2010 Summer Project Week HandN Cancer
- Politecnico di Milano / MIT / MGH: Marta Peroni
- MIT: Polina Golland
- MGH: Greg Sharp
The project aims at allowing plan adaptation throughout the treatment course by means of CBCT. This imaging technique is at its early stage in radio– and proton-therapy application and it is not yet considered a reliable tool for planning. We would like to use Deformable registration algorithms and automatic segmentation for:
- dose adaptation on the basis of a displacement field obtained by image co-registration, which could nevetheless lead to non consistent plans both from physics and technology point of view (i.e. are not deliverable).
- organ shape prediction both intra- and across patients
One crucial step of this approach is the rigid alignment prior to deformable registration. For CT/CBCT co-registration, this is even more difficult given the presence of immobilization devices (e.g. thermoplastic masks) and the different image quality.
We tested out both Versor and Quaternions (currently in slicer) and the plan for the week is to gain a better understanding of both in terms of:
- stopping conditions (i.e. no stopping condition for quaternions?)
- computational velocity
- improvement to reduct influence of metal artifacts and different image quality on the final output
The plan includes the implementation of possible solution that address this problem.
A side problem includes the removal of the immobilization devices + tube artifacts from CBCT images. At the moment we proceed with a combination of registration and erode/dilate morphological operations given a contour on a previous image, but we would be happy to discuss about this :) (NB: maybe smt similar to the skull stripping algorithm would help us?)
- developed monitor algorithm for registration
- gained a better understanding of the methods
- fixed affine/rigid registration
- tests on segmentation of the body contour of a CBCT with BRAINSfit are indicating we might be able to get rid of all the structures external to the patient