Smart Brain Tumor Classification Using Deep Learning with Channel Attention Networks
TL;DR
A deep learning technique classifies MRI scans as pituitary tumor, glioma, meningioma, or no tumor, and provides explanations for its results, showing that the proposed model has strong discriminative power and high reliability for medical image classification tasks.
Abstract
Early detection of brain tumors helps doctors accurately diagnose, plan treatments, and predict disease outcomes. The Magnetic Resonance Imaging (MRI) scans make it easier to find tumors. In this paper, a deep learning technique classifies MRI scans as pituitary tumor, glioma, meningioma, or no tumor, and provides explanations for its results. The techniques preprocess images by normalizing intensity, standardizing size, and applying data augmentation such as rotation, zooming, horizontal flipping, and intensity scaling. To address class imbalance, class weights are used to ensure the model learns from all tumor types equally. The proposed model uses the VGG16 architecture with channel attention units to highlight important structural differences and key tumor areas. The classification report on the independent test dataset also demonstrates high predictive performance across all classes. The model achieved precision, recall, and F1 scores comparable to those of the glioma class: 0.99, 0.95, and 0.97, respectively. The meningioma class achieved a precision of 0.95, a recall of 0.96, and an F1-score of 0.96. All pituitary tumor cases are correctly detected. The no-tumor class achieved 100% precision, recall, and F1 Scores, demonstrating the model's reliability. The proposed model achieved 98% classification accuracy on 1,311 test images. This shows its effectiveness for brain tumor detection and classification. The results show that the proposed model has strong discriminative power and high reliability for medical image classification tasks. The model consistently achieves a macro-average AUC of 0.98 or higher for all tumor types.