first. The average area under the curve for this model was 0.80 when its ability to detect presence of active DD was tested on 10% of the data that were not used for the model's training; an average sensitivity of 0.77 and an average specificity of 0.67 was achieved. This model was then restricted so that only explanatory variables that could be practically recorded in a nonresearch, external setting were included. Validation of this model demonstrated an average area under the curve of 0.78, a sensitivity of 0.88, and a specificity of 0.66 for 1 of the time points (precalving). Lower sensitivity and specificity were achieved for the other 2 time points. Our study adds further evidence to the relationship between DD and foot skin temperature using a large data set with multiple measurements per animal. Additionally, we highlight the potential for infrared thermography to be used for routine on-farm diagnosis of active DD lesions.