Region Growing for Medical Image Segmentation Using a Modified Multiple-seed Approach on a Multi-core CPU Computer

Abstract

Region growing is known as a simple and fast algorithm to segment an image. Many papers on medical image segmentation have reported the use of this algorithm in a variety of applications, for example, to detect cardiac disease and breast cancer and to delineate tumor volumes. One approach compares the initial seed pixels with the unassigned pixels. Another approach compares the outermost pixels with their unassigned neighbor pixels at each iteration. The first leads to consistent segmented areas but is very sensitive to noise. The second may result in inaccurate segmentations especially in cases where the pixel attributes change gradually, but it is robust to noise. In this paper we propose a method which is based on the modification of the multiple-seed approach by combining the above two approaches to obtain the advantages of each method. We consider the speed of segmentation for user convenience in segmenting the images from a DICOM file. A fast segmentation will make the application more user-friendly in displaying a series of images. This is achieved by using parallel programming on the multi-core computers that are commonly available. For hardware compatibility, an application program interface (API), openMP, is used to parallelize the program. Each outermost pixel can be expanded in a parallel manner. In our method, a global threshold is defined based on the initial seed, and the region is expanded by comparing the neighborhood pixel intensity with the global threshold, instead of using statistical region calculation. The evaluation of the performance is measured based on the time required to segment an image.