Master's Thesis: Deep Learning for Metal Artifact Reduction in CT with Heterogeneous Knee Implants

26.08.2026, Studentische Hilfskräfte, Praktikantenstellen, Studienarbeiten

Problem Statement

Metal implants such as total knee arthroplasties, fixation plates, and screws can cause severe artifacts in CT images, reducing the visibility of surrounding anatomy and limiting their diagnostic value. Developing learning-based metal artifact reduction (MAR) methods is challenging because paired CT images with and without metal artifacts are generally not available.

This thesis will investigate a synthetic-data-driven approach for MAR in heterogeneous knee implants. The first objective is to develop a pipeline for realistically introducing implants into artifact-free knee CTs, for example through implant registration or automated implant generation. The resulting synthetic dataset will then be used to train a deep learning-based MAR model, which will subsequently be evaluated on real clinical CT data containing different types of knee implants.

What We Offer

• Access to Real Clinical Data: The thesis provides access to unique clinical datasets with high potential for impactful research and publication.
• Interdisciplinary Environment: You will work in a highly educated team, fostering collaboration between computer science and clinical experts.
• Expert Feedback: Continuous supervision and feedback from medical professionals and machine learning experts will guide the research process.

Requirements

• Strong Coding: Proficiency in Python and PyTorch, with the ability to develop and debug deep learning pipelines independently.
• Deep Learning Fundamentals: Solid understanding of convolutional neural networks and modern deep learning methods for image reconstruction, synthesis, or image-to-image translation.
• Independent Research: Ability to read, understand, and implement methods from current deep learning and medical imaging literature.
• 3D Image Processing: Interest in working with volumetric data, spatial transformations, image registration, and 3D image processing.
• Medical Imaging: Experience with CT data, DICOM/NIfTI, or libraries such as MONAI, SimpleITK, or 3D Slicer is beneficial but not required.

Goals

• Develop a Synthetic Implant Generation Pipeline: Create a pipeline to realistically introduce heterogeneous knee implants, such as knee prostheses, plates, and screws, into artifact-free CT volumes and generate corresponding CT images with simulated metal artifacts.
• Develop a Deep Learning-Based MAR Model: Use the generated paired synthetic data to train a model that reduces metal artifacts while preserving the underlying anatomical structures.
• Evaluate on Real Clinical Data: Assess how well the developed MAR approach generalizes to real CT scans containing different knee implant types, using quantitative image-quality metrics and qualitative evaluation of the reconstructed anatomy.

Application

Send an email with your CV and transcript of records to tim.mach@tum.de.

References

1. Lee, Jimin, et al. "Deep learning-based metal artifact reduction in CT for total knee arthroplasty." Scientific Reports 15.1 (2025): 39587.
2. Zhang, Yanbo, and Hengyong Yu. "Convolutional neural network based metal artifact reduction in X-ray computed tomography." IEEE Transactions on Medical Imaging 37.6 (2018): 1370–1381.
3. Lin, Wei-An, et al. "DuDoNet: Dual domain network for CT metal artifact reduction." 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2019.
4. Wellenberg, R. H. H., et al. "Metal artifact reduction techniques in musculoskeletal CT-imaging." European Journal of Radiology 107 (2018): 60–69.

Kontakt: tim.mach@tum.de