Grey Level Co-Occurrence Matrix (GLCM) and Multi-Scale Non-Negative Sparse Coding For Classification of Medical Images

P. Dhanalakshmi and Dr.G. Satyavathy

A huge number of deaths happening in the world are caused by cancer, among the mortalities in the whole world highest causality is caused by cancer specifically lung cancer. Cancer can be detected at later stages by means of visual interpretation of database, hence paving way to late diagnosis of cancer which only increases the mortality rates caused by cancer. So, cancer can be detected way before by employing image processing tools. In this study, classification of lung cancer is applied by utilizing projected Grey Level Co-Occurrence Matrix (GLCM) + a classifier algorithm named Enhanced Particle Swarm Optimization Kernel Support Vector Machine (EPSOKSVM). This projected work contains four key stages like multiple scale layer decomposition, using GLCM for feature extraction, non-negative sparse coding method employing hybrid fisher discriminative analysis and EPSOKSVM classification. Initially, clinical pictures were disintegrated to form several scale layers; therefore it is possible to extract varied visual data‟s from various scale layers. Afterwards GLCM is employed for texture feature mining and the additional explanatory aspects are mined utilizing entropy, energy, contrast, correlation, homogeneity, and dissimilarity texture features in the lung cancer image database of LIDC-IDRI. So as to acquire this discriminative sparse depiction of medical images, the non-negative sparse coding technique along with hybrid fisher discriminative analysis is built for each and every scale layer. Then EPSOKSVM algorithm is executed for achieving better lung cancer results with advanced classification details from the database provided. By employing kernel feature the training and testing procedure is carried out and kernel feature usage allows us to gain result with high accuracy. The investigational outcomes attest that the projected GLCM+EPSOKSVM algorithm delivers improved classification outcomes regarding the parameters like advanced accurateness, precision, F-measure and recall values along with reduced error rate.

Volume 11 | 10-Special Issue

Pages: 481-493

DOI: 10.5373/JARDCS/V11SP10/20192835