BRAIN TUMOR SEMI AUTOMATIC SEGMENTATION ON MRI-T1 WEIGHTED IMAGE USING ACTIVE CONTOUR

• Main Author: Habbie Thias, Ahmad
• Format: Final Project
• Language: Indonesia
• Subjects: Kedokteran dan kesehatan
• Online Access: https://digilib.itb.ac.id/gdl/view/42479
• Institution: Institut Teknologi Bandung

Brain tumor is a collection of abnormal cells in brain tissue. One of the method to diagnose brain tumor is using magnetic resonance imaging (MRI) to take brain image. Usually the physician will segment brain tumor cells in brain image manually. Segmentation is process to differenciate normal and tumors cell in brain image. A large number of images is a challenge in the manual segmentation process. Those needs can be satisfied using Computer Aided Detection (CAD). In this work, we will compare and analyze the performance of snake active contour, active contour without edge, and morphological geodesic active contour segmentation algorithms, for 3049 brain MRI T1 images that have glioma, meningioma, or pituitary tumor. The performance of these three algorithms will be quantified using the Jaccard similarity index method and the Hausdorff distance. The best segmentation results were obtained by the morphological geodesic active contour method with an average accuracy of the Jaccard similarity index of 71.18% and the average Hausdorff distance of 4.04. While the snake active contour algorithm and active contour without edge only produce the average Jaccard index of 59.26% and 48.78 and the Hausdorff distance averages 4.65 and 4.52. Analysis of the effect of initial contours on changes in segmentation results was also carried out in this work. The percentage similarity at the randomly shifted initial contour is 78.83%.