Skip to main navigation Skip to search Skip to main content

KI-gestützte Auflösungssteigerung poröser Materialien in der klinischen Computertomographie

  • Peter Bostelmann

    Student thesis: Master's Thesis

    Abstract

    The importance of porous materials for medical implants is increasing, as their structure can promote biological integration and support bone tissue ingrowth. In particular, structured and additively manufactured implants enable patient-specific geometries and tailored microstructures and are applied in orthopaedics and preclinical implant research. Computed tomography (CT) plays a central role in the analysis of such structures due to its non-destructive and three-dimensional imaging capabilities. Modern photon-counting CT systems provide advantages in terms of spatial resolution and image quality. However, the visualization of fine structural details remains limited by physical constraints such as voxel size, slice thickness, and image noise. Micro-computed tomography (µCT) is therefore frequently used as a high-resolution reference method. Due to its limited applicability in clinical practice, machine learning approaches, particularly super-resolution methods, are gaining increasing attention as a means of computationally transforming low-resolution CT data into higher-resolution images. The aim of this thesis is the training and validation of a super-resolution model for the reconstruction of low-resolution clinical CT data into higher-resolution images. Various datasets are used for training, including artificially generated structured materials such as additively manufactured samples as well as datasets of aluminium foams. Another focus lies on image preprocessing for generating suitable training pairs and adapting the model to the specific characteristics of the datasets used. Finally, the reconstructed images are evaluated both qualitatively and using established image quality metrics to investigate the extent to which super-resolution methods can improve the visualization of fine structural features in clinical CT data. The results demonstrate both a qualitative improvement in image representation and an increase in the contrast-to-noise ratio, highlighting the potential of the proposed approach.
    Date of Award2026
    Original languageGerman (Austria)
    Awarding Institution
    • FH OÖ
    SupervisorSascha Senck (Supervisor)

    Studyprogram

    • Applied Technologies for Medical Diagnostics

    Cite this

    '