de | en

Dr. rer. nat. Florian Schaff

Photo von Dr. rer. nat. Florian Schaff.
Phone
+49 89 289-10802
+49 89 289-12844
Room
E-Mail
florian.schaff@tum.de
Links
Page in TUMonline
Group
Biomedical Physics

Courses and Dates

Offered Bachelor’s or Master’s Theses Topics

Estimation of Compton scattering in X-ray imaging using neural networks

Compton scattering is one of the two primary interactions of X-rays with matter in X-ray imaging (next to photoelectric absorption). In contrast to photoelectric absorption, Compton scattering is an inelastic scattering process during which X-ray photons are deflected and transfer some of their energy to the interaction partner, typically an electron. As a consequence, photons may still reach the detector after Compton interaction. In X-ray imaging, this leads to a smoothly varying background, which reduces contrast and is detrimental to quantitative imaging. Furthermore, the Compton scatter background is not uniform, but instead depends on the materials and their distribution within an imaged object. This makes a straight-forward analytical correction difficult, and existing tools to estimate the Compton background are limited in their accuracy and applicability.

 

The goal of this thesis is to develop methods to a) estimate the Compton scattering background from simple radiographs, and b) correct these images for it. This will be done using machine learning, in particular convolutional neural networks. The student will generate Monte-Carlo simulations based on the Geant4 platform, design and train neural networks, apply them for Compton scatter correction on clinical radiography images, and compare the results to existing approaches.

 

The project involves mostly data preparation, and computational work (85%, primarily Python, with potentially some C++ for Geant4), as well as experimental data collection (15%). The project will involve collaboration with the Radiology department at the TUM Hospital Klinikum rechts der Isar.

 

Basic experience in scientific programming, Monte-Carlo simulations, neural networks, and/or X-ray imaging are desirable.

 

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

suitable as
  • Master’s Thesis Applied and Engineering Physics
Supervisor: Franz Pfeiffer
Estimation of Compton scattering in X-ray imaging using neural networks

Compton scattering is one of the two primary interactions of X-rays with matter in X-ray imaging (next to photoelectric absorption). In contrast to photoelectric absorption, Compton scattering is an inelastic scattering process during which X-ray photons are deflected and transfer some of their energy to the interaction partner, typically an electron. As a consequence, photons may still reach the detector after Compton interaction. In X-ray imaging, this leads to a smoothly varying background, which reduces contrast and is detrimental to quantitative imaging. Furthermore, the Compton scatter background is not uniform, but instead depends on the materials and their distribution within an imaged object. This makes a straight-forward analytical correction difficult, and existing tools to estimate the Compton background are limited in their accuracy and applicability.

 

The goal of this thesis is to develop methods to a) estimate the Compton scattering background from simple radiographs, and b) correct these images for it. This will be done using machine learning, in particular convolutional neural networks. The student will generate Monte-Carlo simulations based on the Geant4 platform, design and train neural networks, apply them for Compton scatter correction on clinical radiography images, and compare the results to existing approaches.

 

The project involves mostly data preparation, and computational work (85%, primarily Python, with potentially some C++ for Geant4), as well as experimental data collection (15%). The project will involve collaboration with the Radiology department at the TUM Hospital Klinikum rechts der Isar.

 

Basic experience in scientific programming, Monte-Carlo simulations, neural networks, and/or X-ray imaging are desirable.

 

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

suitable as
  • Master’s Thesis Biomedical Engineering and Medical Physics
Supervisor: Franz Pfeiffer
Top of page