Abstract: Sparse-view sampling can reduce the radiation hazard caused by Cone-Beam Computed Tomography (CBCT) while bringing streak artifact to the reconstructed images. Sinogram inpainting can generate projection data for missing angles and improve the quality of reconstructed image. A Residual Encoder-Decoder Generative Adversarial Network (RED-GAN) was proposed for sinogram inpainting, replacing the U-net generator in Pix2pixGAN with Residual Encoder-Decoder (RED) module; In addition, a conditional discriminator was constructed to improve the performance of generator. The sinogram inpainting performance of RED-GAN was tested under 1/2, 1/3 and 1/4 sparse-view sampling conditions, compared with linear interpolation method, RED-CNN and Pix2pixGAN. The result indicated that test results of RED-GAN under 1/2, 1/3 and 1/4 sparse-view sampling conditions are better than the comparison method. Under the 1/4 sparse-view sampling condition the network has the most obvious effect. In the sinogram domain, Root Mean Squared Error (RMSE) decreased by 7.2%, Peak Signal to Noise Ratio (PSNR) increased by 1.5% and Structural Similarity (SSIM) increased by 1.4%; In the reconstruction image domain, RMSE decreased by 4.6%, PSNR increased by 1.3% and SSIM increased by 1.0%. The RED-GAN proposed is suitable for high-quality CBCT reconstruction of incomplete projection data and has potential application value in the field of fast scanning low-dose CBCT.

Key words: cone-beam computed tomography (CBCT), sparse-view, sinogram inpainting, generative adversarial network (GAN), deep learning

摘要： 稀疏投影可有效降低锥束CT（CBCT）扫描剂量和扫描时间，但会导致重建图像中出现大量条状伪影。正弦图修复可以生成缺失角度的投影数据，提高重建图像质量。提出一种用于正弦图修复的残差编解码–生成对抗网络（RED-GAN），利用RED模块替换Pix2pixGAN中的U-net生成器，并利用PatchGAN的条件判别器对修复后正弦图和真实正弦图进行鉴别。利用真实锥束CT投影数据进行网络训练，分别在1/2、1/3、1/4稀疏采样条件下测试网络，并与线性插值法、RED-CNN和Pix2pixGAN对比。RED-GAN的正弦图修复结果均优于线性插值法、残差编解码卷积神经网络（RED-CNN）及Pix2pixGAN,并在1/4稀疏采样条件下最为明显，在正弦图域中，均方根误差（RMSE）下降7.2%，峰值信噪比（PSNR）上升1.5%，结构相似性（SSIM）上升1.4%；在重建图像域中，RMSE下降4.6%，PSNR上升1.3%，SSIM上升1.0%。结果表明，提出的RED-GAN适用于稀疏投影锥束CT重建，在快速低剂量CT扫描领域具有潜在应用价值。

关键词: 锥束CT, 稀疏投影, 正弦图修复, 生成对抗网络, 深度学习