Associative Classification of Mammograms Based on Parallel Mining of Image Blocks

Abstract

One of the main objectives of data mining as a promising multidisciplinary¯eld in computer science is to provide a classi¯cation model to be used for decision support purposes. In the medical imaging domain, mammograms classi¯cation is a di±cult diagnostic task which calls for development of automated classi¯cation systems. Associative classi¯cation, as a special case of association rules mining, has been adopted in classi¯cation problems for years. In this paper, an associative classi¯cation framework based on parallel mining of image blocks is proposed to be used for mammograms discrimination. Indeed, association rules mining is applied to a commonly used mammography image database to classify digital mammograms into three categories, namely normal, benign and malign. In order to do so,¯rst images are preprocessed and then features are extracted from non-overlapping image blocks and discretized for rule discovery. Association rules are then discovered through parallel mining of transactional databases which correspond to the image blocks, and¯nally are used within a unique decision-making scheme to predict the class of unknown samples. Finally, experiments are conducted to assess the e®ectiveness of the proposed framework. Results show that the proposed framework proved successful in terms of accuracy, precision, and recall, and suggest that the framework could be used as the core of any future associative classi¯er to support mammograms discrimination.

Métodos de apoio ao diagnóstico médico por imagens usando regras de associação e redes complexas

“…Estas regras são então empregadas na classificação. Em geral, as abordagens baseadas em regras de associação alcançam altos valores de precisão quando comparados a outros métodos de classificação (Shooshtari et al, 2012;Rodda & Mogalla, 2011;Dua et al, 2009). Outra vantagem dos classificadores associativosé que por seu modelo de aprendizado ser baseado em regras, permite um fácil entendimento e interpretação do conhecimento obtido (Rodda & Mogalla, 2011).…”