Morphological thresholding based Fuzzy C-Means result refinement in liver tumor segmentation

Purpose: Many computer-aided techniques are used to diagnosis unhealthy tissues like liver tumor by partitioning an image into a number of regions with similar attributes. One of the most widely used methods for image segmentation is Fuzzy C-Means (FCM). It classifies all points of an image to C clusters, by allocating a membership degree to each point. According to our observations, FCM may produce numerous false positive (FP) pixels in case of liver and tumor tissue intensities approximately are close. Our purpose in this study is improving standard FCM results specially by reducing FPs and volume error percentage.Material and methods: The proposed refinement method uses morphological operations and statistical analysis on sequential slides of Liver Computed Tomography (CT) images to improving FCM results.Results: Our technique produced low FPs in comparison with standard FCM, FCMLSM and soshi et al. method. It can achieve 0.9990 accuracy, 0.9371sensitivity (0.0239 less than soshi et al.’s sensitivity), 0.9418 precision, 0.9394 Dice and 0.7112 % volume error.Conclusions: The suggested method provides more acceptable evaluation criteria than standard FCM and very lower volume error than other mentioned methods in this work, because of low FP production.

کلیدواژه ها:

Image Segmentation ، Fuzzy C-Means ، liver tumor segmentation ، morphological operation