Topological analysis for osteoporosis assessment in X-Ray images using random forest model

Abstract

This paper deals with the characterization of the texture of the bone. The objective is to combine clinical parameters and topological attributes extracted from X-Ray images for osteoporosis assessment. A total of 120 women are included in this study, divided into two populations, composed of 60 healthy subjects, and 60 osteoporotic patients. Nine features are involved and a random forest model is used to differentiate the patients. Different configurations are tested, trained, and validated using k-fold cross-validation technique with different values of k. The aim is to seek the combination giving the best accuracy for discriminating between the populations. Several classifiers are also tested and compared. The obtained results affirm that the tenfold cross-validation technique with the random forest model combining the topological and the clinical parameters outperform the other classifiers, and provide high accuracy (ACC = 87,5%) demonstrating the efficiency of this model