A Hybrid Two-Stage CNN-SVM Model for Bone X-Rays Classification and Abnormality Detection

Abstract

This paper introduces a novel automatic reliable hybrid two-stage method for bone x-rays abnormality detection. For this purpose, 10 different pre-trained convolutional neural networks architectures with different number of layers are examined. The introduced method considers the seven extremity upper bones, namely shoulder, humerus, forearm, elbow, wrist, hand, and finger. The enhanced images are fed into the first stage to classify the bone type into one of the seven bones. Thereafter, the abnormality is detected in the second stage using a specific classifier according to the bone type. Thus, the classification step consists of eight different classifiers: one for the bone classification stage and seven for the abnormality detection stage. Finally, support vector machine layer is examined as a last layer of the classification in the second stage. The best average sensitivity and specificity achieved by the first stage are 95.78% and 99.45%, and 83.25% and 83.25% for the second stage, respectively. All the experiments were carried out using MURA dataset.