Brain Tumor Identification and Classification of MRI Images Using Deep Learning Techniques

This work presents a hybrid FAHS–ELM–SVM framework for automated brain tumor identification and segmentation from MRI images. The system integrates skull stripping, adaptive histogram equalization, and multi-domain feature extraction to enhance tumor visibility and suppress noise. Preprocessing employs Gaussian filtering, Laplacian response, entropy mapping, and wavelet decomposition to generate rich spatial–frequency descriptors. Balanced sampling ensures robust learning from both tumor and non-tumor pixels. Extracted features are normalized and passed through an Extreme Learning Machine (ELM) for nonlinear projection into a discriminative space. A posterior-optimized SVM classifier with RBF kernel then learns class boundaries while penalizing false positives. During testing, probability maps are generated and refined using morphological filtering to obtain accurate tumor masks. The system outputs tumor classification (normal/abnormal), segmentation maps, and performance metrics including accuracy, sensitivity, specificity, Dice score, and texture statistics. Overall, the proposed pipeline provides an efficient and reliable solution for brain tumor detection using enhanced feature engineering and hybrid machine learning.

Keywords: Brain Tumor Detection, MRI Segmentation, FAHS Enhancement, Extreme Learning Machine (ELM), Support Vector Machine (SVM), Skull Stripping, Wavelet Features, Adaptive Histogram Equalization, Probability Map, Medical Image Analysis.