Liver cancer is the fifth most commonly diagnosed cancer and the third most common cause of death
from cancer worldwide. A precise analysis of the lesions would help in the staging of the tumor and
in the evaluation of the possible applicable therapies. In this paper we present the workflow we have
developed for the semi-automatic segmentation of liver tumors in the datasets provided for the MICCAI
Liver Tumor Segmentation contest. Since we wanted to develop a system that could be as automatic
as possible and to follow the segmentation process in every single step starting from the image loading
to the lesion extraction, we decided to subdivide the workflow in two main steps: first we focus on the
segmentation of the liver and once we have extracted the organ structure we segment the lesions applying
an adaptive multi-thresholding system.

An entropy based multi-thresholding method for semi-automatic segmentation of liver tumors

A semi-automatic scheme was developed for the segmentation of 3D liver tumors from computed tomography (CT) images. First a support vector machine (SVM) classifier was trained to extract tumor region from one single 2D slice in the intermediate part of a tumor by voxel classification. Then the extracted tumor contour, after some morphological operations, was projected to its neighboring slices for automated sampling, learning and further voxel classification in neighboring slices. This propagation procedure continued till all tumor-containing slices were processed. The method was tested using 3D CT images with 10 liver tumors and a set of quantitative measures were computed, resulted in an averaged overall performance score of 72.

Semi-automatic Segmentation of 3D Liver Tumors from CT Scans Using Voxel Classification and Propagational Learning

Automatic segmentation of liver tumorous regions often fails due to high noise and large variance of tumors. In this work, a semi-automatic algorithm is proposed to segment liver tumors from computed tomography (CT) images. To cope with the variance of tumors, their intensity probability density functions (PDF) are modeled as a bag of Gaussians unlike the previous works where the tumor is modeled as a single Gaussian, and employ a three-dimensional seeded region growing (SRG) method. The bag of Gaussians are initialized at manually selected seeds and updated during growing process iteratively. There are two criteria to be ful&#64257;lled for growing: one is the Bayesian decision rule, and the other is a model matching measure. Once the growing is terminated, morphological operations are performed to re&#64257;ne the result. This method, showing promising performance, has been evaluated using ten CT scans of livers with twenty tumors provided by the organizer of the 3D Liver Tumor Segmentation Challenge 2008.

Semi-automatic Segmentation of Liver Tumors from CT Scans Using Bayesian Rule-based 3D Region Growing

This paper describes an interactive approach to the identification of coronary arteries in 3D angiography images. The approach is based on a novel multiple hypothesis tracking methodology which is complemented with a standard minimal path search, and it allows for a complete segmentation with little manual labor. When evaluated using the 3D CT angiography data supplied with the MICCAI'08 workshop 3D Segmentation in the Clinic: A Grand Challenge II, 98% of the target coronary arteries could be segmented in about 5 minutes per data set with the same spatial accuracy achieved in manual segmentations by human experts.

Coronary Centerline Extraction Using Multiple Hypothesis Tracking and Minimal Paths

An algorithm for minimally interactive coronary artery tracking is presented. Tracking ability and accuracy results are demonstrated on 16 images CTA images.
First, a region of interest is automatically selected and a denoising filter applied. Then, for each voxel the probability of belonging to a coronary vessel is estimated from a feature space and a vesselness measure is used to obtain a cost function. The vessel starting point is obtained automatically, while the end point is provided by the user. Finally, the centerline is obtained as the minimal cost path between both points.

Minimally Interactive Knowledge-based Coronary Tracking in CTA using a Minimal Cost Path

An automatic seeding method for coronary artery segmentation and skeletonization is presented. The new method includes automatic removal of the rib cage, tracing of the ascending aorta and initial planting of seeds for the coronary arteries. The automatic seeds are then passed on to a �virtual contrast injection� algorithm performing segmentation and skeletonization. In preliminary experiments, most main branches of the coronary tree were segmented and skeletonized without any user interaction.

An Automatic Seeding Method For Coronary Artery Segmentation and Skeletonization in CTA

Automatic segmentation and tracking of the coronary artery tree from Cardiac Multislice-CT images is an important goal to improve the diagnosis and treatment of coronary artery disease. This paper presents a semi-automatic algorithm (one input point per vessel) based on morphological grayscale local reconstructions in 3D images devoted to the extraction of the coronary artery tree. The algorithm has been evaluated in the framework of the Coronary Artery Tracking Challenge 2008 [1], obtaining consistent results in overlapping measurements (a mean of 70% of the vessel well tracked). Poor results in accuracy measurements suggest that future work should refine the centerline extraction. The algorithm can be efficiently implemented and its general strategy can be easily extrapolated to a completely automated centerline extraction or to a user interactive vessel extraction.

Coronary Artery Tracking in 3D Cardiac CT Images Using Local Morphological Reconstruction Operators

In CT images, tumors located in a liver are generally identified by intensity difference between tumor and liver. The intensity of the tumor can be lower and or higher than that of the liver. However, the main problem of liver tumor detection from is related to low contrast between tumor and liver intensities. Tumor sometimes presents in a very small dimension and makes the detection even more difficult. In this work, we focus on contrast enhancement of CT images containing liver and tumor based on the histogram processing as a necessary preprocessing for liver tumor identification. Results show that using our proposed method, the contrast of the CT images can be enhanced and results in relatively accurate identification of tumors in the liver.

Contrast Enhancement for Liver Tumor Identification

In this paper, we present a fully automated system that detects
and segments potential liver cancer tumors from a thin slice CT
data. The system is targeted toward a tumor whose volume is
larger than $1 cm^3$, and is efficient as the average
computation time per volume in our experiment is roughly 3.5
minutes.

The system first reduces the volume size by 4x4x4 to reduce the
computation and memory requirements. It then detects candidate
locations as local minima of the intensity fields after a
variant of textit{elastic quadratic} smoothing. It then
provides a rough segmentation at each candidate by fitting a
plane at sampled points near the periphery of the concave
region in the intensity profiles. The rough segmentation is
used to estimate the range of intensity values in the tumor,
which is used to obtain a more accurate segmentation by a
method originally developed for pulmonary nodules. The result
of the second segmentation is interpolated at the resolution of
the original data.

The development of the system is a part of the 2008 3D
Segmentation in the Clinic: A Grand Challenge competition. Four
CT volumes containing 10 tumors were used for the development
of the algorithm. Additional six CT volumes containing 10
tumors were used to test the segmentation performance.

Efficient Automated Detection and Segmentation of Medium and Large Liver Tumors: CAD Approach

We present in this paper an application of minimal surfaces and
Markov random fields to the segmentation of liver tumors. The
originality of the work consists in applying these models to the
region adjacency graph of a watershed transform. We detail the
assumptions and the approximations introduced in these models by
using a region graph instead of a pixel graph. This strategy leads
to an interactive method that we use to delineate tumors in 3D CT
images. We detail our strategy to achieve relevant segmentations of
these structures and compare our results to hand made segmentations
done by experienced radiologists. This paper summarizes our
participation to the MICCAI 2008 workshop called: "3D segmentation in the clinic : A
Grand Challenge II".