A Novel KCGSO-CNN Model For Automated Brain Tumor Segmentation and Classification From MRI Scans

Abstract

Brain tumor detection from magnetic resonance imaging (MRI) is a critical medical image analysis task because early and accurate localization of abnormal tissue can support diagnosis, treatment planning, and clinical decision-making. Manual analysis of MRI slices is time-consuming and may vary with observer experience, while conventional segmentation techniques are often sensitive to noise, intensity non-uniformity, and complex tumor boundaries. This paper presents an automated brain tumor detection framework using convolutional neural network (CNN) classification and K-Means with Galaxy-based Search Optimization (KCGSO) segmentation. In the proposed approach, MRI images are first preprocessed using resizing, denoising, contrast-limited adaptive histogram equalization, and min-max normalization. K-means clustering provides initial cluster centers, and the galaxy-based optimization process refines these centers to improve separation between tumor and non-tumor regions. The optimized mask is refined using morphological postprocessing, and texture, shape, intensity, and statistical features are extracted from the segmented region. Finally, a CNN classifier predicts normal and tumor classes, with extension to glioma, meningioma, and pituitary tumor categorization. The proposed method combines unsupervised segmentation, metaheuristic optimization, and deep feature learning to improve detection reliability. Experimental analysis indicates that the CNN-KCGSO framework provides improved segmentation quality, classification accuracy, and computational efficiency compared with conventional clustering and CNN-only baselines.