Optimized MRI Image Enhancement and Brain Tumor Classification Using Xception
DOI:
https://doi.org/10.5281/zenodo.24271895Abstract
Early and accurate analysis of brain tumors from magnetic resonance imaging requires both spatial localization and reliable category prediction. This work presents an integrated dualmodule deep learning framework for MRI-based tumor segmentation, four-class classification, and visual explainability. The segmentation path compares G-Net Light, U-Net, and Residual U-Net, whereas the classification path evaluates MobileNetV2, ResNet50, ResNet18, DenseNet121, VGG19, ResNet152, EfficientNet, Xception, and ConvNeXt-Tiny. Classification training uses 299 × 299 images with horizontal flipping, 20-degree rotation, color jitter, and a WeightedRandomSampler; segmentation deployment uses 256 × 256 input tensors. The recorded results identify RU-Net as the strongest segmentation model with Dice ≈0.69 and IoU ≈0.60, while Xception provides the strongest classification performance with accuracy, precision, recall, and weighted F1-score of approximately 0.982. The selected models are deployed through a Flask application with SQLite authentication. For each valid MRI, the application returns the predicted class, confidence, segmentation mask, red overlay, GradCAM heatmap, and LIME explanation.
Downloads
Published
Issue
Section
License

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.













