Brain tumors are among the most complex disorders of the nervous system, and their accurate diagnosis requires non-invasive methods based on intelligent MRI image processing. In this study, a multi-stage framework has been developed for the precise identification and categorization of brain tumors. The key innovations of this study can be summarized as follows. First, the proposed Self-Gated U-Net architecture is employed to segment tumor regions from MRI images. Subsequently, two image representations in the HSV color space, in heatmap form, are generated, and regions of interest (ROIs) are extracted from these representations using the segmentation mask. From these regions and images, a comprehensive feature set is extracted, including texture features such as Gray-Level Co-occurrence Matrix (GLCM), Local Binary Pattern (LBP), and Histogram of Oriented Gradients (HOG), as well as deep features derived from convolutional networks and various geometric features. Next, dimensionality reduction was performed separately for each feature group using the Linear Discriminant Analysis (LDA) algorithm. After feature integration, the combined features were mapped into a lower-dimensional, structured space using a parametric t-SNE neural network, preserving the nonlinear relationships among the data. Finally, brain tumor classification was performed using a fuzzy k-nearest neighbor (k-NN) classifier, where decisions were based on calculating the membership degrees of samples and their pairwise similarities. The results of implementing and executing the proposed method, in comparison with conventional approaches, demonstrate higher accuracy, robustness, and discriminative capability in classifying different brain tumor types. This framework can be used as a supportive tool to improve the precision of radiological diagnoses.
Sharafi et al. (Wed,) studied this question.