改进深度残差卷积神经网络的LDCT图像估计

针对低剂量计算机断层扫描（Low-Dose Computed Tomography，LDCT）重建图像出现明显条形伪影的现象，提出了一种基于残差学习的深度卷积神经网络（Deep Residual Convolutional Neural Network，DR-CNN）模型，可以从LDCT图像预测标准剂量计算机断层扫描（Normal-Dose Computed Tomography，NDCT）图像。该模型在训练阶段，将数据集中的LDCT图像和NDCT图像相减得到残差图像，将LDCT图像和残差图像分别作为输入和标签，通过深度卷积神经网络（Convolution Neural Network，CNN）学习输入和标签之间的映射关系；在测试阶段，利用此映射关系从LDCT图像预测残差图像，用LDCT图像减去残差图像得到预测的NDCT图像。实验采用50对大小为512×512的同一体模的常规剂量胸腔扫描切片和投影域添加噪声后的重建图像作为数据集，其中45对作为训练集，其他作为测试集，来验证此模型的有效性。通过与非局部降噪算法、匹配三维滤波算法和K-SVD算法等目前公认效果较好的图像去噪算法对比，所提模型预测的NDCT图像均方根误差小，且信噪比略高于其他算法处理结果。

Improved deep residual convolutional neural network for LDCT image estimation

A new algorithm, Deep Residual Convolutional Neural Network（DR-CNN）, is proposed to remove streak artifacts within the reconstructed image of Low-Dose Computed Tomography（LDCT）, which can estimate Normal-Dose Computed Tomography（NDCT） image from LDCT image. In training phase, the residual images are obtained by subtracting LDCT images, LDCT images are taken as inputs, NDCT images are taken as labels, and the mapping relationship between them can be learned by a deep Convolutional Neural Network（CNN）. In test phase, the residual image can be estimated from an LDCT image using the mapping relationship learning in training phase, and the estimated NDCT image can be obtained by subtracting the residual image from the LDCT image. Fifty pairs of normal-dose thoracic scan sections of the same phantom and reconstructed images with noises added to the projection field are used as data sets, of which 45 pairs are used as training sets and others are used as test sets to verify the validity of this algorithm. Compared with the-state-of-art methods, such as non-local means, block-matching and 3D filtering and K-SVD algorithms, the estimated NDCT image has a smaller value of the Root Mean Square Error（RMSE） and a higher Peak Signal-to-Noise Ratio（PSNR） than other algorithms.