This study investigates medical image classification employing Convolutional Neural Networks (CNNs), Support Vector Machine (SVM) and Genetic Algorithm (GA) focusing on hyperparameter optimization for the task of Brain Tumor diagnosis with Magnetic Resonance Imaging (MRI) scans. Different data categories and attributes determine the most effective CNN model for categorizing images correctly according to pre-established classification features. The aim is to employ a hybrid CNN-SVM model to facilitate the exhaustive task of multi-class classification and determine optimal hyperparameters suitable for the diagnostic task using a GA. This paper analyzes critical factors affecting CNN performance, focusing on parameters that substantially influence the model effectiveness. The considered models (CNN, CNN-SVM, CNN-SVM-GA) are compared to contribute to the knowledge of CNN applications in image analysis in the medical radiological field.