Attention-Based CNN for MRI Brain Tumor Classification with Enhanced Sensitivity and Interpretability

Abstract

Brain tumor is one of the most serious and life-threatening neurological condition, and getting its diagnosis as early as possible is important for effective treatment. Magnetic Resonance Imaging (MRI) is the preferred modality because it has a high spatial resolution and better contrast for soft tissue. However, interpreting MRI images manually a long time and can lead to differences between observers. This research study presents an Attention-Based Convolutional Neural Network (CNN) model for the automated classification of brain tumors. The model selectively highlights spatial and channel-wise features, thereby improving tumor localization
and classification. We compared our model to traditional machine learning models like Logistic Regression, Support Vector Machines (SVM), Random Forest, Gradient Boosting, XGBoost, and K-Nearest Neighbors (KNN), as well as deep learning architectures like Basic CNN, VGG16, and ResNet50. Experimental results show that the proposed Attention-Based CNN achieved an accuracy of 0.9900, precision of 0.9862, recall of 0.9930, F1-score of 0.9896, and an AUC-ROC of 0.9928, outperforming all other models. These findings demonstrate its potential as a reliable decision-support system for radiologists, offering high diagnostic accuracy, interpretability through attention maps, and robustness across varied MRI data.