Brain tumors are critical health conditions caused by the irregular growth of malignant or benign cells in brain tissues. These tumors can affect cognitive functions and lead to severe neurological damage if not diagnosed early. Accurate identification and classification are essential for timely medical intervention and treatment planning. Magnetic Resonance Imaging (MRI) remains the most reliable imaging modality for brain tumor diagnosis due to its non-invasive nature and clarity. This research presents a deep learning model with a customized activation function and redesigned hidden layers to enhance prediction capability. The system autonomously extracts relevant features from MRI scans, eliminating the need for manual preprocessing steps. It demonstrates exceptional performance in recognizing tumor regions and classifying them with high confidence. The model's precision and recall values indicate reliable performance across various datasets and conditions. Experimental evaluation reveals a classification accuracy of 98.6% and a precision score of 97.8%, outperforming existing models. When benchmarked against YOLOv5, Mask RCNN, AFPNet, and FCNN, our approach yielded better results in consistency and robustness. The model also achieves a notably lower cross-entropy loss, confirming its stability during training. These findings emphasize the significance of automated deep learning frameworks in medical imaging analysis. The study contributes to early detection strategies that can significantly improve patient prognosis and survival outcomes.

MNET, Convolutional Neural Network, Magnetic Resonance Imaging, Brain Tumors, MRI, Deep Learning,