الخلاصة:
Early and accurate brain tumor detection is critical for effective treatment
planning and improving patient outcomes. Traditional diagnostic methods, such
as biopsies, are invasive and can delay timely intervention. This research
proposes an automated, non-invasive approach for brain tumor classification
using Magnetic Resonance Imaging (MRI) and a deep learning model based on
transfer learning with EfficientNetB0. We curated a comprehensive dataset of
3264 MRI scans, encompassing four distinct categories: glioma, meningioma,
pituitary tumors, and healthy brain tissues. Images were acquired in various
planes (sagittal, axial, and coronal) and pre-processed to ensure consistency
and enhance model performance. Data augmentation techniques, including
rotation and resizing, were employed to increase the dataset’s diversity and
improve the model’s robustness. The EfficientNetB0 architecture, renowned for
its computational efficiency and scalability achieved through compound scaling,
serves as the foundation of our model. We leveraged transfer learning by
utilizing pre-trained weights from ImageNet and fine-tuning the model with a
custom classification head comprising a Global Average Pooling layer, a Dropout
layer for regularization, and a Dense layer with SoftMax activation for multi
class classification. Performance evaluation on a held-out test set demonstrated
a remarkable accuracy of approximately 97% in classifying tumor types.
Furthermore, we analyzed the model’s performance using precision, recall, F1
score, and a confusion matrix to provide a comprehensive assessment of its
diagnostic capabilities. This research highlights the potential of transfer learning
with EfficientNetB0 for developing accurate and efficient automated brain tumor
classification systems. Our findings suggest that this approach can contribute
significantly to improved diagnostic support, potentially reducing the need for
invasive procedures and facilitating timely treatment interventions.
