A Primary Study on Application of Artificial Neural Network in Classification of Pediatric Fracture Healing Time of the Lower Limb

Abstract

In this study we examined the lower limb fracture in children and classified the healing time using supervised and unsupervised artificial neural network (ANN). Radiographs of long bones from 2009 to 2011 of lower limb fractures involving the femur, tibia and fibula from children ages 0 to 13 years, with ages recorded from the date and time of initial injury was obtained from the pediatric orthopedic unit in University Malaya Medical Centre. ANNs was developed using the following input: type of fracture, angulation of the fracture, displacement of the fracture, contact area of the fracture and age. Fracture healing time was classified into two classes that is less than 12 weeks which represent normal healing time in lower limb fractures and more than 12 weeks which could indicate a delayed union. This research was designed to evaluate the classification accuracy of two ANN methods (SOM, and MLP) on pediatric fracture healing. Standard feed-forward, back-propagation neural network with three layers was used in this study. The less sensitive variables were eliminated using the backward elimination method, and the ANN network was retrained again with minimum variables. Accuracy rate, area under the curve (AUC), and root mean square errors (RMSE) are the main criteria used to evaluate the ANN model results. We found that the best ANN model results was obtained when all input variables were used with overall accuracy percentage of 80%, with RMSE value of 0.34, and AUC value of 0.8. We concluded here that the ANN model in this study can be used to classify pediatric fracture healing time, however extra efforts are required to adapt the ANN model well by using its full potential features to improve the ANN performance especially in the pediatric orthopedic application.