3D Reconstruction of hepatic tumours from interventional magnetic resonance images
Postdoctoral project
Alexandra Branzan-Albu
Denis Laurendeau (Supervisor)
Problem: Our research work is focussed on the geometric modeling of hepatic tumours and consists in developing segmentation and 3D reconstruction techniques which are adapted for MR liver images.
Using a database consisting of sequences of 2D images corresponding to parallel and equidistant anatomical slices, a large difference is noted between the horizontal intra-slice resolution (ca. 1.56 mm) and the vertical inter-slice resolution (ca. 10 mm). This difference cannot be minimized during the acquisition process, due to the respiratory motion artefact and to technical limitations. Therefore, 3D segmentation approaches are not reliable in our case and we must consider 2D segmentation techniques. We concentrate on two particular features of abdominal MRI data. The first one is the inhomogeneous texture of the liver tumours at a certain stage of their evolution. For instance, large-sized liver tumours develop a lobular appearance. The second feature is the non-uniform sharpness of the tumour boundary which may contain sharp segments alternating with “blurred” segments. Due to infiltration into surrounding tissues, malignant liver lesions often present contours that do not reveal a clear-cut transition.
To detect the region of interest, we create an isolabel tumour contour map using a multi-threshold technique and a similarity measure for the contours. In order to extract the isolabel contour map of the tumour, we need a minimum amount of information about its location in the liver. The radiologist is asked to select one single reference pixel located inside the tumour. The tumours characterized by “blurred” contours of variable sharpness are detected with a pixel aggregation algorithm based on local texture information.
2. 3D Reconstruction
The results of the 2D segmentation of liver tumours are further used in the 3D reconstruction of the tumour. We have developed a 3D reconstruction approach using shape-based interpolation and contour-based extrapolation. While interpolation generates intermediate slices between every pair of adjacent input slices, extrapolation performs a smooth closing of the external surface of the model. Surface rendering is accomplished through the generation of a triangular mesh using a parametric representation of 2-D intermediate slice contours.
Expected results: The semi-automatic 2D segmentation method for hepatic tumours and the technique for the 3D reconstruction of geometric models have been successfully tested and validated on a large database provided by the iMRI Interventional Unit at St-François d’Assise Hospital in Québec City. At the present time, we are developing an entirely automatic method for liver and liver tumour segmentation. Research involving the synthesis of the 3D geometric model of a tumour and the corresponding biomechanical viscoelastic model is also ongoing in the laboratory.
Calendar: September 2001 - May 2003
Last modification: Sep 28 2007 2:25PM by branzan


©2002-. Computer Vision and Systems Laboratory. All rights reserved