Recognition of Osteoporosis Based on Texture Analysis and a Support Vector Machine

Abstract

To explore a new approach for osteoporosis recognition with images, a texture analysis was made of 27 bone tissue images (16 of which from a SHAM group, and 11 from an OVX group). Texture features were then extracted through a co-occurrence matrix and a run-length matrix, and the texture features with significant differences between the two groups were chosen and used as the features in the classification course based on a Support Vector Machine (SVM). The results show that there are obvious statistic differences between the SHAM group and the OVX group in terms of texture features. Furthermore, the highest recognition accuracy was achieved at 92.59%. A SVM based on a linear function showed the highest accuracy rate and the lowest error rate in recognition. This new approach can be used to recognize osteoporosis effectively and thus can provide a valuable reference to the clinical application in osteoporosis diagnosis with medical images.

Keywords

• Texture analysis
• Support vector machine (SVM)
• Osteoporosis
• Co-occurrence matrix
• Run-length matrix