<![CDATA[The automated categorization of brain cancers from MRI is essential for improving diagnostic precision. Traditional Convolutional Neural Networks (CNNs) are proficient in local feature extraction but are constrained in their ability to capture long-range spatial relationships, hence impairing performance on intricate malignancies. We propose a hybrid parallel architecture that merges a CNN with a Vision Transformer (ViT) to combine local and global feature modeling. We assessed our dual-branch model in comparison to a conventional CNN baseline using a curated dataset of 15,000 MRI images categorized into three classes: glioma, meningioma, and pituitary. The hybrid model exhibited enhanced performance, attaining 98.40% accuracy and 0.0783 loss, in contrast to the baseline's 97.40% accuracy and 0.1187 loss. The substantial decrease in misclassifications was validated by additional metrics, such as enhanced recall for the meningioma category. The integration of local and global variables produces a more precise, stable, and generalizable classification framework, demonstrating significant potential as a basis for dependable AI-driven Clinical Decision Support Systems (CDSS) in neuroradiology.]]>
Copyrights © 2026
