APPLICATION OF AN ARTIFICIAL NEURAL NETWORK FOR THE QUANTITATIVE CLASSIFICATION OF TROCHLEAR DYSPLASIA

A method that provides the ability to accurately classify trochlea dysplasia would be a valuable tool in treating patients with unstable patellas and anterior knee pain. In this study, a reference frame and a standardised femoral parameter measurement method for three-dimensional models were established to measure key femoral parameters. An artificial neural network was then trained with these parameters and the matching output from qualitative classifications by three orthopaedic surgeons for each knee. The neural network was then evaluated to test its ability for quantitative classification of normal and dysplastic knees. The maximum agreement between the qualitative and quantitative classification methods was found to be 80.6%, whereas agreement between the surgeons was 69.44%. This was achieved for a neural network with 8 input parameters and 19 hidden-layer neurons. The study shows that there is merit in making use of a trained artificial neural network as an additional tool for classification of trochlear dysplasia.