Biomedical multiregional image segmentation with using of Local Mass method

Abstract

This thesis deals with biomedical image segmentation with the help of the local centre of mass method. Segmentation of medical images is very crucial for diagnostic and surgical purposes. Therefore, new methods are constantly being developed to enable high-quality segmentation that will manoeuvre the influence of noise and artefacts. The method used for this work is Local mass segmentation, a type of unsupervised image segmentation that segments pixels based on the centre of mass of the region using the information of the neighbouring pixels and that of the entire image. This work aims to apply this segmentation method to biomedical images generated from different imaging modalities (Computed Tomography, X-Rays, Magnetic Resonance Imaging and Skiagraphy) that made up the dataset. The dataset contains images of skeletal structures. The algorithm segments and identifies objects of interest using unique colours that will differentiate each structure. Also, the algorithm was subjected to thorough testing to determine its effectiveness, robustness and segmentation performance using simulation environments in Matlab software. The simulation included using noise generators and median filters. The result is graphically presented and analysed using descriptive statistical parameters. In addition, selected evaluation metrics were used to analyse and better understand the results.