ESR 4: Real-time Physically-based Registration of Pre- and Intra-operative Medical Images – Københavns Universitet

RAINBOW > Projects > ESR 4: Real-time Physi...

ESR 4: Real-time Physically-based Registration of Pre- and Intra-operative Medical Images

Host institution: University of Copenhagen (UCPH, DK)
Supervisor: Sune Darkner, UCPH 
Clinical expertise co-supervisor: Julien Jomier, Chief Executive Officer, Kitware (FR)


Tracking soft-tissue deformations in real-time from intra-operative images is an important task for navigation and augmented reality applications in various clinical procedures such as localization of breast tumors, prostate cancer image-guided radiotherapy. A pre-operative MRI or CT volume is acquired prior to treatment for diagnosis purpose, where 3D geometries of relevant anatomical structures can be reconstructed using traditional segmentation tools. Main objective is to create novel real-time physically-based deformable registration techniques to align these 3D geometries to intra-operative images. Traditionally this is challenging due to uncertainty about material parameters, rest configurations and contact forces. To tackle this, methods to appropriately weight image registration “forces” versus mechanical forces will be investigated. In particular uncertainty about physical parameters as well as measured signals will be taken into account to produce registration results with confidence intervals. To narrow such intervals, additional images (and possibly measurements on tissues) will be required. Due to heavy computation (registration is a constrained optimization process involving elasticity, contact computation and deformation of potentially large 3D images), parallelization techniques on the GPU will be used.

Expected Results

One expected output will be recommendations to improve the acquisition methodology and protocol by identifying the data leading to the most reliable registration. To achieve solutions in terms of computational time ITK (Kitware) a state-of-the-art open source platform and methods developed at UCPH dedicated to image processing and interactive simulation, will be coupled and enhanced.

Planned research stays

  • UCPH (12 months): Training on basic image registration, locally order less registration methods, solid simulation and statistics.
  • Kitware (10 months): Training on image processing, segmentation and registration specific to the ITK platform and collect data for test case.
  • UCPH (14 months): Develop physically-based registration methods, including model uncertainty, demonstrate methods for chosen clinical scenarios, and thesis writing

Ekstra information / Sidebar