University of Twente Student Theses


A Data-Driven Model for Direct Segmentation in Computed Tomography Imaging

Karlashchuk, M.O. (2022) A Data-Driven Model for Direct Segmentation in Computed Tomography Imaging.

[img] PDF
Abstract:Computed tomography (CT) is a powerful tool in medical imaging that is used to study anatomical structures in the human body. To achieve this, CT measurement data is reconstructed as an image, after which semantic segmentation is performed to analyse body structures or to detect diseases. These so-called sequential methods perform reconstruction and segmentation separately and are frequently used in practice, just as joint methods which perform these tasks simultaneously. However, these methods have several disadvantages. The reconstructed images could be not suited for segmentation or might contain noise that results in incorrectly predicted segmentation masks, which might be fatal in practical applications. Next to this, reconstructing an image is not necessary to detect the location of a certain body substructure, since measurement data contains this information too. In this thesis, a direct deep learning model is evaluated that performs segmentation using extracted features from measurement data. The model learns information based on the geometry of the measurements and uses this to predict the relation between sinusoids in sinograms and coordinates in a corresponding segmentation mask. The direct segmentation model is compared to a joint model, which is adapted from the aforementioned geometry-based model. The joint model reconstructs images and predicst segmentation masks based on the same learned relation. Both models are assessed on their performance in segmentation and reconstruction.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Programme:Applied Mathematics MSc (60348)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page