The thermometry data mining in the diagnostics of mammary glands

The paper deals with the microwave thermometry data and proposes a method for forming informative features, based on qualitative descriptions of medical knowledge about the behavior of the temperature fields of mammary glands. There are some already known qualitative characteristics of breast cancer, which were form a basis for quantitative features, e.g. feature ‘a large thermal asymmetry between mammary glands’ can be described by temperature differences between corresponding points of right and left mammary glands. If the value of such difference is sufficiently large, then it may be an indication of pathology. After preprocessing, which involves weighting and cleaning, informative features may be applied in various classification algorithms, such as logistic regression, which yields about seventy percents of accuracy on a test sample, or the more complex ones that yield better accuracy: neural networks, genetic algorithm, and fuzzy classification. The significance of proposed features consists of the fact that they were formed from qualitative characteristics and each of them has a qualitative description, therefore they are of interest for further study and can be applied in diagnosis-advisory systems.