Stefano Viazzi

Lameness detection based on multivariate continuous sensing of milk yield, rumination, and neck activity

The objective of this study was to develop and validate a mathematical model to detect clinical lameness based on existing sensor data that relate to the behavior and performance of cows in a commercial dairy farm. Identification of lame (44) and not lame (74) cows in the database was done based on the farms daily herd health reports. All cows were equipped with a behavior sensor that measured neck activity and ruminating time. The cows performance was measured with a milk yield meter in the milking parlor. In total, 38 model input variables were constructed from the sensor data comprising absolute values, relative values, daily standard deviations, slope coefficients, daytime and nighttime periods, variables related to individual temperament, and milk session-related variables. A lame group, cows recognized and treated for lameness, to not lame group comparison of daily data was done. Correlations between the dichotomous output variable (lame or not lame) and the model input variables were made. The highest correlation coefficient was obtained for the milk yield variable (rMY=0.45). In addition, a logistic regression model was developed based on the 7 highest correlated model input variables (the daily milk yield 4d before diagnosis; the slope coefficient of the daily milk yield 4d before diagnosis; the nighttime to daytime neck activity ratio 6d before diagnosis; the milk yield week difference ratio 4d before diagnosis; the milk yield week difference 4d before diagnosis; the neck activity level during the daytime 7d before diagnosis; the ruminating time during nighttime 6d before diagnosis). After a 10-fold cross-validation, the model obtained a sensitivity of 0.89 and a specificity of 0.85, with a correct classification rate of 0.86 when based on the averaged 10-fold model coefficients. This study demonstrates that existing farm data initially used for other purposes, such as heat detection, can be exploited for the automated detection of clinically lame animals on a daily basis as well.

Analysis of individual classification of lameness using automatic measurement of back posture in dairy cattle

Currently, diagnosis of lameness at an early stage in dairy cows relies on visual observation by the farmer, which is time consuming and often omitted. Many studies have tried to develop automatic cow lameness detection systems. However, those studies apply thresholds to the whole population to detect whether or not an individual cow is lame. Therefore, the objective of this study was to develop and test an individualized version of the body movement pattern score, which uses back posture to classify lameness into 3 classes, and to compare both the population and the individual approach under farm conditions. In a data set of 223 videos from 90 cows, 76% of cows were correctly classified, with an 83% true positive rate and 22% false positive rate when using the population approach. A new data set, containing 105 videos of 8 cows that had moved through all 3 lameness classes, was used for an ANOVA on the 3 different classes, showing that body movement pattern scores differed significantly among cows. Moreover, the classification accuracy and the true positive rate increased by 10 percentage units up to 91%, and the false positive rate decreased by 4 percentage units down to 6% when based on an individual threshold compared with a population threshold.

Manual and automatic locomotion scoring systems in dairy cows: A review

The objective of this review was to describe, compare and evaluate agreement, reliability, and validity of manual and automatic locomotion scoring systems (MLSSs and ALSSs, respectively) used in dairy cattle lameness research. There are many different types of MLSSs and ALSSs. Twenty-five MLSSs were found in 244 articles. MLSSs use different types of scale (ordinal or continuous) and different gait and posture traits need to be observed. The most used MLSS (used in 28% of the references) is based on asymmetric gait, reluctance to bear weight, and arched back, and is scored on a five-level scale. Fifteen ALSSs were found that could be categorized according to three approaches: (a) the kinetic approach measures forces involved in locomotion, (b) the kinematic approach measures time and distance of variables associated to limb movement and some specific posture variables, and (c) the indirect approach uses behavioural variables or production variables as indicators for impaired locomotion. Agreement and reliability estimates were scarcely reported in articles related to MLSSs. When reported, inappropriate statistical methods such as PABAK and Pearson and Spearman correlation coefficients were commonly used. Some of the most frequently used MLSSs were poorly evaluated for agreement and reliability. Agreement and reliability estimates for the original four-, five- or nine-level MLSS, expressed in percentage of agreement, kappa and weighted kappa, showed large ranges among and sometimes also within articles. After the transformation into a two-level scale, agreement and reliability estimates showed acceptable estimates (percentage of agreement ≥ 75%; kappa and weighted kappa ≥ 0.6), but still estimates showed a large variation between articles. Agreement and reliability estimates for ALSSs were not reported in any article. Several ALSSs use MLSSs as a reference for model calibration and validation. However, varying agreement and reliability estimates of MLSSs make a clear definition of a lameness case difficult, and thus affect the validity of ALSSs. MLSSs and ALSSs showed limited validity for hoof lesion detection and pain assessment. The utilization of MLSSs and ALSSs should aim to the prevention and efficient management of conditions that induce impaired locomotion. Long-term studies comparing MLSSs and ALSSs while applying various strategies to detect and control unfavourable conditions leading to impaired locomotion are required to determine the usefulness of MLSSs and ALSSs for securing optimal production and animal welfare in practice.

