Malaysian Journal of Mathematical Sciences, January 2020, Vol. 14, No. 1
MRI-Based Brain Tumor Segmentation Using Gaussian and Hybrid Gaussian Mixture Model-Spatially Variant Finite Mixture Model with Expectation-Maximization Algorithm
Pravitasari, A. A., Qonita, S. F., Nur Iriawan, Irhamah, Fithriasari, K., Purnami, S. W., and Ferriastuti, W.
Corresponding Email: nur_i@statistika.its.ac.id
Received date: 10 March 2019
Accepted date: 29 December 2019
Abstract:
A brain tumor is found in the nervous system and several studies have been conducted to assist medical personnel in dealing with the problem, one of which is through detection by using image-based medical segmentation of Magnetic Resonance Imaging (MRI). This is mostly used to separate the Region of Interest (ROI) or segments considered important in the medical point of view from others (Non-ROI), especially noise. The method commonly used is the model-based clustering with a Gaussian Mixture Model (GMM). However, this is limited by the consideration of the image between the pixels independent, thereby making the segmentation results lack noise robustness. In order to minimize this, the Markov Random Field (MRF) model, which fully considers the spatial dependencies between pixels, was used. The combination was named Spatially Variant Finite Mixture Model (SVFMM), with the initial parameter generated from the GMM, making the proposed model a hybrid GMM-SVFMM. In the inference process, the maximum likelihood estimation method was employed to estimate the proposed model parameters using the Expectation-Maximization (EM) algorithm. The results from the correct classification ratio (CCR) showed that MRI-based brain image segmentation coupled with hybrid GMM-SVFMM was able to provide more accurate results in separating ROI from noise compared to GMM, with CCR of 0.9876 for hybrid GMM-SVFMM and 0.8735 for GMM.
Keywords: Expectation-Maximization, GMM, Hybrid GMM-SVFMM, Markov Random Fields, Image Segmentation
Indexing
