To develop a highly accurate and efficient framework for the automated detection of brain tumors in MRI scans.
Key Findings:
Achieved P = 95.5%, R = 95.7%, mAP@50 = 97.9%, and mAP@50-95 = 82.6% on an internal test dataset of 5,731 MRI scans.
Reduced parameters and GFLOPs by 37.7% and 30.9% compared to baseline models.
Demonstrated strong cross-dataset generalization with mAP@50 = 96.6% and mAP@50-95 = 79.8% on an external test set without fine-tuning.
Interpretation:
CMRA-DETR shows competitive or superior performance compared to existing models, indicating its potential for practical application in clinical settings.
Limitations:
Performance may vary with different MRI scan qualities and types of brain tumors not included in the study.
Dependence on specific architectural modifications may limit adaptability to other imaging modalities.
Conclusion:
CMRA-DETR is a promising framework for AI-assisted brain tumor detection, particularly suitable for resource-limited clinical devices.
