Liver tumor segmentation and volume computation with user-guided 
3D active contours

Pavel Katz

Master of Science Thesis, 2006.

School of Computer Science and Engineering
The Hebrew University of Jerusalem, Israel

Tumor volume computation is a crucial task routinely performed in radiology
departments in order to assess the effectiveness of treatments.  Today,
radiologists perform this task by estimating the tumor volume using a set
of simple guidelines. To improve this estimate, the precise 3D segmentation
of the tumor is necessary. This segmentation is especially difficult for
liver tumors due to the low image contrast and presence of irrelevant
artifacts.  To date, no effective method suitable for this task has been
developed.

In this thesis we present a software system for liver tumors segmentation
and volume computation. This system is a semi-automatic segmentation
framework which combines automatic segmentation with a set of
semi-automatic tools. The automatic algorithm is based on the 3D Active
Contours method. Semi-automatic tools include initial contour placement,
image preprocessing, mesh manipulation, anatomical landmarks placement, and
manual contour guidance.  We have implemented a software prototype of this
framework and have conducted experiments with synthetic images and real CT
datasets. The relative error of our results when compared to ground-truth
segmentation is 10%-20%, depending on the tumor type. This result proves
the system to be suitable for tumor volume determination from the clinical
point of view.