Merging Scheme-based Classification of Medical X-ray Images

Abstract

Due to rapid growth of computerized medical imagery, the research area of medical image classification has been very active for the past decade. This paper presents an approach to achieve high recognition rate from classification of medical x-ray images. The methodology is based on local binary pattern as a feature extraction technique and support vector machine (SVM) as a classifier. This classification model was built based on merging scheme where overlapped classes were combined with each other and SVM classifier was re-trained to construct the model. The overlapped classes used in merging scheme are detected based on their accuracy, miss-classification ratio and similarity in their body anatomy. The proposed algorithm was evaluated on a database consisting of 36 classes of medical X-ray images which are suffering from high inter-class similarity and intra-class variability. The accuracy rate obtained for this model is 91%.