We consider the problem of diagnosing breast cancer on the basis of combined thermometry. Diagnosis is carried out using two-dimensional generalized signs. Each such sign is formed on the basis of two one-dimensional. We propose an algorithm for the formation of sets of such features using tapering ellipses. On the basis of a set of such signs a classifier is built. By varying the semi-axes of the ellipses, the physician can control the sensitivity and specificity of the classifier. The doctor can fix the threshold value of specificity and achieve the maximum value of sensitivity. As a result of computational experiments, we built a family of 300 sets of two-dimensional signs and implemented a classifier customized by the doctor.
Keywords: combined thermometry, diagnostic algorithms, data analysis, algorithm sensitivity, algorithm specificity, logistic regression, custom classifier, breast screening, mammology