CT Reconstruction from X-ray Videos with Conditional GAN Image Translation
Masayuki Fujita, Hitoshi Iba · 2023
Advancements in medical imaging technology have led to the emergence of rotatable X-ray scanning devices, which can capture multiple X-ray images from various angles (X-ray videos) with reduced radiation exposure and subsequent burden compared to computed tomography (CT) scans. However, the potential utility of X-ray videos remain underexplored. This study introduces a novel CT reconstruction technique that combines the Filtered Back Projection method (FBP) with image-to-image translation using a conditional Generative Adversarial Network (conditional GAN). While conventional FBP from X-ray videos produces CT images marred by artifacts and noise due to perspective distortion, our results demonstrate that the Pix2Pix image translation network can effectively mitigate these artifacts, yielding CT scans of superior qualitative and quantitative value.