Medical Image Segmentation Based On Wavelets And FCM

Image segmentation is a very vital step in diagnosis of clinic. Compared with common image, medical images have heterogeneity, partial volume effect and noise. Fuzzy c-means clustering (FCM) is one of the most commonly used algorithm for medical image segmentation. It is unsupervised, soft segmentation. And the calculation of it is very simple. However, when FCM is applied to medical image segmentation, there are also several obvious drawbacks. FCM algorithm is sensitive to initial value and noise points, and it does not take into consideration any spatial dependence.To solve the above problem, the paper put forward an unsupervised multiresolution image segmentation, which combines wavelet transform and improved FCM, which considers neighboring pixels. Firstly, apply wavelet transform to image, making it decomposes into different resolution images. Because low-frequency image of the coarse scale have the global information of the image and a small amount of data, meanwhile is not sensitive to noise. The traditional FCM is applied to low-resolution image to get the initial image segmentation. Then, according to the properties of intrascale clustering and interscale persistence of wavelet coefficients, attach labels to image elements from coarse to fine scale. Finally, an improved FCM based on obtained the cluster centroid in the initial segmentation and label of fine-scale, and the neighboring pixels, will be adopted for final image segmentation.The proposed algorithm was validated on MR images, abdomen medical images. Experimental results show that the algorithm has higher recognition and high efficiency and stability. The method gets good results and greatly inhibits the effects of noise, and shortens the time of segmentation, which is important in practical applications.