In X-ray angiography imaging, one of the major challenges that cardiovascular specialists face is the unwanted shadows that appear from stationary and moving organs within the chest during a sequence of recorded images. These shadows reduce the quality of the images and make it harder for specialists to diagnose artery blockages. Several methods have been developed to address this issue; however, many of them require processing all frames of an input sequence, which is time-consuming. Moreover, the output of many of these methods still contains a significant amount of shadow, and the quality of the coronary arteries in the corrected images is not always satisfactory.

In this study, the Frangi filter is first used to detect the coronary arteries and remove them from the image frame containing the contrast agent. Next, the regions of the arteries are represented as a mask image, along with the frame containing the contrast agent and the closest frame without the contrast agent. These are provided as input to the deep learning model, DeepFillv2-Grayscale. This advanced network is designed to reconstruct the missing parts of the images. In this study, a specific version of this network, which operates on grayscale images, was trained specifically on X-ray angiography images.

In the final stage, the reconstructed background image is subtracted from the frame containing the contrast agent, resulting in improved images of the arteries with reduced shadows. The proposed approach can effectively reduce processing time compared to traditional methods. According to SSIM and Precision metrics, the proposed method achieves values of 0.96 and 0.97, respectively, showing superiority over well-known advanced methods such as FPCP-RPCA, OTS-RPCA, DECOLOR, and OSTD.