Texture features’ based classification of MR images of normal and herniated intervertebral discs

Abstract

Disc herniation is considered as a very common spine abnormality resulting in severe pain in back and legs. Besides it has great impact on economy of suffering patients also there is a concern about the shortage of radiologists and hence demand for computer aided diagnosis system. In this paper statistical texture features have been used for the classification of a normal intervertebral disc and a herniated intervertebral disc from MRI sequences acquired in sagittal plane. The main objective of this work was to appraise about the capability of texture features obtained from the intervertebral disc MR images and distinguish between normal intervertebral disc and herniated intervertebral disc using three different classifiers, namely, BPNN, KNN and SVM. The regions of interest (ROI) from patients with herniated discs were extracted by experienced radiologist from SKIMS institute, Srinagar. Three techniques where applied to each ROI to obtain texture features, which are, grey level run length matrix (GLRLM), grey level co-occurrence matrix (GLCM), and grey level difference method (GLDM). The results obtained show that GLRLM texture features ascertain a good discrimination capability to differentiate between a normal intervertebral disc and a herniated disc when SVM was used. Texture features extracted from GLCM present a good discrimination ability to differentiate between a normal intervertebral disc and a herniated disc when K-NN and BPNN classifiers were used. It is found that the selected set of features of the GLCM can discriminate a normal intervertebral disc from a herniated one, much accurately, on using K-NN and BPNN classifiers. On comparing the classification accuracies of K-NN and BPNN it is found that BPNN gives better results. K-NN is a simple algorithm to understand and implement but is slower because it starts learning from the testing data. As far as SVM is considered, selected set of features of GLRLM discriminates a normal intervertebral disc from a herniated one with good classification accuracy as compared to the others.