Effect of merging levels of locomotion scores for dairy cows on intra- and interrater reliability and agreement

Locomotion scores are used for lameness detection in dairy cows. In research, locomotion scores with 5 levels are used most often. Analysis of scores, however, is done after transformation of the original 5-level scale into a 4-, 3-, or 2-level scale to improve reliability and agreement. The objective of this study was to evaluate different ways of merging levels to optimize resolution, reliability, and agreement of locomotion scores for dairy cows. Locomotion scoring was done by using a 5-level scale and 10 experienced raters in 2 different scoring sessions from videos from 58 cows. Intra- and interrater reliability and agreement were calculated as weighted kappa coefficient (κw) and percentage of agreement (PA), respectively. Overall intra- and interrater reliability and agreement and specific intra- and interrater agreement were determined for the 5-level scale and after transformation into 4-, 3-, and 2-level scales by merging different combinations of adjacent levels. Intrarater reliability (κw) ranged from 0.63 to 0.86, whereas intrarater agreement (PA) ranged from 60.3 to 82.8% for the 5-level scale. Interrater κw=0.28 to 0.84 and interrater PA=22.6 to 81.8% for the 5-level scale. The specific intrarater agreement was 76.4% for locomotion level 1, 68.5% for level 2, 65% for level 3, 77.2% for level 4, and 80% for level 5. Specific interrater agreement was 64.7% for locomotion level 1, 57.5% for level 2, 50.8% for level 3, 60% for level 4, and 45.2% for level 5. Specific intra- and interrater agreement suggested that levels 2 and 3 were more difficult to score consistently compared with other levels in the 5-level scale. The acceptance threshold for overall intra- and interrater reliability (κw and κ ≥0.6) and agreement (PA ≥75%) and specific intra- and interrater agreement (≥75% for all levels within locomotion score) was exceeded only for the 2-level scale when the 5 levels were merged as (12)(345) or (123)(45). In conclusion, when locomotion scoring is performed by experienced raters without further training together, the lowest specific intra- and interrater agreement was obtained in levels 2 and 3 of the 5-level scale. Acceptance thresholds for overall intra- and interrater reliability and agreement and specific intra- and interrater agreement were exceeded only in the 2-level scale.

Image-processing technique to measure pig activity in response to climatic variation in a pig barn

In the past decades, the increasing scale of intensive pig farms led farmers to use automatic tools to monitor the welfare and health of their animals. Visual observation and manual monitoring, usually practiced in small-scale farms, is unreliable in large-scale husbandry, and is expensive and time consuming. Environmental parameters are crucial information for the efficient management of piggery buildings, as they have a significant effect on production efficiency, health and welfare of confined animals. The aim of the present study was to evaluate the relationship between pig activity and environmental parameters in a pig building by means of image analysis. The barn for 350 fattening pigs was open-space, mechanically ventilated and subdivided into 16 pens with fully slatted floor. The room was equipped to monitor the ventilation rate, internal and external temperature and relative humidity every minute. For the experiments, two adjacent pens were selected, each 5.9 by 2.6 m, with ~16 pigs in each. Pigs were continuously monitored during 30 days using an infrared-sensitive CCD camera that was mounted 5 m above the floor. Recorded data were processed in real time by Eyenamic, an innovative software that continuously and automatically registers the behaviour of a group of animals, intended as the activity and occupation indices of the pigs. A preliminary virtual subdivision of the two pens in four zones (two zones for each pen) was performed to evaluate differences in activity/occupation indices in ‘front’ and ‘back’ zones of the pen. Recorded images were visually observed in the laboratory to estimate pig activity type in relation to the indices calculated by Eyenamic software. The occupation index showed higher values (up to 0.75 units) in Zones 1 and 4 placed near the corridor. There was a significant relation between pig occupation index measured in the two pens and ventilation rate, temperature and humidity. The interaction between ventilation and humidity and temperature and humidity significantly affected pig movements during the day. Pigs tended to stay in the part of the pen far from the external wall, where air velocity was higher, probably because this is a ‘central zone’ in the barn, characterised by a reasonable air movement (~0.30 m/s). On the contrary, the part of the pen nearest to the external wall, characterised by a humid floor surface and by a limited air speed, was occupied by animals at the trough mainly during feeding times and for defecation and urination.

