Detection of lameness and determination of the affected forelimb in horses by use of continuous wavelet transformation and neural network classification of kinematic data.

Objective: To investigate continuous wavelet transformation and neural network classification of gait data for detecting forelimb lameness in horses.

Animals: 12 adult horses with mild forelimb lameness.

Procedure: Position of the head and right forelimb foot, metacarpophalangeal (ie, fetlock), carpal, and elbow joints was determined by use of kinematic analysis before and after palmar digital nerve blocks. We obtained 8 recordings from horses without lameness, 8 with right forelimb lameness, and 8 with left forelimb lameness. Vertical and horizontal position of the head and vertical position of the foot, fetlock, carpal, and elbow joints were processed by continuous wavelet transformation. Feature vectors were created from the transformed signals and a neural network trained with data from 6 horses, which was then tested on the remaining 2 horses for each category until each horse was used twice for training and testing. Correct classification percentage (CCP) was calculated for each combination of gait signals tested.

Results: Wavelet-transformed vertical position of the head and right forelimb foot had greater CCP (85%) than untransformed data (21%). Adding data from the fetlock, carpal, or elbow joints did not improve CCP over that for the head and foot alone.

Conclusions And Clinical Relevance: Wavelet transformation of gait data extracts information that is important for the detection and differentiation of forelimb lameness of horses. All of the necessary information to detect lameness and differentiate the side of lameness can be obtained by observation of vertical head movement in concert with movement of the foot of 1 forelimb.