Prediction of continuous and discrete kinetic parameters in horses from inertial measurement units data using recurrent artificial neural networks.

Vertical ground reaction force (GRFz) measurements are the best tool for assessing horses' weight-bearing lameness. However, collection of these data is often impractical for clinical use. This study evaluates GRFz predicted using data from body-mounted IMUs and long short-term memory recurrent neural networks (LSTM-RNN). Twenty-four clinically sound horses, equipped with IMUs on the upper-body (UB) and each limb, walked and trotted on a GRFz measuring treadmill (TiF). Both systems were time-synchronised. Data from randomly selected 16, 4, and 4 horses formed training, validation, and test datasets, respectively. LSTM-RNN with different input sets (All, Limbs, UB, Sacrum, or Withers) were trained to predict GRFz curves or peak-GRFz. Our models could predict GRFz shapes at both gaits with RMSE below 0.40 N.kg-1. The best peak-GRFz values were obtained when extracted from the predicted curves by the all dataset. For both GRFz curves and peak-GRFz values, predictions made with the All or UB datasets were systematically better than with the Limbs dataset, showing the importance of including upper-body kinematic information for kinetic parameters predictions. More data should be gathered to confirm the usability of LSTM-RNN for GRFz predictions, as they highly depend on factors like speed, gait, and the presence of weight-bearing lameness.

Improving gait classification in horses by using inertial measurement unit (IMU) generated data and machine learning.

For centuries humans have been fascinated by the natural beauty of horses in motion and their different gaits. Gait classification (GC) is commonly performed through visual assessment and reliable, automated methods for real-time objective GC in horses are warranted. In this study, we used a full body network of wireless, high sampling-rate sensors combined with machine learning to fully automatically classify gait. Using data from 120 horses of four different domestic breeds, equipped with seven motion sensors, we included 7576 strides from eight different gaits. GC was trained using several machine-learning approaches, both from feature-extracted data and from raw sensor data. Our best GC model achieved 97% accuracy. Our technique facilitated accurate, GC that enables in-depth biomechanical studies and allows for highly accurate phenotyping of gait for genetic research and breeding. Our approach lends itself for potential use in other quadrupedal species without the need for developing gait/animal specific algorithms.

Validation of distal limb mounted inertial measurement unit sensors for stride detection in Warmblood horses at walk and trot.

BACKGROUND: Inertial measurement unit (IMU) sensor-based techniques are becoming more popular in horses as a tool for objective locomotor assessment. OBJECTIVES: To describe, evaluate and validate a method of stride detection and quantification at walk and trot using distal limb mounted IMU sensors. STUDY DESIGN: Prospective validation study comparing IMU sensors and motion capture with force plate data. METHODS: A total of seven Warmblood horses equipped with metacarpal/metatarsal IMU sensors and reflective markers for motion capture were hand walked and trotted over a force plate. Using four custom built algorithms hoof-on/hoof-off timing over the force plate were calculated for each trial from the IMU data. Accuracy of the computed parameters was calculated as the mean difference in milliseconds between the IMU or motion capture generated data and the data from the force plate, precision as the s.d. of these differences and percentage of error with accuracy of the calculated parameter as a percentage of the force plate stance duration. RESULTS: Accuracy, precision and percentage of error of the best performing IMU algorithm for stance duration at walk were 28.5, 31.6 ms and 3.7% for the forelimbs and -5.5, 20.1 ms and -0.8% for the hindlimbs, respectively. At trot the best performing algorithm achieved accuracy, precision and percentage of error of -27.6/8.8 ms/-8.4% for the forelimbs and 6.3/33.5 ms/9.1% for the hindlimbs. MAIN LIMITATIONS: The described algorithms have not been assessed on different surfaces. CONCLUSIONS: Inertial measurement unit technology can be used to determine temporal kinematic stride variables at walk and trot justifying its use in gait and performance analysis. However, precision of the method may not be sufficient to detect all possible lameness-related changes. These data seem promising enough to warrant further research to evaluate whether this approach will be useful for appraising the majority of clinically relevant gait changes encountered in practice.

Automated attention flags in chronic disease care planning.

OBJECTIVES: To assess the value of computerised decision support in the management of chronic respiratory disease by comparing agreement between three respiratory specialists, general practitioners (care coordinators), and decision support software. METHODS: Care guidelines for two chronic obstructive pulmonary disease projects of the SA HealthPlus Coordinated Care Trial were formulated. Decision support software, Care Plan On-Line (CPOL), was created to represent the intent of these guidelines via automated attention flags to appear in patients' electronic medical records. For a random sample of 20 patients with care plans, decisions about the use of nine additional services (eg, smoking cessation, pneumococcal vaccination) were compared between the respiratory specialists, the patients' GPs and the CPOL attention flags. RESULTS: Agreement among the specialists was at the lower end of moderate (intraclass correlation coefficient [ICC], 0.48; 95% CI, 0.39-0.56), with a 20% rate of contradictory decisions. Agreement with recommendations of specialists was moderate to poor for GPs (kappa, 0.49; 95% CI, 0.33-0.66) and moderate to good for CPOL (kappa, 0.72; 95% CI, 0.55-0.90). CPOL agreement with GPs was moderate to poor (kappa, 0.41; 95% CI, 0.24-0.58). GPs were less likely than specialists or CPOL to decide in favour of an additional service (P<0.001). CPOL was 87% accurate as an indicator of specialist decisions. It gave a 16% false-positive rate according to specialist decisions, and flagged 61% of decisions where GPs said No and specialists said Yes. CONCLUSIONS: Automated decision support may provide GPs with improved access to the intent of guidelines; however, further investigation is required.