A Spatially Constrained Density-Weighted Clustering Method for Brain Tumor Segmentation in MRI Images

The article proposes a modified K-means clustering algorithm to improve the accuracy of identifying tumors in brain MRI images, a crucial challenge in diagnosing and treating neurological disorders. The algorithm incorporates iterative refinement with tolerance-based convergence, density-based weighting for centroid adjustment, and spatial constraints to enhance segmentation accuracy. The proposed method introduces new methods to improve the segmentation of brain MRI images. A process of repeating small improvements on cluster centroids is used, helping to achieve low intra-cluster variation and ensuring that the algorithm reaches a good calculation point without repeating unnecessary parts. An important aspect is the added density-based weighting which updates centroid positions based on the local distribution of data points. As a result, the algorithm works better with groups that are uneven in size or shape. In addition, spatial issues are covered so that clustering is influenced by where the pixels are and how bright they are, so as to hold anatomical structures. All these methods make it possible to exactly segment the different and complicated parts of the brain in MRI scans. Results from experiments on benchmark brain scans show that the proposed method gets better results compared to both traditional and advanced segmentation methods. The model shows an average increase of 6.41% in DSC, 10.52% in JI and 6.41% in F1 Score, accompanied by a 38.43% decline in MSE. These outcomes suggest that the algorithm is a promising tool for clinical use in brain MRI analysis.