A Hybrid CNN-Transformer Framework for Robust Coronary Artery Stenosis Detection in X-ray Angiography
Abstract
Abstract Background Coronary Artery Disease (CAD), characterized by coronary artery stenosis, remains a leading global cause of morbidity and mortality. Timely detection is crucial to preventing severe outcomes like myocardial infarction. However, current diagnostic methods relying on manual interpretation of angiograms are time-consuming and subjective. This study addresses these limitations by proposing a Hybrid CNN-Transformer framework to enhance detection accuracy in noisy imaging environments. Methods We introduced the Real-Time Detection Transformer (RT-DETR), which integrates a ResNet-50 backbone with an Attention-based Intra-scale Feature Interaction (AIFI) module. Unlike traditional CNNs, this architecture captures global context to distinguish vessels from artifacts. The model was trained on 8,325 annotated angiographic images from 100 patients, utilizing advanced Mosaic augmentation. Results The proposed model achieved a state-of-the-art Mean Average Precision (mAP50) of 97.8%, surpassing the baseline EfficientDet-D3 (96.6%). Significantly, detection of small lesions (APSmall) improved to 59.2%. Due to its anchor-free design, the model reached an inference speed of 38 FPS, making it 3.5 times faster than previous methods. Conclusion By effectively bridging high accuracy with real-time performance, the proposed RT-DETR framework offers a robust and efficient solution for automated coronary stenosis detection, facilitating timely clinical decision-making.
Read the paper
Related studies
- The potential and limitations of neutrons, electrons and X-rays for atomic resolution microscopy of unstained biological molecules
- X-ray-Computed Tomography Contrast Agents
- Reactivity of Surface Species in Heterogeneous Catalysts Probed by In Situ X-ray Absorption Techniques
- Single-crystal X-ray diffraction studies on structural transformations of porous coordination polymers
- Investigation of solid/vapor interfaces using ambient pressure X-ray photoelectron spectroscopy
- Resonant X-Ray Scattering Studies of Charge Order in Cuprates
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.