Norbert Vígh
Supervisor(s): Vanda Benešová
Slovak Technical University
Abstract: Deep learning methods have recently found applications in several fields, including the processing of medical imaging data. We explore the application of convolutional neural networks (CNN) for automatically processing magnetic resonance imaging (MRI) scans in ceT1- and T2-weighted modalities to assist doctors with executing accurate and time-efficient tumor diagnostics.
The main challenge of the work is the multimodal registration of coronal and axial scans, which are perpendicular to each other and, therefore, cannot be registered directly. We use a generative adversarial network (GAN) architecture to convert between modalities, making it easier to register them. The resulting registered scans can be used for a wide variety of further tasks, utilizing the complementary information contained in different MRI modalities, i.e. image segmentation.
Keywords: Computer Vision, Image Processing
Full text: Year: 2024