Soft feature evaluation indices for the identification ofsignificant image cytometric factors in assessment of nodal involvementin breast cancer patients

摘要

This paper is focused on statistical, artificial neural networksand fuzzy logic based feature evaluation indices for determining themost/least clinically significant image cytometric prognostic factorsfor assessment of nodal involvement in breast cancer patients. Sevendifferent prognostic factors, {tumour type, tumour grade, DNA ploidy,S-phase faction, G₀G₁/G₂M ratio,minimum (start) and maximum (end) nuclear pleomorphism indices}, areassessed by means of a multilayer feedforward backpropagation neuralnetworks based feature evaluation index as an artificial neural networkapproach, a fuzzy logic-based feature evaluation index derived from thefuzzy k-nearest neighbour classifier as a fuzzy logic method, and alogistic regression-based statistical analysis. The results suggest thatthe artificial neural network and fuzzy based indices may be morereliable than their statistical counterpart. Overall results obtainedfor all the three methods highlight the fact that only one method'soutcome may not be adequate to reliably determine the most/leastclinically important factors for assessment of nodal involvement inbreast cancer patients. Our results appear to suggest that S-phasefraction and tumour type may be the most and least clinicallysignificant markers, respectively, and should be closely investigatedfor the assessment of breast cancer nodal involvement