The effect of routine hoof trimming on locomotion score, ruminating time, activity, and milk yield of dairy cows

The objective of this study was to quantify the effect of hoof trimming on cow behavior (ruminating time, activity, and locomotion score) and performance (milk yield) over time. Data were gathered from a commercial dairy farm in Israel where routine hoof trimming is done by a trained hoof trimmer twice per year on the entire herd. In total, 288 cows spread over 6 groups with varying production levels were used for the analysis. Cow behavior was measured continuously with a commercial neck activity logger and a ruminating time logger (HR-Tag, SCR Engineers Ltd., Netanya, Israel). Milk yield was recorded during each milking session with a commercial milk flow sensor (Free Flow, SCR Engineers Ltd.). A trained observer assigned on the spot 5-point locomotion scores during 19 nighttime milking occasions between 22 October 2012 and 4 February 2013. Behavioral and performance data were gathered from 1wk before hoof trimming until 1wk after hoof trimming. A generalized linear mixed model was used to statistically test all main and interactive effects of hoof trimming, parity, lactation stage, and hoof lesion presence on ruminating time, neck activity, milk yield, and locomotion score. The results on locomotion scores show that the proportional distribution of cows in the different locomotion score classes changes significantly after trimming. The proportion of cows with a locomotion score ≥3 increases from 14% before to 34% directly after the hoof trimming. Two months after the trimming, the number of cows with a locomotion score ≥3 reduced to 20%, which was still higher than the baseline values 2wk before the trimming. The neck activity level was significantly reduced 1d after trimming (380±6 bits/d) compared with before trimming (389±6 bits/d). Each one-unit increase in locomotion score reduced cow activity level by 4.488 bits/d. The effect of hoof trimming on ruminating time was affected by an interaction effect with parity. The effect of hoof trimming on locomotion scores was affected by an interaction effect with lactation stage and tended to be affected by interaction effects with hoof lesion presence, indicating that cows with a lesion reacted different to the trimming than cows without a lesion did. The results show that the routine hoof trimming affected dairy cow behavior and performance in this farm.

Relation between observed locomotion traits and locomotion score in dairy cows

Lameness is still an important problem in modern dairy farming. Human observation of locomotion, by looking at different traits in one go, is used in practice to assess locomotion. The objectives of this article were to determine which individual locomotion traits are most related to locomotion scores in dairy cows, and whether experienced raters are capable of scoring these individual traits consistently. Locomotion and 5 individual locomotion traits (arched back, asymmetric gait, head bobbing, reluctance to bear weight, and tracking up) were scored independently on a 5-level scale for 58 videos of different cows. Videos were shown to 10 experienced raters in 2 different scoring sessions. Relations between locomotion score and traits were estimated by 3 logistic regression models aiming to calculate the size of the fixed effects on the probability of scoring a cow in 1 of the 5 levels of the scale (model 1) and the probability of classifying a cow as lame (locomotion score ≥3; model 2) or as severely lame (locomotion score ≥4; model 3). Fixed effects were rater, session, traits, and interactions among fixed effects. Odds ratios were calculated to estimate the relative probability to classify a cow as lame when an altered (trait score ≥3) or severely altered trait (trait score ≥4) was present. Overall intrarater and interrater reliability and agreement were calculated as weighted kappa coefficient (κw) and percentage of agreement, respectively. Specific intrarater and interrater agreement for individual levels within a 5-level scale were calculated. All traits were significantly related to the locomotion score when scored with a 5-level scale and when classified as (severely) lame or nonlame. Odds ratios for altered and severely altered traits were 10.8 and 14.5 for reluctance to bear weight, 6.5 and 7.2 for asymmetric gait, and 4.8 and 3.2 for arched back, respectively. Raters showed substantial variation in reliability and agreement values when scoring traits. The acceptance threshold for overall intrarater reliability (κw ≥0.60) was exceeded by locomotion scoring and all traits. Overall interrater reliability values ranged from κw=0.53 for tracking up to κw=0.61 for reluctance to bear weight. Intrarater and interrater agreement were below the acceptance threshold (percentage of agreement <75%). Most traits tended to have lower specific intrarater and interrater agreement in level 3 and 5 of the scale. In conclusion, raters had difficulties in scoring locomotion traits consistently, especially slight alterations were difficult to detect by experienced raters. Yet, the locomotion traits reluctance to bear weight, asymmetric gait, and arched back had the strongest relation with the locomotion score. These traits should have priority in locomotion-scoring-system guidelines and are the best to be used for the development of automated locomotion scoring systems.

How do pigs behave before starting an aggressive interaction? Identification of typical body positions in the early stage of aggression using video labelling techniques.

