Brain Mr Image Segmentation Based On Fuzzy C-means Algorithm Is Studied

Medical image segmentation is an important foundation and key step in themedical image processing technology. In many of the medical image segmentationalgorithms, fuzzy clustering algorithm is considered to be one of the most suitablemethods for medical image segmentation. Though fuzzy c-means algorithm is widelyused, it has some disadvantages. Aiming to solve the problem, this dissertation mainlyfocuses on some modified fuzzy c-means algorithms and their application to MR brainimage segmentation. The main results are listed as follows.1. An error in the fuzzy local information c means (FLICM) algorithm is pointedout and the corresponding proof procedure is provided.2. Fuzzy kernel c-means (KFCM) algorithm and kernel-based fuzzy c-meansincorporating spatial constraints (SKFCM) algorithm are commonly used in brain MRimage segmentation. The former has poor quality while the latter is slow computationspeed for noise image segmentation. Motivated by these considerations, a newalgorithm, namely kernel-based enhanced fuzzy c-means (KEnFCM), is proposed. Thealgorithm uses ‘kernel method’ to modify the objective function in the enhanced fuzzyc-means algorithm, which overcomes the disadvantages of KFCM and SKFCMalgorithms. Experiments show the robustness and efficiency of the proposed algorithm.3. For overcoming the disadvantages that each sample point has equal effect tocluster in fuzzy c-means algorithm, a new method, namely weighted fuzzy c-meansincorporating two-dimension gray histogram algorithm (2DWFCM), is proposed basedon weighted fuzzy c-means (WFCM) algorithm. First calculate the one-dimension grayhistogram of smoothing image, and the two-dimension gray histogram’s diagonalinformation of the raw image and smoothing image, then combine their product as anew weighted coefficient of WFCM algorithm. Experiments show that the proposedalgorithm is a better improved FCM algorithm and more robust than fast FCM andWFCM algorithms.4. Weighted kernel fuzzy c-means incorporating two-dimension gray histogramalgorithm (2DWKFCM) is proposed. The new algorithm uses kernel induced distance instead of the Euclidean distance of the2DWFCM algorithm. Experiments performedon image segmentation show that the proposed algorithm is more adaptive and robust.The proposed three modified FCM algorithms of this dissertation improve therobustness and efficiency for noise image segmentation and provide some new idea andmethods for the development of medical image segmentation.