Classificação de imagens mamográficas por meio de redes neurais convolucionais

Abstract

With the increasing data increase and the need for classification, it is essential to develop more effective and efficient learning techniques. Learning techniques have been proposed in the literature in several fields of application. In this context, different architectures of convolutional neural networks have stood out. Therefore, the present work aims to study, develop and validate learning techniques to obtain improvements in the process of classifying bioimages, more specifically mammographic images. For that, an experimental evaluation was carried out considering: i-) performance analysis of different architectures of convolutional neural networks; ii-) increase of data through data augmentation strategies; iii-) alteration of network hyperparameters to assess their performance under different configurations; iv-) evaluation of the characteristics that best describe the data sets.