Could Digital Technologies Lead to Improved Lameness
Detection on Dairy Farms? - Abstract
Sarah Hertle 1, Isabella Lorenzini 1, Sophia Sauter 1 and Bernhard Haidn 1
1
    Bavarian State Research Center for Agriculture, Prof.-Dürrwachter-Platz 2, Poing, 85586, Germany


                 Summary 1
                 Lameness is a major and common problem in dairy farms, which is unfortunately still
                 underestimated by most farmers. Regularly and systematically observing the gait pattern of
                 every single cow on a farm is a very time-consuming and error-prone task. Indirect automatic
                 lameness detection could enable farmers to detect lame cows earlier based on their individual
                 behaviour and performance, which are recorded by animal-specific sensor systems. Preceding
                 studies about indirect automatic lameness detection already showed that prediction models
                 could correctly distinguish between lame and non-lame cows with a probability of over 85%.
                 In these studies, pedometers (ENGS Dairy Solutions, Israel) which could register activity, lying
                 and feeding behaviours, were the only employed animal attached sensors.
                 The aim of the dissertation project “Automatic lameness detection on dairy farms - suitability
                 of digital technologies for recording behaviour and performance data” as part of the
                 experimental field DigiMilch is to refine and further develop these algorithms with data from
                 many different sensor systems installed on eight Bavarian dairy farms.
                 Individual animal behaviour and performance data recorded by boluses, pedometers, collars
                 and milking robots from different manufacturers are collected on three public research farms
                 and five commercial dairy farms throughout Bavaria. The required reference data to train
                 prediction models in combination with the acquired information by sensor systems consist of
                 locomotion scores performed through video recordings, diagnoses of the claw trimmings and
                 a pain test carried out before the trimming on each claw. According to the applied locomotion
                 score, cows with an irregular, asymmetric and uneven gait are scored as “lame” (score=3),
                 cows without gait alterations, but showing signs like head bobbing, an arched back or a
                 compensatory posture are considered “unsound” (score=2), and animals who do not present
                 any of these features are categorised as “sound” (score=1). This score has been validated by
                 calculating the inter- and intra-rater agreement and by creating a comparable three-step lesion
                 score which rates cows according to their visible lesions and pain reaction in groups from one
                 to three. After finishing data collection on the farms, the data is split in two parts to train and
                 then test specific prediction models with the collected information about claw health, behaviour
                 and performance data of every animal.
                 To assess the validity of the reference method, the lesion and locomotion score were compared.
                 The analysis of a dataset of 110 cows revealed results (cohen’s kappa (K)=0.72, confidence
                 interval (CI)=0.58-0.86, percentage of agreement (PA)=80%), which imply that the agreement
                 between those scores might be moderate to almost perfect. For the intra-rater agreement, the
                 computed PA yielded 93% and the K 0.89 (CI=0.84-0.94), indicating an almost perfect
                 agreement. The inter-rater reliability resulted in a PA=82 % and K=0.72 (CI=0.64-0.81),
                 meaning it was substantial to almost perfect.
                 The validation of the locomotion score showed that in this study the three-point locomotion
                 scoring system was a reliable reference system for claw health. In order to generalise and
                 confirm the presented results additional comparable data sets from other farms and observers
                 should be examined. Further analysis and modelling with the recorded sensor and claw health
                 data is planned after finishing the data collection. This wider approach with various sensor
                 systems increases the practical relevance of indirect automatic lameness detection and could
                 support farmers in optimising the animal health on their farm with minimal investment.

Proceedings of HAICTA 2022, September 22–25, 2022, Athens, Greece
EMAIL: Sarah.Hertle@lfl.bayern.de (A. 1); Isabella.Lorenzini@lfl.bayern.de (A. 2); SophiaAnna-Maria.Sauter@lfl.bayern.de (A. 3);
Bernhard.Haidn@lfl.bayern.de (A. 4)
              ©️ 2022 Copyright for this paper by its authors.
              Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
              CEUR Workshop Proceedings (CEUR-WS.org)


                                                                                     2
Keywords
Automatic lameness detection, claw health, sensor systems, dairy farming

Acknowledgements
The project is supported by funds of the Federal Ministry of Food and Agriculture (BMEL)
based on a decision of the Parliament of the Federal Republic of Germany. The Federal Office
for Agriculture and Food (BLE) provides coordinating support for digitalisation in agriculture
as funding organisation.




                                         3