A Model Based On Genetic Algorithm For Colorectal Cancer Diagnosis

In this paper we present a method based on genetic algorithm capable of analyzing a significant number of features obtained from fractal techniques, Haralick texture features and curvelet coefficients, as well as several selection methods and classifiers for the study and pattern recognition of colorectal cancer. The chromosomal structure was represented by four genes in order to define an individual. The steps for evaluation and selection of individuals as well as crossover and mutation were directed to provide distinctions of colorectal cancer groups with the highest accuracy rate and the smallest number of features. The tests were performed with features from histological images H&E, different values of population and iterations numbers and with the k-fold cross-validation method. The best result was provided by a population of 500 individuals and 50 iterations applying relief, random forest and 29 features (obtained mainly from the combination of percolation measures and curvelet subimages). This solution was capable of distinguishing the groups with an accuracy rate of 90.82% and an AUC equal to 0.967.