Syndrome approach for computer recognition of fuzzy systems and its application to immunological diagnostics and prognosis of human cancer

Abstract

We propose a novel pattern recognition method for the processing of a great number of “fuzzy” features in small groups of objects with small differences of features among the groups and with the missing of some data. Our algorithm dealing with statistical weighted syndromes (SWS) is based on the three-level feedback processing of fuzzy sets: recasting procedures of continual features to the gradations, optimization procedures of selection of the informative subareas (syndromes) in the feature space, and statistically weighted calculation of these subareas for prognosis of the class number function. At the training phase of the SWS method, the special likelihood function for the distribution of objects on the subareas is constructed. At all algorithm steps for control of stabilization and reproducing of the results, the jackknife procedure is used. The SWS method has been applied to analysis of flow cytometry measurements of the lymphocyte subpopulations in blood before treatment of osteosarcoma patients. We selected the small sets of the features from among about 40 immunological features which allowed us to predict correctly the widespread metastases and to forecast successfully the outcome of preoperation chemotherapy of the patients with osteosarcoma. Thus, for the first time, the information about lymphocyte subpopulations in the blood is effectively used for individual diagnostics, staging and prognosis of patients with the low-immunogeneic solid human tumor.