Dilated SE-DenseNet for brain tumor MRI classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate brain tumor classification is crucial for effective treatment planning.
• Current machine learning models face challenges in achieving high diagnostic accuracy for complex medical images.
• Convolutional Neural Networks (CNNs) show promise but require architectural improvements for enhanced performance.

Purpose of the Study:

• To develop and evaluate an advanced CNN for improved MRI-based brain tumor classification.
• To enhance the DenseNet-121 architecture with dilated convolutions and attention mechanisms.
• To compare the proposed model's performance against established state-of-the-art models.

Main Methods:

• Utilized DenseNet-121 architecture as a base.
• Incorporated dilated convolutional layers for expanded receptive fields.
• Integrated Squeeze-and-Excitation (SE) networks for feature recalibration.
• Trained and validated the model on a comprehensive Kaggle brain tumor dataset.

Main Results:

• The proposed model achieved superior performance in brain tumor classification compared to ResNet-101, VGG-19, original DenseNet-121, MobileNet-V2, ViT-L/16, and Swin-B.
• Demonstrated significant improvements across key metrics including F1-score, accuracy, precision, and recall.
• Validated the effectiveness of architectural enhancements in medical image analysis.

Conclusions:

• The enhanced CNN model represents a significant advancement in MRI-based brain tumor classification.
• Architectural modifications like dilated convolutions and attention mechanisms are effective for medical image analysis.
• Machine learning holds substantial potential for improving diagnostic accuracy in various medical imaging applications.