Advancing brain tumor detection: harnessing the Swin Transformer's power for accurate classification and performance...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate brain tumor detection is critical for diagnosis and treatment.
• Existing medical imaging classification methods have limitations.

Purpose of the Study:

• To develop and evaluate a Swin Transformer-based methodology for brain tumor image classification.
• To classify brain tumors into four categories: glioma, meningioma, non-tumor, and pituitary.

Main Methods:

• Utilized the Swin Transformer architecture for brain tumor image classification.
• Implemented a pipeline involving sophisticated preprocessing and feature extraction.
• Trained and evaluated the model on a dataset of 2,870 brain tumor images.

Main Results:

• Achieved an outstanding accuracy of 97% in brain tumor classification.
• Outperformed conventional models like Convolutional Neural Networks (CNN), Deep Convolutional Neural Networks (DCNN), and Vision Transformer (ViT).
• Detailed performance evaluation using 21 matrices, including confusion matrix, loss, and accuracy graphs.

Conclusions:

• The Swin Transformer methodology demonstrates robust and highly accurate brain tumor classification.
• This approach shows significant potential as a pioneering model in medical image analysis.
• The study contributes to advancements in precise tumor identification and classification.