Received: June 18, 2018 128 International Journal of Intelligent Engineering and Systems, Vol.11, No.6, 2018 DOI: 10.22266/ijies2018.1231.13 Medical Image Segmentation Based on Extreme Learning Machine Algorithm in Kernel Fuzzy C-Means Using Artificial Bee Colony Method Hemalatha Kivudujogappa Lingappa 1 * Hosahally Narayangowda Suresh 2 Sunil Kumar Manvi 1 1 Reva University, India 2 Bangalore Institute of Technology, India * Corresponding author’s Email: hema.sjcit@gmail.com Abstract: In image segmentation field, the Fuzzy C-Means (FCM) algorithm is a well-known algorithm for its simplicity and membership function that can control the overlapped clusters effectively with a predefined number of clusters. Despite the fact, the standard FCM algorithm is noise sensitive. To solve the issue, we proposed a new method of clustering named Kernel Fuzzy C-means (KFCM) clustering. KFCM performed well in terms of clustering however, for pattern recognition KFCM has issues. The first one is grouping the similar objects in a single partition due to non- awareness of patterns and the second one is misclassification of data due to the standard structure of the membership subspace plane. Non-awareness of patterns of KFCM is solved by an Extreme Learning Machine (ELM) algorithm and Artificial Bee colony (ABC) algorithm utilized for optimizing the structure of the membership subspace plane. Experimental results showed that the proposed KFCM algorithm performed better segmentation for pattern recognition. At last effectiveness of the proposed algorithm has been evaluated based on comparing the K Means, FCM, spatial FCM, and KFCM algorithms in terms of centroids, segmentation accuracy, and pixel error. The proposed methodology improved the segmentation accuracy up to 0.8-5.5% compared to the existing methods. Keywords: Artificial bee colony, Clustering, Extreme learning machine, Fuzzy c-means, Image segmentation, Kernel fuzzy c-means. 1. Introduction Image segmentation is crucial research field since it has a vast amount of real world application, for example, robot vision, object recognition, geographical imaging and color imaging and medical imaging, etc. Generally, the major task of the image segmentation technique is to make the partitions of an image into non-overlapping, constituent regions based on the features such as gray level, color, tone, texture etc. [1-3]. Commonly, the image segmentation methodology is processed in four different ways, such as graph partitioning method [4], [5], model based technique [6], morphological strategies and clustering based technique [7, 8]. In the last decades, most of the segmentation research methodologies are done by the cluster-based technology. Among them, the fuzzy technique is a standard method and it can preserve more image info than the other clustering methodology [9, 10]. The membership value of every pixel of an image is assigned by the FCM on the basis of closeness of the image pixel to the centroid of each cluster, as the members of the similar partition have more similarity. The cluster centroids and the corresponding membership values are updated at each iteration, which leads the FCM algorithm classifies the image by generating the partitions of similar pixels in image space. FCM algorithm is efficient for image segmentation [11]. However, FCM in getting struggle to handle the noisy image, particularly when, there is no any prior knowledge of the noise [12]. While handling the noisy image, the FCM methodology failed to distinguish the pixels, which contain noise. The reason behind this is abnormal feature data, which is the main drawback of FCM. In order to solve this kind of problem we plan to utilize