Quantitative Analysis and Objective Comparison of Clustering Algorithms for Medical Image Segmentation

The paper describes the implementation of non-hierarchical methods k-means and fuzzy c-means on nosily images from different medical modalities as computed tomography and magnetic resonance. Modern devices are created on the basis of advanced technology, both during the actual acquisition of the image and subsequently during its processing. The problem is caused by the unexpected disturbance of the image by parasitic noise, which may already occur in the electronics of the device or in dependence on the phenomena caused by the external environment. The testing was carried out on 3 datasets of medical images and the evaluation per individual images was determined based on the correlation factor and the mean quadratic error. The result is evaluation of non-hierarchical clustering techniques for the creation of mathematical models of tissue depending on the noise intensity.