A novel hybrid transformer-based framework (H-ConvNeXt-Swin) to classify brain tumors using MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Accurate brain tumor classification from MRI is crucial for effective treatment planning.
• Existing deep learning models face challenges in capturing both local and global features in MRI scans.
• Hybrid models offer potential for improved performance by integrating diverse architectural strengths.

Purpose of the Study:

• To introduce and evaluate a novel hybrid deep learning model for brain tumor classification using MRI.
• To combine the strengths of ConvNeXt and Swin Transformer for enhanced feature representation.
• To assess the model's performance against state-of-the-art methods on a large, multi-dataset MRI collection.

Main Methods:

• A hybrid deep learning architecture integrating ConvNeXt for local features and Swin Transformer for long-range dependencies was developed.
• The model was trained and validated on a combined dataset of 7,023 MRI scans across glioma, meningioma, pituitary, and no tumor categories.
• Performance was evaluated using accuracy, precision, and F-score, with attention-based visualization for explainability.

Main Results:

• The proposed H-ConvNeXt-Swin model achieved a classification accuracy of 95.37%, outperforming existing models.
• The hybrid architecture demonstrated competitive precision and F-score values.
• Attention-based visualizations highlighted diagnostically relevant regions, providing model interpretability.

Conclusions:

• The hybrid ConvNeXt-Swin model presents a highly effective approach for brain tumor classification from MRI.
• This model offers superior performance and interpretability compared to current state-of-the-art methods.
• Future research will focus on clinical validation and extending the model for tumor segmentation and localization.