Cell Segmentation Based on FOPSO Combined With Shape Information Improved Intuitionistic FCM
Abstract: Fuzzy c-means (FCM) clustering algorithms have been proved to be effective image segmentation techniques. However, FCM clustering algorithms are sensitive to noises and initialization. They cannot effectively segment cell images with inhomogeneous gray value distributions and complex touching cells. Aiming to overcome these disadvantages, this paper proposes a cell image segmentation algorithm using fractional-order velocity based particle swarm optimization (FOPSO) combined with shape information improved intuitionistic fuzzy c-means (SI-IFCM) clustering. Iterations are carried out between FOPSO and SI-IFCM to achieve final cell segmentation. Experimental results demonstrate that the proposed algorithm has advantages on cell image segmentation, with the highest recall (90.25%) and lowest false discovery rate (0.28%) compared with the stateof-the-art algorithms. Existing system: The FCM method clusters pixels in cell images to achieve segmentation. The computation is simple and the convergence is fast. FCM with soft clustering
concept is more suitable for describing the inhomogeneous gray value distributions in cell images than hard clustering methods. Moreover, FCM has nice expansibility and could be combined with many other methods. Thus, we select FCM as the basis of our proposed algorithm. However, conventional FCM is sensitive to noises and initialization. Furthermore, conventional FCM does not consider the spatial or shape information in images. Thus, conventional FCM may not be effective enough for segmenting dim or touching cells. Some improved FCM based algorithms have been proposed to overcome these shortcomings. For example, to better describe the image data with more uncertainty, an intuitionistic fuzzy cmeans (IFCM) clustering algorithm was proposed. IFCM considers both the non membership and intuitionistic degree information. The intuitionistic membership degree is the combination of the membership and intuitionistic degrees. This algorithm can be optimized for the segmentation of cell images. Some other FCM based clustering algorithms have also been proposed. Proposed system: Propose an automatic nuclei detection algorithm using generalized Laplacian of Gaussian filter (LoG). In spite of the computational efficiency, these algorithms produce disconnected contours and they are sensitive to initialization. The region accumulation algorithms use the similarity and connectivity of the pixels to segment regions in images. For example, the marker-based watershed algorithms use a marker image to find cell boundaries. The FogBank algorithm proposed by Chalfoun et al , based on the watershed algorithm, uses histogram binning to reduce noises and geodesic distance to detect the cell shapes. However, these algorithms may result in unsatisfactory cell boundaries due to the inaccurate classification of cell region or over-segmentation. The deformable model based algorithms, including active contour and level set, use the deformable contours to match the actual cell contours. These algorithms may have inaccurate boundaries because they use local optimization algorithm which may find local extremes. Advantages: Clustering algorithms can be combined with intelligent optimization algorithms to make them more flexible. Among them, the particle swarm optimization (PSO) algorithm has advantages of fast convergence and broad applicability.
The local shape information has been incorporated into the SI-IFCM and FOPSO to improve the performance on touching cells. The final FOPSO combined with SIIFCM uses alternate optimization to avoid local extremes and sensitivity to initialization. Comparisons with other algorithms show that our algorithm has advantages over state-of-the-art algorithms. Disadvantages: Aiming to overcome these disadvantages, this paper proposes a cell image segmentation algorithm using fractional-order velocity based particle swarm optimization (FOPSO) combined with shape information improved intuitionistic fuzzy c-means (SI-IFCM) clustering. Similarly, it also has poor performance in PFA, because it is still sensitive to complex and inhomogeneous gray value distributions in cell images. The FogBank algorithm has the same problem. It has the second-best results in PCD and PND, since it uses histogram binning and geodesic distance to improve the watershed algorithm. Modules: Cell image segmentation: Cell image segmentation is fundamental and important for biomedical image analysis. The main challenges are the inhomogeneous gray value distributions and touching cells in images. The inhomogeneous gray value distributions include the uneven intensities in one cell, the various patterns of different cells, and the inhomogeneous intensities in background. There are many algorithms proposed for cell image segmentation. These algorithms can be divided into following categories: intensity thres holding, edge detection, region accumulation, deformable models, graph cuts, gradient vector flow, morphology, supervised and unsupervised classification, and fuzzy c-means (FCM) clustering based algorithms. Laplacian of Gaussian filter: Propose the two-step cell splitting algorithm based on two-step binarization and clump splitting. Ruberto and Putzu propose a cell segmentation algorithm based on intuitionistic fuzzy set thresholding. It uses intuitionistic fuzzy set theory to search
the optimal thresholds. The edge detection based algorithms detect possible cell contours for segmentation. Xu et al. propose an automatic nuclei detection algorithm using generalized Laplacian of Gaussian filter (LoG). In spite of the computational efficiency, these algorithms produce disconnected contours and they are sensitive to initialization. The region accumulation algorithms use the similarity and connectivity of the pixels to segment regions in images. For example, the marker-based watershed algorithms use a marker image to find cell boundaries. The Fogbank algorithm proposed by Chalfoun et al., based on the watershed algorithm, uses histogram binning to reduce noises and geodesic distance to detect the cell shapes. However, these algorithms may result in unsatisfactory cell boundaries due to the inaccurate classification of cell region or over-segmentation. The deformable model based algorithms, including active contour and level set, use the deformable contours to match the actual cell contours. Parameters of SI-IFCM: The fuzzy weighting exponent has been proved to be effective in FCM based algorithms. So, is used in this paper. As we aim to segment only the cells and background, the number of centroids is chosen as. Experiments about the effects of on cell images are carried out and the results are shown in Figs. 9 and 10. In Fig. 9, the horizontal axis represents the value of and the vertical axis represents the average number of missing cells. In Fig. 10, the horizontal axis represents the value of and the vertical axis represents the average number of wrongly segmented cells. In Fig. 9, the curve, which is indicated by the first legend representing image set 1, decreases? And the curve, which is indicated by the second legend representing image set 2, smoothly increases. In Fig. 10, the curve, which is indicated by the second legend representing image set 1, smoothly increases, decreases, then increases again. And the curve, which is indicated by the first legend representing image set 2, increases. From the curves representing image set 2 in Fig. 9 and Fig. 10, we obtain 0.1 as the optimal value for for image set 2. From the curve indicated by the first legend in Fig. 9, the number of missing (not detected) cells changes smoothly when changing for image set 1.