Softwarové prostředí pro automatickou detekci fokálních kortikálních dysplázií

Abstract

The diploma thesis „The Software Environment for the Automatic Detection of Focal Cortical Dysplasia“ is divided into a theoretical and a practical part. The theoretical part deals with the description of focal cortical dysplasia and its diagnosis, as well as its imaging using magnetic resonance imaging and an overview of SW algorithms for its detection. The practical part focuses on classification using convolutional neural networks and image segmentation. The classification is carried out using the GoogLeNet and ResNet-101 networks, and different settings of their hyperparameters is tested. Overall, the ResNet-101 network showed better results. Segmentation of focal cortical dysplasia was implemented using template matching, regional segmentation and active contours. After testing, it is found that the template matching method is unsuitable for bearing detection. Segmentation using active contours showed the best results, therefore, at the end of the work, it is used in the user interface, which is used for loading the MR image, viewing it and subsequent segmentation.