Particle Swarm Estimation of Distribution Algorithm for Lymphoma Classification through Automatic Biopsies Analysis

In this paper we propose a new system for lymphoma classification through automatic biopsy images analysis. The system is composed by two main modules: a computer vision module that extracts numerical features from the biopsy images acquired from a microscope, and a machine learning module that, basing on the numerical features of the images, builds an automated classifier that predicts the class label (i.e., the lymphoma type) of a new and unrecognized biopsy image coming in the system. Particle Swarm Estimation of Distribution Algorithm (PSEDA), a recently proposed meta-heuristic technique that hybridize Particle Swarm Optimization (PSO) and Estimation of Distribution Algorithms (EDAs), has been employed in order to perform the training of the classifier. Experiments were conducted on a standard and publicly available dataset of lymp-nodes tissue biopsy images, and they show that our approach results in a good classification accuracy with respect to other state-of-the-art and evolutionary classification schemes.