Computer-aided classification of femur fractures

Classification or determining the types of fractured bones is the most important step of fracture treatment. Different fracture cases may be observed in daily life and each of them may require a specific treatment. According to the AO Müller [1] method, prominent of those most known fracture classifications, different kinds of 117 long- bone fractures in the human body may be seen. It is a difficult and burdensome procedure for any physician to know all those kinds of fracture types and treatments by heart. For this reason, decision support systems which will lessen the burden of work are needed to help the diagnosis and the treatment. Based on this need, in this thesis, we propose an auxiliary tool called a DIaphyseal Femur FRActure Classifier SysTem (DIFFRACT). DIFFRACT is a fully-automated computer aided diagnosis system, which is composed of different design steps. In first step of the proposed system, Niblack thresholding method is used to eliminate the background information in the X-ray images. According to the various experiments performed, Niblack, compared to the other segmentation methods, leads to less informational loss in the region of fractured bones although it makes more noise. Therefore, in the DIFFRACT, Niblack is preferred as a segmentation method. In the next step, a new method, called as SVM-based sensitive noise remover (SSNR) is proposed to eliminate the noises occurred in the segmentation step. With the help of SSNR, bone fragments and noises can be sensitively differentiated. In the feature extraction step, a lot of new feature extraction methods such as number of fragments (NOF), angle of fracture (AFE), bone completeness indicator (BCI) and fracture region map (FRM) are developed in order to classify the fractures. In the last operational step, multi-class support vector machine is used to classify the types of fractures. 196 X-ray images including different types of fracture are used to evaluate the performance of DIFFRACT. According to the 10-fold cross validation method; DIFFRACT successfully classified the cases of AO-32 fracture with 89,87% accuracy. In addition, to evaluate the usability of DIFFRACT in the clinical setting, the differentiation performance was compared with two experienced physicians, and it has been observed that its success is close to the physicians. Therefore, the DIFFRACT may be used as supplementary tool for the determination of fractured femur bones by physicians. It may facilitate decision making process of the physicians.