Evaluation of pediatric brain tissue segmentation from MRI using a clustering method

Image segmentation in medical images such as magnetic resonances (MR) is extremely important for faster and more reliable diagnoses. Since nowadays most segmentation systems are manual, new automatic methods are being developed to achieve more consistent results. Starting from a code which uses an atlas-free BTS (afBTS) algorithm to segment the brain MR images of neonates, this study aims to improve the obtained segmentation creating new code versions. Four different clustering methods are going to be used: k-means algorithm, hierarchical clustering, region-growing algorithm and statistical parametric mapping (SPM). With these methods and other changes, twelve new versions are going to be developed. In order to assess the segmentations, the Sørensen-Dice index and the correct pixel recognition will be used. Based on the results and conclusions obtained from this project, different alternatives and future developments will be proposed to deal with the weakest points in the new code versions. Keywords:Outgoin