Application Of Fusing Fuzzy Connectedness In Medical Image Segmentation

Accurate segmentation of medical images is a key step in computer-aided diagnosis and quantitative analysis of interested objects.However,fuzzy nature,intensity inhomogeneity,weak boundaries often occur in medical images due to the imaging mechanism.Now,accurate segmentation of medical images is still a challenging problem.Fuzzy connectedness is an image segmentation algorithm based on the fuzzy set theory.It defines an object by adjacency function and affinity function which reflects the continuity and similarity between adjacent pixels.As fuzzy connectedness takes both gray and spatial features into consideration,it can achieve acceptable segmentation results in the case of dealing with the blurred image.In this paper,we present two novel medical image segmentation algorithms based on fuzzy connectedness.The specific work is as follows:1.To alleviate the initial contour sensitivity of Region-Scalable Fitting level set model(RSF-level set),we propose to employ fuzzy connectedness to guide the initial contour setting of the RSF-level set.Firstly,the edge of the interested object is extracted by using the fuzzy connectedness,it is then mapped into the initial contour of the RSF-level set.The computation complexity and cost of initial contour setting is reduced effectively.Secondly,the curve evolution control parameters of the RSF-level set could be estimated automatically in terms of the analysis of morphological characteristics of the initial contour and the fuzzy connectedness parameters.Finally,the curve converges quickly driven by the regional force and boundary force.The proposed scheme is compared with other four initial contour guiding mechanisms,i.e.K-means clustering,fuzzy C-means clustering,mean-shift clustering and pulse coupled neural network(PCNN).Experimental results demonstrate that the initial contours extracted by fuzzy connectedness approach the real object boundaries better.The average Dice coefficient of the proposed method can achieve 94.97%,which is 6.44%,5.96%,2.16%and 11.72%higher than the other four guiding mechanisms respectively.2.In order to solve the problem of the traditional pulse coupled neural network with multiple parameters and single characteristics,a new idea of fusion of fuzzy connectedness and pulse coupled neural network is proposed.Firstly,the gray similarity features and distance space features(fuzzy connectedness matrix)of the image are extracted by fuzzy connectedness algorithm.Then,the fuzzy connectedness matrix is used as the stimulus input of the pulse coupled neural network,and the iterative threshold is adaptively selected according to the image edge gradient information.Finally,the optimal iterations of the pulse coupled neural network are adjusted automatically by calculating the minimum cross entropy based on fuzzy connectedness,and the final segmentation results are obtained.The experimental results show that the proposed algorithm is significantly more accurate and automated than the classical pulse coupled neural network.The average Dice coefficient is 96.98%,which is 13.43%,12.64%,15.65%,12.32%and 18.42%higher than the standard pulse coupled neural network,intersecting cortical model,spiking cortical model,simplified pulse coupled neural network and the pulse coupled neural network based on minimum cross entropy,respectively.