Fully Automatic Liver Volumetry Using 3D Level Set Segmentation for Differentiated Liver Tissue Types in Multiple Contrast MR Datasets

Abstract

Abstract.Modern epidemiological studies analyze a high amount of magnetic resonance imaging (MRI) data, which requires fully automatic segmentation methods to assist in organ volumetry. We propose a fully automatic two-step 3D level set algorithm for liver segmentation in MRI data that delineates liver tissue on liver probability maps and uses a distance transform based segmentation refinement method to improve segmentation results. MR intensity distributions in test subjects are extracted in a training phase to obtain prior information on liver, kidney and background tissue types. Probability maps are generated by using linear discriminant analysis and Bayesian methods. The algorithm is able to differentiate between normal liver tissue and fatty liver tissue and generates probability maps for both tissues to improve the segmentation results. The algorithm is embedded in a volumetry framework and yields sufficiently good results for use in epidemiological studies.

Automatic 3D model-based method for liver segmentation in MRI based on active contours and total variation minimization

“…information about computing time. Finally, Gloger et al in [16] report tests over datasets of healthy and fatty livers. The runtime analysis was performed on a 1.8 GHz Intel core 2 Duo Processor with 3 GB of RAM, and it ranges from 11.22 ± 2.78 min for healthy livers and 15.37 ± 4.96 min for fatty livers.…”

“…Gloger et al develop a fully automatic three-step 3D segmentation approach in MRI based upon a modified region growing methodology and a further thresholding technique [15]. Other approach of same authors can be found in [16] to estimate liver volume. Therein, it is proposed a 3D level set algorithm that delineates liver tissue on liver probability.…”

Liver segmentation in MRI: A fully automatic method based on stochastic partitions

“…So, the results presented in this paper cannot be compared in a direct manner with the results of these authors but initial conclusions can be extracted from this comparison. In the About MRI methods (table 5), in the fully automatic algorithms only healthy patients were used for the method validation (or fat livers in two cases), i.e, patients with tumors were not used [20,9,35,22]. Besides that, two of them used few images for validation [35,22] and other two used only a type of coefficient for the validation procedure [20,9] so a direct comparison with our results is difficult.…”

“…The liver segmentation methods found in the state-of-the-art in MRI are based mainly on level-set methods [17,18,19,20,21,22], where the drawbacks of these algorithms (difficult training, high computational cost, or high user iteration) are noticeable. Specially, in [18], a level-set method (a fast marching algorithm) and fuzzy theory are applied in the liver segmentation task, but the computational cost of this algorithm needs to be improved (as the authors themselves recognize) and, additionally, non-uniform intensity problems are not solved.…”

“…Specially, in [18], a level-set method (a fast marching algorithm) and fuzzy theory are applied in the liver segmentation task, but the computational cost of this algorithm needs to be improved (as the authors themselves recognize) and, additionally, non-uniform intensity problems are not solved. In [19,20], level-sets and probabilistic maps are used, and a training process is required with a high user iteration for manual segmentation. In [21], another level-set method called active contour is applied in T1 MR images of the liver, and the radiologist's knowledge is required to define the region of interest.…”