<![CDATA[The classification of gliomas based on Magnetic Resonance Imaging (MRI) still faces challenges due to image quality issues such as low contrast, noise, and intensity variations. This study aims to evaluate the impact of applying Contrast Limited Adaptive Histogram Equalization (CLAHE), transfer learning architectures, and optimization strategies on the performance of binary glioma classification using an ablation study framework with a patient-level data split to prevent data leakage. This study uses a dataset consisting of 4,000 MRI images from 1,402 unique patients. The results show that the combination of CLAHE, ResNet50, and the Adam optimizer delivers the best performance with an accuracy of 100% on the dataset and under the experimental conditions used, while Grad-CAM visualizations qualitatively demonstrate the model’s focus on relevant anatomical areas. This study contributes through an integrated evaluation framework that combines preprocessing, transfer learning, optimization strategies, and model interpretability. However, validation on a multi-institutional dataset and quantitative XAI evaluation are still needed to test generalization capabilities and strengthen the interpretation of results.]]>
Copyrights © 2026