The aim of this study was to identify, quantify, and describe pre-signs of aggression in pigs and the early stages of aggressive interactions. The experiment was carried out at a commercial farm on a group of 11 male pigs weighing on average 23 kg and kept in a pen of4 m x 2.5 m. In total 8 hours were videorecorded during the first 3 days after mixing. As a result, 177 aggressive interactions were identified and labelled to find pre-sign body positions before aggressive interactions, attack positions and aggressive acts performed from these positions. A total of 12 positions were classified as pre-signs (P1-P12) and 7 of them were identified immediately at the start of aggressive interactions (P6-P12). Most common pre-sign positions were P3-pigs approaching and facing each other (24%) and P2-initiator pigs approaching from the lateral side (18%). In 80% of the cases the duration of pre-signs was 1-2 sec 72% of all aggressive interactions were short (1 to 10 sec). The most frequent attack positions were P12-inverse parallel (39.5%), P7-nose to nose, 90 degrees (19.77%) and P9-nose to head (13.5%). The most frequent aggressive acts from attack positions were head knocking (34.4%), pressing (34.4%) and biting of different body parts (29.4%). Head knocking was mostly observed in relation to P7 and P2 positions and biting was common in the P7 position. In conclusion, pigs adopt specific pre-signs and body positions before the escalation of aggressive interactions. This could be used as potential sign to identify a beginning aggression.

Lameness detection in dairy cattle: single predictor v . multivariate analysis of image-based posture processing and behaviour and performance sensing

The objective of this study was to evaluate if a multi-sensor system (milk, activity, body posture) was a better classifier for lameness than the single-sensor-based detection models. Between September 2013 and August 2014, 3629 cow observations were collected on a commercial dairy farm in Belgium. Human locomotion scoring was used as reference for the model development and evaluation. Cow behaviour and performance was measured with existing sensors that were already present at the farm. A prototype of three-dimensional-based video recording system was used to quantify automatically the back posture of a cow. For the single predictor comparisons, a receiver operating characteristics curve was made. For the multivariate detection models, logistic regression and generalized linear mixed models (GLMM) were developed. The best lameness classification model was obtained by the multi-sensor analysis (area under the receiver operating characteristics curve (AUC)=0.757±0.029), containing a combination of milk and milking variables, activity and gait and posture variables from videos. Second, the multivariate video-based system (AUC=0.732±0.011) performed better than the multivariate milk sensors (AUC=0.604±0.026) and the multivariate behaviour sensors (AUC=0.633±0.018). The video-based system performed better than the combined behaviour and performance-based detection model (AUC=0.669±0.028), indicating that it is worthwhile to consider a video-based lameness detection system, regardless the presence of other existing sensors in the farm. The results suggest that Θ2, the feature variable for the back curvature around the hip joints, with an AUC of 0.719 is the best single predictor variable for lameness detection based on locomotion scoring. In general, this study showed that the video-based back posture monitoring system is outperforming the behaviour and performance sensing techniques for locomotion scoring-based lameness detection. A GLMM with seven specific variables (walking speed, back posture measurement, daytime activity, milk yield, lactation stage, milk peak flow rate and milk peak conductivity) is the best combination of variables for lameness classification. The accuracy on four-level lameness classification was 60.3%. The accuracy improved to 79.8% for binary lameness classification. The binary GLMM obtained a sensitivity of 68.5% and a specificity of 87.6%, which both exceed the sensitivity (52.1%±4.7%) and specificity (83.2%±2.3%) of the multi-sensor logistic regression model. This shows that the repeated measures analysis in the GLMM, taking into account the individual history of the animal, outperforms the classification when thresholds based on herd level (a statistical population) are used.

Comparison of locomotion scoring for dairy cows by experienced and inexperienced raters using live or video observation methods.

Lameness is considered a major problem in dairy production. Lameness is commonly detected with locomotion scores assigned to cows under farm conditions, but raters are often trained and assessed for reliability and agreement by using video recordings. The aim of this study was to evaluate intra- and inter-rater reliability and agreement of experienced and inexperienced raters for locomotion scoring performed live and from video, and to calculate the influence of raters and the method of observation (live or video) on the probability of classifying a cow as lame. Using a five-level locomotion score, cows were scored twice live and twice from video by three experienced and two inexperienced raters for three weeks. Every week different cows were scored. Intra- and inter-rater reliability (expressed as weighted kappa, ?w)) and agreement (expressed as percentage of agreement, PA) for live/live, live/video and video/video comparisons were determined. A logistic regression was performed to estimate the influence of the rater and method of observation on the probability of classifying a cow as lame in live and video observation. Experienced raters had higher values for intra-rater reliability and agreement for video/video than for live/live and live/video comparison. Inexperienced raters, however, did not differ for intra- and inter-rater reliability and agreement for live/live, live/video and video/video comparisons. The logistic regression indicated that raters were responsible for the main effect and the method of observation (live or from video) had a minor effect on the probability for classifying a cow as lame (locomotion score =3). In conclusion, under the present experimental conditions, experienced raters performed better than unexperienced raters when locomotion scoring was done from video. Since video observation did not show any important influence in the probability of classifying a cow as lame, video observation seems to be an acceptable method for locomotion scoring and lameness assessment in dairy cows.