Validation of an automated system for at-slaughter assessment of footpad dermatitis and hock burn in broiler chickens

The automated assessment of footpad dermatitis (FPD) and hock burn via computer vision at the slaughterline could allow continuous monitoring of these important welfare indicators in broiler chickens. Such a system can be used to monitor the occurrence of FPD and hock burn over subsequent flocks and allow bench marking at different aggregation levels (farms, chains, countries, etcetera). This study aimed to compare the performance of a camera system (ChickenCheck software, CLK GmbH) against three human assessors. FPD (4 point scale based on relative size of the lesions) and hock burn (3 point scale based on absolute size of the lesions) were scored on both the left and right foot/hock according to severity from images captured by the camera system. A total of 50 images (100 feet/hocks) were used in the training dataset used to train the human assessors and 500 images (1,000 feet/hocks) were used in the validation dataset to determine measures of agreement and reliability for the discrete scoring system. Additionally, 100 images were used for lesion annotation (200 feet/hock) to determine the agreement between the camera system and the human assessors in capturing the total area of the foot pad or hock affected by lesions. Despite showing substantial to good intra-observer reliability (kappa: 0.78-0.98) and moderate to substantial inter-observer reliability (kappa: 0.46-0.66) during training, the human assessors showed fair to moderate agreement (kappa: 0.33-0.47) for FPD and hock burn when the dataset included a larger number of images with greater variation in lesion severity and appearance. However, the human assessors showed moderate to substantial agreement with the scores assigned by the camera system (0.60-0.70). More importantly, human assessors and camera system indicated a similar size of lesions on the footpad (R2: 0.93) and hock burn (R2: 0.97), suggesting that lesion size is correctly annotated by the camera system. However, the small but systematic differences found in the size of the affected area of FPD (+2.47%) and hock burn (-0.38 cm2) by the camera system compared to the human assessors need to be addressed to further improve the performance of the automated system. A validated system for automatic assessment of FPD and hock burn at slaughter allows continuous monitoring of all birds that are slaughtered and is a valuable tool to improve broiler welfare.