Detection and classification of brain tumor using hybrid deep learning models.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology
• Deep Learning for Diagnostics

Background:

• Accurate brain tumor classification is vital for effective treatment and patient outcomes.
• Magnetic Resonance Imaging (MRI) is a key non-invasive tool for brain tumor diagnosis.
• Deep learning, particularly Convolutional Neural Networks (CNNs), has enhanced medical image interpretation.

Purpose of the Study:

• To classify brain tumors (glioma, meningioma, pituitary) using a transfer learning approach with EfficientNets.
• To evaluate the performance of fine-tuned EfficientNet models (B0-B4) on the CE-MRI Figshare dataset.
• To analyze the impact of data augmentation and visualize model decisions using Grad-CAM.

Main Methods:

• Transfer learning-based fine-tuning of pre-trained EfficientNet models (B0-B4) initialized with ImageNet weights.
• Addition of custom layers for brain tumor classification task.
• Utilized the CE-MRI Figshare dataset and incorporated data augmentation techniques.
• Employed Grad-CAM for visualizing model attention and highlighting tumor regions.

Main Results:

• EfficientNetB2 demonstrated superior performance among the tested EfficientNet models.
• Achieved high performance metrics: 99.06% overall test accuracy, 98.73% precision, 99.13% recall, and 98.79% F1-score.
• Grad-CAM visualization effectively identified tumor locations, aiding in understanding model predictions.

Conclusions:

• Fine-tuned EfficientNetB2 offers a highly accurate and robust solution for classifying common brain tumor types from MRI.
• The transfer learning approach with EfficientNets shows significant promise for improving automated brain tumor diagnosis.
• Data augmentation and visualization techniques enhance model reliability and interpretability in medical imaging applications.