Segmentation of brain tumors from MRI with k-means clustering and MLGIF

Abstract

The precise detection of brain tumors in magnetic resonance imaging (MRI) is crucial for diagnosis and therapy planning. Conversely, conventional approaches often face challenges such as intensity changes, complex tumor shapes, and susceptibility to noise. This study introduces a novel hybrid framework that integrates histogram matching, k-means clustering, and a Morphological Local and Global Intensity Fitting (MLGIF) model to tackle these issues. The first stage in histogram matching is normalizing the intensity distributions of MRI data. K-means clustering is used to provide an initial segmentation of the tumor regions. The MLGIF-based active contour model enhances the precision of tumor border segmentation while maintaining computational economy by integrating both local and global intensity inputs. The BraTS 2013 dataset was used to conduct comprehensive evaluations to determine the efficacy of the suggested framework. The Dice coefficient was 94.18%, while the Jaccard index was 89.11%. The results demonstrated that our method effectively segmented brain tumors and had promise for real-world therapeutic applications.