This paper presents a comparison study between 10 automatic and six interactive methods for liver segmentation from contrast-enhanced CT images. It is based on results from the "MICCAI 2007 Grand Challenge" workshop, where 16 teams evaluated their algorithms on a common database. A collection of 20 clinical images with reference segmentations was provided to train and tune algorithms in advance. Participants were also allowed to use additional proprietary training data for that purpose. All teams then had to apply their methods to 10 test datasets and submit the obtained results. Employed algorithms include statistical shape models, atlas registration, level-sets, graph-cuts and rule-based systems. All results were compared to reference segmentations five error measures that highlight different aspects of segmentation accuracy. All measures were combined according to a specific scoring system relating the obtained values to human expert variability. In general, interactive methods reached higher average scores than automatic approaches and featured a better consistency of segmentation quality. However, the best automatic methods (mainly based on statistical shape models with some additional free deformation) could compete well on the majority of test images. The study provides an insight in performance of different segmentation approaches under real-world conditions and highlights achievements and limitations of current image analysis techniques.

Objective We propose a simultaneous extraction method for 12 organs from non-contrast three-dimensional abdominal CT images. Materials and methods The proposed method uses an abdominal cavity standardization process and atlas guided segmentation incorporating parameter estimation with the EM algorithm to deal with the large fluctuations in the feature distribution parameters between subjects. Segmentation is then performed using multiple level sets, which minimize the energy function that considers the hierarchy and exclusiveness between organs as well as uniformity of grey values in organs. To assess the performance of the proposed method, ten non-contrast 3D CT volumes were used. Results The accuracy of the feature distribution parameter estimation was slightly improved using the proposed EM method, resulting in better performance of the segmentation process. Nine organs out of twelve were statistically improved compared with the results without the proposed parameter estimation process. The proposed multiple level sets also boosted the performance of the segmentation by 7.2 points on average compared with the atlas guided

This study verified the effectiveness of two-stage segmentation with spatial standardization of pancreas in delineating the pancreas region, patient-specific probabilistic atlas guided segmentation in reducing false negatives, and a classifier ensemble in boosting segmentation performance.

This paper proposes a unique fitter called an iris filter, which evaluates the degree of convergence of the gradient vectors within its region of support toward a pixel of interest. The degree of convergence is related to the distribution of the directions of the gradient vectors and not to their magnitudes. The convergence index of a gradient vector at a given pixel is defined as the cosine of its orientation with respect to the line connecting the pixel and the pixel of interest. The output of the iris filter is the average of the convergence indices within its region of support and lies within the range [-1,1]. The region of support of the iris filter changes so that the degree of convergence of the gradient vectors in it becomes a maximum, i.e., the size and shape of the region of support at each pixel of interest changes adaptively according to the distribution pattern of the gradient vectors around it. Theoretical analysis using models of a rounded convex region and a semi-cylindrical one is given. These show that rounded convex regions are generally enhanced, even if the contrast to their background is weak and also that elongated objects are suppressed. The filter output is 1/pi at the boundaries of rounded convex regions and semi-cylindrical ones. This value does not depend on the contrast to their background. This indicates that boundaries of rounded or slender objects, with weak contrast to their background, are enhanced by the iris filter and that the absolute value of 1/pi can be used to detect the boundaries of these objects. These theoretical characteristics are confirmed by experiments using X-ray images.

No abstract