This chapter presents a method based on level sets to segment organs using computer tomography (CT) medical images. Initially, the organ boundary is manually set in one slice as an initial solution, and then the method automatically segments the organ in all other slices, sequentially. In each step of iteration it fits a Gaussian curve to the organ’s slice histogram to model the speed image in which the level sets propagate. The parameters of our method are estimated using genetic algorithms (GA) and a database of reference segmentations. The method was tested to segment the liver using 20 different exams and five different measures of performance, and the results obtained confirm the potential of the method. The cases in which the method presented a poor performance are also discussed in order to instigate further research.
TopIntroduction
In medical imaging analysis, image-guided surgery and organs visualization, segmentation is a crucial step. The segmentation process is particularly arduous in abdominal computer tomography (CT) images because different organs lie within overlapping intensity value ranges and are often near to each other anatomically. Therefore, usually it is not possible to define accurately the boundaries of organs and lesions using simple threshold based segmentation. On the other hand more complex algorithms involve more parameters that must be properly adjusted in order to generate good results for a wide set of exams.
Numerous techniques have been proposed in the literature for extraction of organs contours in abdominal CT scans. They can be roughly divided in two main groups: model driven and data driven approaches (Masutani et al, 2005).
Model driven techniques (e.g. Lamecker et al, 2004; Soler et al, 2000) use pre-defined models to segment the desired object from the available images. In this kind of technique a model describing the organ to be segmented is defined in terms of object characteristics such as position, texture and spatial relation to other objects, and the algorithm searches the images for instances that fit the given model.
Data driven techniques (e.g. Fujimoto et al, 2001; Kim et al, 2000) try to emulate the human capacity of identifying objects using some similarity information present on image data, automatically detecting and classifying objects and features in images. Many of them use known techniques such as region growing and thresholds, combined with some prior knowledge about the object being analyzed.
Level set methods (Osher & Sethian, 1998) are model driven methods that rely on partial differential equations to model deforming isosurfaces. These methods have been used successfully in medical image processing but usually require human intervention to set an initial solution and indicate explicitly when the model should stop expanding. In semi automatic level sets based methods, the user is minimally required to manually set parameters that segment correctly the desired object, which is a time consuming trial and error task.
The manual definition of parameter values on level sets methods is a complex task, because their relation with the final result is unclear and there is no guarantee that the optimal set of values will be determined by an trial and error approach. Such parameters in the implementation of traditional level sets are related to the curves mean curvature, propagation rate and advection of the curve to certain characteristics of the image. Therefore, there is a demand for automatic methods to define such parameters automatically. These methods can use important information provided by a dataset or a specialist to built or adjust a given model. In this way the process learns subjective information from the data, adding this knowledge to the model, making it more robust to deal with problems similar to the ones presented in the training dataset.
Genetic Algorithms (GA) can be used to find optimal parameters in different applications. GAs are a computational search technique to find approximate solutions to optimization problems, based in the biological evolution of species as presented by Charles Darwin (Darwin 1859).
In this chapter we propose a model driven method based on level sets to segment the liver using CT images. From an initial user-defined liver segment in one slice, the method segments the liver through all other slices, using a Gaussian fit to define the speed image where the level sets propagates. The initial solution at each slice is defined as the region previously segmented on an adjacent slice.
We propose also the use of an evolutionary approach to determine appropriate parameter values based on a set of reference segmentation results. Experiments using five exams as training set and other 15 exams for validation indicated the good performance of the proposed method. The cases with poor performance are discussed to instigate further research.