Analysis of the Influence of Color Normalization in the Classification of Non-Hodgkin Lymphoma Images

In this work, a method is proposed to analyze the influence of color normalization in the classification lymphoma images. The approach combines multidimensional fractal techniques, curvelet transforms and Haralick features. The method considered a feature selection technique and different classification approaches to evaluate the combinations, such as decision tree, random forest, support vector machine, naive bayes and k-star. The classifications were analyzed considering three common lymphoma classes: mantle cell lymphoma, follicular lymphoma and chronic lymphocytic leukemia. The best result was achieved for the extraction from input images, features obtained mostly from lacunarity and percolation from curvelet sub-images, using random forest classifier. The tests were considered with 10-fold cross-validation and the result was a rate of AUC=0.963. The color normalization was not able to provide relevant classification rates. The obtained performance with the analysis over different types of features, classifiers and color normalization influence are important contributions to the identification of the lymphoma cancer.