BTA-DETR: a multi-domain cooperative perception model for brain tumor detection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Brain tumor detection in MRI faces challenges like lesion heterogeneity, blurred boundaries, and noise.
• Accurate detection is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To develop an advanced AI model for improved brain tumor detection in MRI scans.
• To address limitations of existing models in handling complex lesion characteristics.

Main Methods:

• Proposed the Brain Tumor Aware-DETR (BTA-DETR) model, a multi-domain cooperative perception approach.
• Introduced Content-Aware Fusion Unit (CAFU) for adaptive feature re-weighting.
• Developed Learnable Temperature Attention (LTA) for precise localization of blurred boundaries.
• Integrated Frequency-Spatial Cooperative Module (FSCM) for enhanced fine-grained feature representation.

Main Results:

• BTA-DETR improved mAP50 by 4.36 percentage points compared to RT-DETR on the Roboflow dataset.
• The model reduced parameters by 0.82M and computational cost by approximately 9.1%.
• BTA-DETR outperformed the baseline on external Kaggle and BraTS datasets.

Conclusions:

• BTA-DETR demonstrates superior performance in brain tumor MRI detection.
• The model effectively addresses challenges of heterogeneity, blurred boundaries, and noise.
• BTA-DETR offers an efficient and accurate solution for clinical applications.