Enhancing texture detail recovery in low-dose x-ray fluoroscopic images with a multi-frame deep learning framework

• DOI: 10.1117/12.3006331
• Publication Date: 2024-02-19T17:21:53Z
• Authors: Wonjin Kim; Sun-Young Jeon; Jang-Hwan Choi
• Keywords: Texture (cosmology)

The use of low-dose x-ray fluoroscopy imaging has been found to be effective in reducing radiation exposure during prolonged procedures that may result high doses patients. However, the noise generated by protocol can degrade quality fluoroscopic images and impact clinical diagnostic accuracy. This paper proposes a novel framework for denoising algorithm recover extremely small details texture edges denoised images. While existing deep learning–based approaches have shown promising performance, they still exhibit limitations capturing detailed textures objects. To address these limitations, we introduce two-step training denoising. first network uses multi-frame inputs leverage more information from several frames, while second learns residual relationship, which enhance performance recovering miss. Our extensive experiments on clinically relevant phantoms with real demonstrate proposed method outperforms state-of-the-art methods