Comparing local binary, weber local descriptor and local gradient increasing pattern features in differentiating the normal from COVID-19 subjects

• Published in: AIP conference proceedings Vol. 2822; no. 1
• Main Authors: Lavanya, V., Ramesh, M.
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Melville American Institute of Physics 14.11.2023
• Subjects: Classifiers; Computed tomography; Hand tools; Neural networks; Standard error
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0178981

To estimate the texture deformation in lung CT images caused by COVID 19 using local binary pattern and Weber local descriptor Pattern in comparison with local gradient increasing pattern features. Materials and Methods: G-power tool is used to calculate the total number of samples required for the analysis. A total of 176 samples per group is obtained by fixing the standard error rate, effect size, and predefined power parameters in G power tool are set to 0.05, 0.3, and 0.80, respectively. Results: The feature values obtained using Local Gradient Increasing Pattern are found to exhibit the significance of 0.0001 (p<0.05) compared to local binary pattern and Weber Local Descriptor. Significant Local Gradient Increasing Patterns are LGIP11, LGIP19, LGIP23, with mean values of normal as 0.42, 0.48, and 0.45 respectively and COVID as 0.35, 0.39, and 0.41 respectively. Neural network classifier is found to achieve a high accuracy of 0.98 using LGIP features in differentiating normal and COVID subjects compared to logistic regression and K-NN classifier. Conclusion: Local gradient increasing patterns perform significantly better than the local binary pattern and Weber local descriptor in differentiating normal and COVID subjects.