Robust Anisotropic Diffusion Based Edge Enhancement for Level Set Segmentation and Asymmetry Analysis of Breast Thermograms using Zernike Moments.

Breast thermography plays a major role in early detection of breast cancer in which the thermal variations are associated with precancerous state of breast. The distribution of asymmetrical thermal patterns indicates the pathological condition in breast thermal images. In this work, asymmetry analysis of breast thermal images is carried out using level set segmentation and Zernike moments. The breast tissues are subjected to Tukey’s biweight robust anisotropic diffusion filtering (TBRAD) for the generation of edge map. Reaction diffusion level set method is employed for segmentation in which TBRAD edge map is used as stopping criterion during the level set evolution. Zernike moments are extracted from the segmented breast tissues to perform asymmetry analysis. Results show that the TBRAD filter is able to enhance the edges near infra mammary folds and lower breast boundaries effectively. It is observed that segmented breast tissues are found to be continuous and has sharper boundary. This method yields high degree of correlation (98%) between the segmented output and the ground truth images. Among the extracted Zernike features, higher order moments are found to be significant in demarcating normal and carcinoma breast tissues by 9%. It appears that, the methodology adopted here is useful in accurate segmentation and differentiation of normal and carcinoma breast tissues for automated diagnosis of breast abnormalities.