An automatic generalized Gaussian mixture-based approach for accurate brain tumor segmentation in magnetic resonance imaging analysis

Image segmentation is crucial in medical science for feature extraction, analysis, and interpretation, especially brain tumor segmentation, which is challenging. Prior researchers have proposed both semi-automatic and fully automatic methods for this purpose. In the present paper, we propose a new automatic approach that combines thresholding and the Generalized Gaussian Mixture Model (GGMM) with the expectation-maximization algorithm for brain tumor segmentation from Magnetic Resonance Imaging (MRI) histogram data. Our method isolates the tumor region using a thresholding technique and then employs the GGMM to cluster the tumor's regions. Performance analysis of our approach was performed by using ground truth pre-segmented images as a reference, and it has an excellent performance in tumor region detection with high values for metrics such as the Dice coefficient, sensitivity, accuracy, specificity, and precision. Our work was executed randomly on a dataset of nine distinct patient MRIs. The FLAIR MRI modality was used for thresholding, and the T1ce MRI modality was used for segmentation. The results seem promising, indicating successful tumor region detection and segmentation.