Open Positions

Ph.D. positions

If you are interested in contacting us and would like to send us your CV and Cover Letter, please contact directly Prof. Franz Pfeiffer  

Master projects

Research Area: Artificial Intelligence in X-ray Imaging

Patient data, including medical images, is governed by strong privacy laws. Contemporary trends in machine learning include generative AI, capable of creating synthetic data that closely mimics a given training data set. In a medical setting, synthetic CT images hold potential for creating cases for medical training, augmenting training data in machine learning projects, and enabling data sharing without compromising privacy. 
The specific goal of this project is to further develop and implement methods to generate synthetic clinical CT images.

Character of thesis work: Data processing & Machine learning (100%).

Experience in X-ray imaging, Python programming and Machine Learning are desirable.
 
For more information, please contact: Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

Advanced AI algorithms are powerful tools for reducing image artifacts in post-processing. Especially in high-resolution X-ray computed tomography (micro-CT), this technology holds great promise for minimizing acquisition times.
This project aims to implement previously explored AI processing into our routine micro-CT workflow, eventually in form of a Web-based application. Another goal is to develop task-specific refinement methods for increased efficiency and improved imaging results.

Character of thesis work: Machine learning & Data analysis (60%), Image processing (25%), Experimental (15%).

Experience in X-ray imaging, Python programming and Machine Learning are desirable.
 
For more information, please contact: Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

High-resolution X-ray computed tomography (micro-CT) is a powerful, non-destructive tool for examining three-dimensional details of an object at the microscopic level. However, high-quality, high-resolution micro-CT imaging comes at the cost of extended acquisition times, often measured in hours. To overcome this problem, various methods reduce the scan time at the expense of image quality, which is later restored by post-processing algorithms such as denoising.

This project aims to perform a comprehensive comparison to identify the most effective approach for scan time reduction in micro-CT under different conditions.

Character of thesis work: Machine learning & Data analysis (50%), Image processing (25%), Experimental (25%).

Experience in X-ray imaging, Python programming and Machine Learning are desirable.
 
For more information, please contact: Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

Here at TUM, we have developed the first human-sized dark-field CT scanner, which has the potential to ultimately improve the diagnosis of lung diseases. In order to improve image quality using advanced AI denoising algorithms, we need large amounts of training data, which are not available because dark-field CT data is a novel modality.
This project therefore aims to address this limitation by investigating potential methods for synthesizing artificial dark-field CT images, with the ultimate goal of improving the training of AI algorithms for dark-field CT.

Character of thesis work: Data processing & Machine learning (100%).

Experience in X-ray imaging, Python programming and Machine Learning are desirable.
 
For more information, please contact: Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

In medical X-ray imaging, Compton scattering manifests itself as a low-frequency background in radiographs, leading to image degradation. Accurate modeling of Compton scattering is essential for the development of post-processing algorithms to mitigate its effects.
This project will combine physics-based Monte Carlo calculations using Geant-4 with advanced AI processing techniques. 
The goal is to refine models for Compton scatter reduction and make them more applicable to routine applications. The work will primarily be performed on our clinical dark field radiography prototype at Klinikum Rechts der Isar and will include experimental validation of the simulation results.

Character of thesis work: Numerical Simulation (50%), Image processing (25%), Experimental (25%).

Experience in X-ray imaging and Python programming are desirable.
 
For more information, please contact: Henriette Bast (henriette.bast@tum.de), Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

Deep learning has shown promising results in medical diagnosis applications in recent years, specifically in computer vision tasks, such as classifying and detecting disease with CT images. However, most models still need to be fine-tuned for use in practice, for example, by incorporating existing radiological knowledge. Traditionally, radiologists analyze and adjust CT data to better visualize relevant information. e.g., adjusting pixel intensity values of a CT image for higher contrast or scanning a region of interest in the CT volume. 
This project aims to mimic and automate the clinical workflow in the deep-learning pipeline for more efficient and accurate models.

Character of thesis work: Numerical simulation/ image processing (100%).

Basic experience in X-ray computed tomography, deep learning, and/or image processing is desirable.

For more information, please contact: Tina Dorosti (tina.dorosti@tum.de), Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de)
 

Computed tomography relies on transforming data from the 2-D sinogram to the 3-D image domain. At the core of this are algorithms performing efficient forward/backward operations. Several geometric factors need to be considered, e.g. a curved detector, cone beam illumination, or spiral pitch. Research of novel AI image processing methods in CT currently is mostly done using simplistic projection geometries. For clinical adoption, translation to an accurate projection model is essential. 

The goal of this project is to implement a sophisticated projection model to create cone-beam sinogram data from volumetric data for the training of AI algorithms.

Character of thesis work: Numerical simulation / Image Processing (100%).

Basic experience in Computed tomography, X-ray imaging, and/or programming are desirable.
 
For more information, please contact: Johannes Thalhammer (Johannes.thalhammer@tum.de), Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

Grating-based X-ray interferometry (GBI) is an imaging method that uses X-ray gratings to obtain attenuation, refraction (phase contrast) and scattering (darkfield contrast) images simultaneously. Image processing is required to extract those signals from multiple raw data images. Conventionally, this is done on a pixel-by-pixel basis, which disregards any potential inter-pixel correlations.
 
The goal of this project is to incorporate AI-based regularization schemes into GBI processing to utilize the underlying morphology of imaged objects.
 
Character of thesis work: Image Processing (50%), Numerical Simulation (25%), Experimental (25%).

Basic experience in X-ray imaging, and/or Python programming are desirable.
 
For more information, please contact: Dr. Florian Schaff (florian.schaff@tum.de), or Prof. Franz Pfeiffer (franz.pfeiffer@tum.de).
 

Bachelor projects