Brain Tumor Detection Using MRI Images: Evaluating Pre-Trained Deep Learning Models with and without Data Augmentation

In this study, various pre-trained models of deep learning are proposed for brain tumor detection using MRI images. The input data, sourced from the publicly available brain tumor MRI Dataset, undergoes preprocessing with a Multi-Layer Adaptive Guided Side Window Box Filter (MLA-GSWB) to remove noise. Data augmentation techniques, including scaling, cropping, and flipping, are employed to improve model robustness by enhancing tumor regions and feature variability. Feature extraction is performed using a Multiple Discrete Orthonormal S-Transforms (MD-OST) model to capture meaningful patterns. Furthermore, deep learning models such as ResNet 50, Inception, Xception, and custom CNN have been proposed to categorize brain tumors (BT) into glioma, meningioma, pituitary, and no-tumor. Additionally, the model is tested using both augmented and non-augmented versions of the dataset to examine how data augmentation influences its generalization performance. The implementation of the proposed approach is done in Python, and its effectiveness is evaluated through various core metrics, such as Accuracy, Precision, Recall, ROC, and the Confusion Matrix.