Automatic Segmentation and Analysis of Thermograms Using Texture Descriptors for Breast Cancer Detection

Breast cancer is one of the leading causes for high mortality rates among young women, in the developing countries. In Latin American this is an important health problem, for instance in Brazil and Ecuador this is the leading cause of cancer among women around 35 years old. Currently mammography is used as the gold standard for screening breast cancer. However, for young women mammograms are not recommended due the low contrast it presents on dense breasts and alternative techniques must be considered for this purpose. The World Health Organization states that screening programs are the more efficient way to combat this disease. Therefore it is fundamental to address new researches on early detection that are cost-effective and present advantages over the current method (based on the self-examination and mammography). The identification of such disease in early stage increases the prognosis and the survival rate. This article proposes to incorporate a low-cost and non-invasive diagnostic technique based on the use of thermal imaging. A textural analysis (by using statistical descriptors for automatic detection of abnormality in breast thermo grams) is considered for features as well as statistics measures computed from thermo grams' ROI (region of interest). Theses features feed a Nearest Neighbors classifier, where abnormal breasts are was identified with an accuracy of 94.44 %. The results of the study show that using simple textures descriptors, appropriate filtering and enhancement techniques it is possible to detected early onset of breast tumor in women of any age, with breasts of any density or size and even in pregnant women.