Deep Feature Fusion with a Hybrid CNN–Transformer Framework for High-Fidelity Coronary Angiography Enhancement and Fine Vascular Structure Preservation

Abstract

Coronary Artery Disease (CAD) is a major cause of mortality worldwide, and X-ray coronary angiography remains an important imaging modality for evaluating coronary artery morphology and stenosis. However, low contrast, noise, motion blur, and compression artifacts can degrade angiographic images, obscure fine vascular structures, and reduce the reliability of subsequent image interpretation and automated analysis. To address these challenges, we propose a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) with Transformer-based self-attention for high-fidelity coronary angiography enhancement. The CNN component extracts local features for noise suppression, edge refinement, and recovery of fine vessel details, while the Transformer component models long-range spatial dependencies to improve global anatomical continuity and structural coherence. The framework was trained in a supervised manner using high-quality images from the ARCADE dataset and their synthetically degraded counterparts. Quantitative evaluation demonstrated that the proposed Hybrid CNN–Transformer model achieved a PSNR of 36.33 dB and an SSIM of 0.9733, outperforming the Transformer-only model (34.43 dB, 0.9674 SSIM) by 1.90 dB in PSNR and 0.0059 in SSIM. The ablation analysis further showed that the proposed optimized hybrid configuration improved PSNR by 2.42 dB and SSIM by 0.0077 compared with the CNN-only configuration (33.91 dB, 0.9656 SSIM). The proposed model also achieved an MSE of 0.00023281 and an RMSE of 0.015258, representing the lowest reconstruction error among the evaluated methods. Qualitative assessment indicated improved vessel visibility, contrast, edge definition, thin-vessel representation, and preservation of vascular continuity. The results demonstrate that combining local convolutional feature extraction with global Transformer-based contextual modeling provides an effective approach for enhancing coronary angiographic images while preserving fine vascular structures, offering a promising preprocessing framework for subsequent computer-aided cardiovascular analysis.