Analysis of breast thermograms using structure tensors

Asymmetry analysis in breast infrared images plays a significant role in the early diagnosis of breast cancer detection. In this work, an analysis on the characteristics of normal and abnormal breast tissues in thermal images is carried out using structure tensor features. Thermograms with normal tissue and any of three different neoplasms such as carcinoma, nodule and fibro adenoma are considered in this analysis. The images are obtained from online database for mastology research. The breast tissue is segmented by multiplying the raw image with the ground truth mask. Region of interest is defined manually to remove the extra regions such as armpits, neck and shoulder. Left and right breast tissues are separated using an automated algorithm. Normal and abnormal tissues are grouped according to the health and diseased conditions of the separated tissues. Structure tensor features such as coherence, orientation, energy and anisotropy index are extracted from the normal and abnormal tissues. All features except coherence show expected variations among normal and abnormal conditions. Further, it is noticed that the orientation and anisotropy index of three abnormal conditions show distinct variations due to varied metabolic activity. It is found that the structure tensor features could be one among the biomarkers that can be used to identify normal and various pathological conditions in breast thermograms.