Focal Cortical Dysplasia (FCD) is a type of brain injury that is the main cause of Refractory Epilepsy in children, and the third largest cause in adults, present in more than 50% of childhood cases and 20% in adult cases. Patients with this inflammatory disease of epileptic seizures where the drugs are not able to help. The most appropriate clinical treatment in this case is surgery. But to be performed it is necessary if you have an accurate identification of the injury. This process is quite complex because the region of the lesion is not well defined for viewing on Magnetic Resonance Images (MRI). However, with the computational advances, have been suggested several techniques to aid in the classification of medical images. In this work, the use of Genetic Algorithm with Convolutional Neural Networks (CNN) to classify images with the presence of FCD is studied. The CNN-based system should detect and identify the location of FCD in the images. For this, it is proposed that CNN classify rectangular windows of the images in order to locate regions affected by FCD. The size of the windows and the overlap between them has a direct impact on the efficiency of the CNN-based system. In addition, hyper-parameters strongly influence CNN's performance. It is proposed here, using a genetic algorithm to: i) define the size of the windows; ii) define the overlap of the windows; iii) define some of CNN's hyper-parameters. As a result, the AG proved to be quite beneficial, optimizing the CNN in order to obtain an accuracy greater than 90% in the classification