||3D Image Guided Surgery Planning
Ipsen, Niels Bruun
||Paulsen, Rasmus Reinhold (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||Surgical resection is the primary curative treatment for primary and secondary liver cancer.
The estimation of tumor size and location in relation to other tissues is often done on a basis
of 2D Computed Tomography slices. Software able to produce 3D surfaces of liver and
tumor, based on CT slices, could be clinically beneficial when planning liver surgery.
Reconstructing 3D surfaces from 2D data can be done in many different ways. A variety
of segmentation algorithms, as well as surface reconstruction algorithms exist. This
report makes use of a semi automatic segmentation algorithm based on a 2D geometric
deformable model, to obtain point clouds from CT slices. Due to the high CT interslice
distance, compared to the sampling density of the segmentation method used, the point
clouds are anisotropic. Therefore they are pre-processed, in order to obtain more isotropically
distributed data. This processing consists of downsampling and linear interpolation.
After pre-processing, the data is introduced to the Sumatra toolkit, where implicit surface
reconstruction is done, based on a prior energy function, either Difference of Laplacian
or Membrane. The optimal combination of downsampling, interpolation and prior energy
function, reviewed in a visual and “one-way” comparative fashion, was found to be factor
two downsampling, two times interpolation and the Membrane prior energy model. The
setup is used on a case study, with good results; reliable liver surface is produced and tumor
localization is clarified.
However, further tests of the used methods needs to be conducted and improvements of the
algorithm can be done, in terms of automation, adding a prior knowledge and faster performance.
A 3D geometric deformable model, sampling isotropically, with a choice of a prior
knowledge that can be selected for the tissue/organ of interest and where the data can be
pre-processed according to this selection, could provide several improvements. A high level
of automation should be applied, but with some degree of expert interaction.
||In cooperation with Rigshospitalet, MD, Linda Bardram, Surgery Division
||Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Creation date: 2010-06-02
Update date: 2010-06-02