Fitness Functions Evaluation for Segmentation of Lymphoma Histological Images Using Genetic Algorithm

Abstract

For disease monitoring, grade definition and treatments orientation, specialists analyze tissue samples to identify structures of different types of cancer. However, manual analysis is a complex task due to its subjectivity. To help specialists in the identification of regions of interest, segmentation methods are used on histological images obtained by the digitization of tissue samples. Besides, features extracted from these specific regions allow for more objective diagnoses by using classification techniques. In this paper, fitness functions are analyzed for unsupervised segmentation and classification of chronic lymphocytic leukemia and follicular lymphoma images by the identification of their neoplastic cellular nuclei through the genetic algorithm. Qualitative and quantitative analyses allowed the definition of the Rényi entropy as the most adequate for this application. Images classification has reached results of 98.14% through accuracy metric by using this fitness function.