Home | deutsch  | Legals | Data Protection | Sitemap | KIT

Thesis or Internship: Multimodal breast image registration

Thesis or Internship: Multimodal breast image registration
job offer: links:
job posting number:

IPE 02-17


Institute for Data Processing and Electronics

starting date:

On appointment

closing date:

December 12, 2017

contact person:

Torsten Hopp

Multimodal breast image registration

Foto 1 IPE 02-17
Foto 2 IPE 02-17
At Karlsruhe Institute of Technology an automated method for multimodal image registration of 2D Xray mammograms, Tomosynthesis, 3D breast MRI and 3D Ultrasound Computer Tomography volumes is developed. The registration allows localizing suspicious tissue structures in all modalities at a glance. Using this approach the advantages of modalities may be combined. Our method therefore contributes to computer aided multimodal diagnosis of breast cancer. The challenge of the image registration is the considerable difference between the images of the modalities: while X-ray mammograms and Tomosynthesis volumes are acquired during compression of the breast between two plates, MRI images the breast in 3D without applying compression. USCT images breast under buoyancy conditions while in breast MRI the breast is imaged subject to gravity in prone position. Our registration method is based on simulating the deformation of the breast during the different imaging processes using a patient-specific biomechanical model.

The aim of this work is to develop an automated workflow for registering all modalities by integrating the data preprocessing, the biomechanical model, the deformation simulation as well as the postprocessing. The current methods (e.g. image segmentation) should be analyzed in terms of robustness and extended if necessary. Methods for visualization of the results and evaluation of the accuracy should be included in the workflow. Furthermore an image-based optimization of the registration parameters is intended to adapt the registration to each individual patient and determine the best registration accuracy. For this a speedup of the applied algorithms might be necessary, e.g. by using GPU accelerated libraries. The effect of the extended biomechanical model and the optimization of parameters will be evaluated with clinical datasets, which will be provided.

Task description (topics will be adapted to available time frame)
  •  Starting point of the work will be the analysis of the current state of the registration method 
  • Afterwards a concept for an automated workflow should be developed.
  •  Implementation of the workflow based on the existing software and integration of existing functionality as well as development of new image processing methods extending the current state.
  • Development of an evaluation framework, selection of appropriate evaluation and optimization metrics. Development and application of an automated optimization scheme.
  •  Evaluation with clinical datasets and optimization of the simulation parameters.

  • Programming skills in MATLAB required.
  •  Interest in medical imaging and medical image processing, in particular image segmentation and image registration.
  • Basic knowledge in (bio-)mechanical simulations / Finite Element simulations beneficial.

Torsten Hopp,  torsten hoppEjm7∂kit edu