Analýza medicínských snímků

Abstract

This master thesis focuses on semantic segmentation of neonatal brain images obtained through magnetic resonance imaging, which falls under the broader topic of medical image analysis and processing. Deep neural networks for semantic segmentation were selected, implemented, or existing implementations were utilized for available datasets. Mainly fully-convolutional neural networks were used, however experiments with Vision transformer and U-Mamba were conducted too. The chosen neural networks were briefly described, trained for four segmentation tasks, and subsequently evaluated using selected metrics. The outcome is a comparison of these neural networks in the context of the four segmentation tasks.