CDT-CAD: Context-Aware Deformable Transformers for End-to-End Chest Abnormality Detection on X-Ray Images.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Healthcare
• Deep Learning

Background:

• Deep learning excels in medical imaging but faces challenges with slow convergence and high computational costs.
• Integrating context knowledge into deep networks significantly enhances diagnostic accuracy.

Purpose of the Study:

• To introduce CDT-CAD, a context-aware deformable transformer for efficient and accurate end-to-end chest abnormality detection in X-ray images.
• To address the limitations of existing deep learning models in terms of speed and computational demands.

Main Methods:

• CDT-CAD employs an iterative context-aware feature extractor with dilated context encoding and frequency pooling blocks to capture multi-scale and wavelet-domain features.
• A deformable transformer detector utilizes sampled key points for focused feature subspace analysis, accelerating convergence.

Main Results:

• CDT-CAD demonstrated superior performance on the Vinbig Chest and Chest Det 10 datasets.
• The model achieved notable improvements in AP50 and AR metrics compared to existing methods, indicating enhanced detection accuracy and efficiency.

Conclusions:

• CDT-CAD offers an effective solution for chest abnormality detection, improving upon current deep learning approaches.
• The model's context-aware design and deformable transformer architecture contribute to faster convergence and higher accuracy in practical medical imaging scenarios.