A Comparative Analysis of Static and Dynamic Thermal Imaging in Breast Cancer Screening

Breast cancer is affecting higher percentage of women across the world. Early detection can significantly decrease the breast cancer fatality rate. Thermogram is a new imaging modality that fulfils the above requirement by capturing the thermal radiations emitted from the breast region. In the proposed work, frontal breast thermograms from DMR-IR data-base acquired using static and dynamic protocol are considered for analysis. The frontal images are denoised using anisotropic diffusion filtering followed by breast region extraction using level set segmentation without reinitialization. In dynamic case, the selected five frontal dynamic thermograms are registered and fused to obtained fused image. Texture analysis is performed by extracting radiomic features from static frontal and fused frontal dynamic thermograms. Statistically significant features are determined by conducting 2-tailed Wilcoxon rank sum test on the extracted texture features. These significant features are fed to Least-Square-Support-Vector-Machine (LSSVM) employing different kernel types. LS-SVM employing linear kernel produced maximum accuracy of 92.85% for static frontal thermograms whereas LS-SVM employing Radial Basis Function (RBF) kernel produced maximum accuracy of 75%. Thus, thermal analysis using static frontal thermograms outperformed dynamic frontal thermograms.