Brain tumor identification using Gaussian Mixture Model features and Decision Trees classifier

Abstract

This paper concerns a new features type of Glioblastoma (GBM) detection based on the Gaussian Mixture Model (GMM). We address the task of the new features to identify the brain tumor using the T1, T2 weighted and FLAIR MR images. An abnormal area is detected using the multithresholding segmentation with morphological operations of MR images, while discarding those that are either redundant or confusing, thereby improving the performance of the feature-based scheme to detected brain tumor. Decision Tree classifier is applied on GMM features reduced using three principal components to evaluate the performance of cancer and normal area discrimination. The discrimination between GBM and normal area including the images, was compared using three performance indicators, namely, accuracy, false alarm and missed detection, and three modes of MRI images T1, T2 and Flair were employed. The GMM features demonstrated the best performance overall. For the T1 and T2 weighted images, the accuracy performance was 100 % with 0% missed detection and 0% false alarm respectively. In FLAIR mode the accuracy decrease to 94.11 % with 2.95 % missed detection and 2.95 % false alarm. All the experimental result is promising to enhance the precocious GBM diagnosis.