Development of semi-automatic segmentation methods by histogram-based cluster estimation method in two-dimensional medical images

Before the medical images have been interpreted by a computer or an expert, the preparation of medical images for subsequent image processing by segmenting the images into meaningful pieces is of vital importance in the correct orientation and diagnosis phase. The fact that segmentation algorithms can achieve the correct result depends on the correct estimation of the number of clusters. In this study, it is aimed to determine the number of clusters, to estimate cluster centers and to improve a semi-automatic image segmentation algorithm. With this aim, a novel algorithm called Histogram Based Cluster Estimation(HBCE) was developed in the extend of this thesis. Based on the developed HBCE method, cluster numbers and estimated cluster center information were obtained. Segmentation was realized sending with this information as an initial values for K-means, Biogeography Based Optimization (BBO), Particle Swarm Optimization (PSO), Darwinian Particle Swarm Optimization (DPSO), Fractional Order Darwinian Particle Swarm Optimization (FODPSO). The cluster number and cluster centers were also sent to modified versions of the PSO, DPSO, and FODPSO called as respectively mPSO, mDPSO and mFODPSO. The International Skin Imaging Collaboration (ISIC) and its human segmentation results were used for testing. To make the segmentation process easier and more intuitive, an interface has been created in the MATLAB ™ environment. Rand Index, Global Consistency Error (GCE), Dice Score, Jaccard Index, Accuracy, Sensitivity, Originality values were used to test the segmentation performances. As a result of the comparative tests, that the number of clusters and the estimated cluster center values obtained from the HBCE algorithm are given to the segmentation algorithms for only one particle, was observed to increase the performance and stability of the segmentation algorithm. Moreover, the center-based mDPSO algorithm gives better results (94,677% accuracy) than the other algorithms.