A Review on Deep Learning Methods for Glioma Segmentation, Limitations, and Future Perspectives

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Accurate glioma segmentation from MRI is vital for clinical management.
• Tumor complexity presents significant segmentation challenges.
• Existing deep learning methods require evaluation for clinical deployment.

Purpose of the Study:

• To comprehensively review deep learning (DL) models for glioma segmentation.
• To bridge the gap between research performance and clinical applicability.
• To assess models based on accuracy, computational efficiency, and clinical suitability.

Main Methods:

• Systematic review of over 80 state-of-the-art DL models (CNN, Transformer, Hybrid) up to 2025.
• Performance evaluation on BraTS datasets benchmark.
• Suitability analysis considering robustness, efficiency, and tumor delineation.

Main Results:

• Categorization of models into CNN-based, Pure Transformer, and Hybrid architectures.
• Comparison of segmentation accuracy and computational efficiency.
• Identification of trade-offs between technical performance and clinical usability.

Conclusions:

• Deep learning shows promise for automated glioma segmentation.
• Further research is needed to optimize models for real-world clinical environments.
• Balancing performance with efficiency is key for improved diagnostic outcomes.