Feature extraction and classification of thorax x-ray image in the assessment of osteoporosis

Previous studies showed that it was possible to have a prediction or an early detection of osteoporosis by measuring the thickness of the cortex of the clavicle of thorax x-ray image. The drawback of this system was that it was still dependent on the operator of subjective vision applications in the measurement. In addition, the accuracy of the system very much relied on the x-ray image quality. Therefore, it is in urgent need of another system which can automatically classify x-ray image and another method of image processing to identify and acknowledge a certain texture of the based image using a set of classes or texture classification given. In this paper, calculation and analysis of a series of image processing algorithms to perform x-ray image classification are done using the K-Nearest Neighbor (KNN) and feature extraction techniques Gray Level Co-occurrence Matrix (GLCM) on small sample size data of 46 Thorax x-ray images of 44 females and 2 males with the average age of 63 years old. T-score of these images had been measured using DEXA scan before as a justification. The proposed method shows that the clavicle cortex thickness measurement using GLCM and KNN method as feature extraction and image classification has its sensitivity of 100% and specificity of 90%. Furthermore, the accuracy which is obtained from the entire implementation capability in correctly assessing osteoporosis is 97.83%. Thus, it is evident that it is significantly correlated with predetermined T-score of DEXA in the assessment of osteoporosis.