Skip to Main content Skip to Navigation

Classification du cancer du sein par des approches basées sur les systèmes immunitaires artificiels

Abstract : Breast cancer arrives in the world in first place in terms of incidence and mortality among the different cancer localizations in women. Despite the significant progress made in recent decades to improve the management of this type of cancer, more accurate diagnostic tools are still necessary to help experts fight against this fatal disease. In this context, considerable research studies have been carried out to bring new perspectives for the improvement of the diagnosis of breast cancer, by developing Computer-Aided Diagnosis systems (CAD). Many works were directed to detecting the presence of cancerous tissues in the breast and tumor classification using tools from artificial intelligence often inspired by natural systems. In this case, Artificial Immune Systems (AIS) are a research field that bridges the fields of immunology, computer science and engineering. The main developments in artificial immune systems, have focused on three main immunological theories: clonal selection, immune networks and negative selection. We focus in this work on the use of clonal selection algorithms for classification of breast cells in Benign / Malignant. Indeed, these approaches are generally based on two main processes: the shape recognition of the antigen and selection of the specific memory cell to it. The established idea is that only memory cells capable of recognizing the antigen are selected for cloning and mutation. After introducing the principle of these algorithms we will study, through various approaches, their performances. First, we focus on improving CLONALG algorithm, which is a basic algorithm in the field of artificial clonal selection. To enhance the learning of the latter, with a better initialization and controlled diversity, three different methods are proposed appointed Median Filter Clonal ALGorithm (MF-CLONALG), Average Cells Clonal ALGorithm (AC-CLONALG) and Validity Interval Clonal Selection (VI-CS). However, although successful, these approaches require significant computing time. In this context, the second proposed approach aims at reducing the computational rates of these algorithms (and those of the AIS in general) without affecting their performance. The Local Database Categorization Artificial Immune System algorithm (LDC-AIS) uses clustering by K-means for local data categorization, and RBF neural network for learning categories, to accelerate the selection process. The last part of the thesis is dedicated to multimodal optimization. Indeed, after having presented the clonal selection algorithms as competitive tools of pattern recognition and classification, we were interested in exploring this concept, to demonstrate the benefits of cloning and mutation operators in functions optimization’s framework. In response to some drawbacks of the MLP neural network (Multi-Layer Perceptron), an optimization procedure in several stages is proposed, in which the back-propagation is assisted by cloning and mutation processes, for fast and accurate convergence of MLP. Being close to evolutionary techniques, the Multi-Layer Perceptron based Clonal Selection approach (MLP-CS) is compared to an MLP optimized by a genetic algorithm. Each of the approaches proposed in this work is tested and compared to different previous works using two different breast databases which are the Wisconsin Diagnostic Breast Cancer (WDBC), and the Digital Database for Screening Mammography (DDSM).
Complete list of metadata

Cited literature [89 references]  Display  Hide  Download
Contributor : Frédéric Davesne Connect in order to contact the contributor
Submitted on : Tuesday, April 10, 2018 - 2:13:42 PM
Last modification on : Saturday, May 1, 2021 - 3:49:55 AM


Files produced by the author(s)


  • HAL Id : tel-01762795, version 1


Rima Daoudi-Dabladji. Classification du cancer du sein par des approches basées sur les systèmes immunitaires artificiels. Traitement du signal et de l'image [eess.SP]. Université Paris-Saclay; Université d'Evry-Val-d'Essonne, 2016. Français. ⟨NNT : 2016SACLE026⟩. ⟨tel-01762795⟩



Record views


Files downloads