Evaluating Textural Changes of Lung in CT Images using GLCM in Comparison with GLRLM

The aim of this analysis is to identify the textural alterations due to incidence of COVID-19 in lung CT scan images using GLCM matrix in comparison with GLRLM. Materials and Methods: Sample size is calculated using G power analysis and a total of 176 sample sizes are acquired for this novel texture analysis using parameters like effect size (0.3), standard error rate (0.05), maximum rate (0.8) and allocation rate (N2/N1=1). For this analysis the required CT images are collected from Github. For group 1 a total of 94 sample images are taken and for group 2 a total of 82 sample images are taken. For analyzing the textural alterations of CT scan lung images, comparison between Gray Level Co-occurrence Matrix (GLCM) and Gray Level Run Length Matrix (GLRLM) is carried out for this analysis. In the process of evaluation of classifiers 10-fold cross validation is performed. Normal and COVID subjects are classified using Random forest, K-NN, Logistic regression classifiers for better classification. Results and Discussion: Due to incidence of COVID in lunge tissues it is observed that textural alterations are formed in lung CT scan images. From the acquired features values of GLCM and GLRLM it is observed that GLCM is statistically significant than the GLRLM. Contrast, homogeneity and sum of average features are statistically significant (0.0001) in identifying normal and COVID subjects. The mean value of homogeneity for healthy controls is (0.215) and for COVID subjects it is (0.327) such that normal subjects have a gentle surface of the lung and COVID subjects have rough surface and significance value is (p<0.05). GLCM has acquired precision (0.931), F1-score (0.928), Recall (0.929), AUC (0.981), Classification Accuracy (0.929) are obtained using random forest classifiers. From the above values it is observed that COVID subjects have textural variations than the normal subjects. Conclusion: From this analysis it is observed that GLCM provides significantly better classification in differentiating the COVID and normal subjects than GLRLM.