Comparative study of Brain Tumor Segmentation using Different Segmentation Techniques in Handling Noise

Abstract

Image segmentation has been popularly performed for researchers in the field of Biomedical, Informatics Engineering, and Statistical Computation. This study tries to compare several methods for brain tumor image segmentation, especially in handling noise. The methods are K-means Cluster, Fuzzy C-Means (FCM) Cluster, Gaussian Mixture Model (GMM), and Fernandez-Steel Skew Normal (FSSN) Mixture model. K-means and FCM are the popular Partitioning methods for clustering, while GMM is model-based clustering method. The FSSN mixture model is the new model-based clustering introduced in this study. Both GMM and FSSN are formed through a finite mixture model with Bayesian Markov Chain Monte Carlo (MCMC) optimization. The dataset is the MRI brain tumor image from General Regional Hospital (RSUD) Dr. Soetomo Surabaya. Gaussian noise and Salt pepper noise are generated to see the robustness of each method. Various evaluation parameters like Silhouette Index, Partition Coefficient Index, and Misclassification Ratio are calculated for the appropriate methods and comparative analysis is carried out. The results indicate for partitioning methods especially FCM is more robust in handling Gaussian noise, while GMM is more robust in handling Salt and pepper noise. The outstanding result shows that the FSSN mixture model could handle both Gaussian and Salt pepper noise better than other methods.