Optimization of Convolutional Neural Network ensemble classifiers by Genetic Algorithms Miguel A. Molina-Cabello 1 , Cristian Accino 1 , Ezequiel López-Rubio 1 , and Karl Thurnhofer-Hemsi 1 Department of Computer Languages and Computer Science. University of Málaga. Bulevar Louis Pasteur, 35. 29071 Málaga. Spain. {miguelangel,ezeqlr,karlkhader}@lcc.uma.es, cristian.accino@gmail.com, WWW home page: http://www.lcc.uma.es/˜ezeqlr/index-en.html Abstract. Breast cancer exhibits a high mortality rate and it is the most invasive cancer in women. An analysis from histopathological im- ages could predict this disease. In this way, computational image process- ing might support this task. In this work a proposal which employes deep learning convolutional neural networks is presented. Then, an ensemble of networks is considered in order to obtain an enhanced recognition per- formance of the system by the consensus of the networks of the ensemble. Finally, a genetic algorithm is also considered to choose the networks that belong to the ensemble. The proposal has been tested by carrying out several experiments with a set of benchmark images. Keywords: Breast cancer classification · medical image processing · convolutional neural networks. 1 Introduction Medicine fields are being enhanced by employing digital image processing. These images are obtained in medical test such as X-ray image, ultrasound image and resonance imaging, among others. According to this information, digital image processing facilitates the analysis of the medical images due to an improvement of them by emphasizing the parts where medical staff focus on. In addition, image processing can be used in order to predict a disease. In this way, a system like this kind could detect an illness by processing an image input. In fact, image processing is essential for pathology detection [8,3,12,16]. An example of the application of image processing could be found in blood sample images obtained in laboratory by microscopy. The hematocrit is the percentage occupied by red blood cells in relation to the total blood. Early detection of several diseases, like anemia, can be indicated by a decrease or growth of the hematocrit value. The image processing supports the analysis of blood images by counting the red blood cells. Several model kinds can be used to this purpose [2,7,9]