MLG: a mixed local and global model for brain tumor classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Brain tumors pose significant health risks, necessitating accurate classification for effective treatment.
• Computer-aided diagnosis (CAD) systems offer a reliable approach for brain tumor differentiation.

Purpose of the Study:

• To propose a highly accurate Mixed Local and Global (MLG) model for brain tumor classification.
• To enhance brain tumor classification by effectively integrating local and global image features.

Main Methods:

• The MLG model utilizes Convolutional Neural Networks (CNNs) for local feature extraction and Transformers for global feature extraction.
• A gated attention mechanism fuses features from the REMA Block (local) and Biformer Block (global).
• The REMA Block preserves local feature expressiveness, while the Biformer Block focuses on relevant global information.

Main Results:

• The MLG model achieved 99.02% accuracy on the Chen dataset and 97.24% on the Kaggle dataset.
• Performance surpassed existing state-of-the-art models in brain tumor classification tasks.
• Experimental validation on public datasets confirmed the model's effectiveness and superiority.

Conclusions:

• The proposed MLG model demonstrates superior performance in brain tumor classification.
• Effective integration of local and global features via gated attention significantly enhances classification accuracy.
• The MLG model represents a significant advancement in computer-aided systems for brain tumor diagnosis.