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

This paper is focused on statistical, artificial neural networks and fuzzy logic based feature evaluation indices for determining the most/least clinically significant image cytometric prognostic factors for assessment of nodal involvement in breast cancer patients. Seven different prognostic factors, {tumour type, tumour grade, DNA ploidy, S-phase faction, G0G1/G2M ratio, minimum (start) and maximum (end) nuclear pleomorphism indices}, are assessed by means of a multilayer feedforward backpropagation neural networks based feature evaluation index as an artificial neural network approach, a fuzzy logic-based feature evaluation index derived from the fuzzy k-nearest neighbour classifier as a fuzzy logic method, and a logistic regression-based statistical analysis. The results suggest that the artificial neural network and fuzzy based indices may be more reliable than their statistical counterpart. Overall results obtained for all the three methods highlight the fact that only one method's outcome may not be adequate to reliably determine the most/least clinically important factors for assessment of nodal involvement in breast cancer patients. Our results appear to suggest that S-phase fraction and tumour type may be the most and least clinically significant markers, respectively, and should be closely investigated for the assessment of breast cancer nodal involvement