NeXtBrain: Combining local and global feature learning for brain tumor classification.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence
• Computational Pathology

Background:

• Accurate brain tumor diagnosis is critical for treatment and patient outcomes.
• Deep learning models struggle to balance accuracy, generalization, and efficiency in medical image analysis.
• Capturing both local and global tumor features computationally is challenging.

Purpose of the Study:

• To introduce NeXtBrain, a novel hybrid deep learning architecture for brain tumor classification.
• To overcome the limitations of existing models in capturing diverse tumor features and computational costs.
• To achieve high accuracy, robustness, and efficiency in brain tumor diagnosis.

Main Methods:

• Developed NeXtBrain, a hybrid architecture featuring NeXt Convolutional Blocks (NCB) and NeXt Transformer Blocks (NTB).
• NCB utilizes Multi-Head Convolutional Attention and SwiGLU-based MLP for local feature extraction.
• NTB integrates self-attention, convolutional attention, and SwiGLU MLP for global context modeling.

Main Results:

• NeXtBrain achieved 99.78% accuracy and 99.77% F1-score on the Figshare dataset.
• On the Kaggle dataset, NeXtBrain attained 99.78% accuracy and 99.81% F1-score.
• Outperformed 17 state-of-the-art models, including ViT, CNN, and hybrid approaches, with significantly fewer parameters (23.91M) and lower computational cost (10.32 GFLOPs).

Conclusions:

• NeXtBrain offers a computationally efficient and highly accurate solution for brain tumor classification.
• The hybrid architecture effectively captures both fine-grained local and long-range global tumor characteristics.
• NeXtBrain represents a significant advancement in deep learning for medical image analysis, enabling better clinical decision-making.