基于深度学习单投影的CT断层成像三维重建

【摘要】目的:提出一种基于深度学习的计算机断层扫描（CT）单视图断层成像三维（3D）重建方法，在减少数据采集量和降低成像剂量的情况下对不同患者进行CT图像的3D重建。方法:对不同患者的CT图像进行数据增强和模拟生成对应的数字重建放射影像（DRR），并进行数据归一化操作。利用预处理后的数据通过卷积神经网络训练出一个普适于不同患者的神经网络模型。将训练好的神经网络模型部署在测试数据集上，使用平均绝对误差（MAE）、均方根误差（RMSE）、结构相似性（SSIM）和峰值信噪比（PSNR）对重建结果进行评估。结果:定性和定量分析的结果表明，该方法可以使用不同患者的单张2D图像分别重建出质量较高的3D CT图像，MAE、RMSE、SSIM和PSNR分别为0.006、0.079、0.982、38.424 dB。此外，相比特定于单个患者的情况，该方法可以大幅度提高重建速度并节省70%的模型训练时间。结论:构建的神经网络模型可通过不同患者的2D单视图重建出相应患者的3D CT图像。因此，本研究对简化临床成像设备和放射治疗当中的图像引导具有重要作用。

Deep learning-based 3D reconstruction of CT tomography from a single projection

Abstract: Objective To propose a deep learning-based method for the 3D reconstruction of single-view computed tomography (CT) image, thereby realizing the 3D reconstruction of CT image for different patients under the condition of reducing the amount of data collection and decreasing imaging dose. Methods After the CT images of different patients were processed by data enhancement, and the corresponding digitally reconstructed radiograph (DRR) was obtained by simulation, data normalization was carried out. A neural network model universally suitable for different patients was established by training the pre-processed data using deep neural networks, and the trained neural network model was then deployed on the test dataset. Finally, the reconstruction results were evaluated using mean absolute error, root mean square error, structural similarity and peak signal noise ratio. Results The qualitative and quantitative analyses showed that the 3D CT images with high quality could be constructed by the proposed method using a single-view 2D image of different patients, with mean absolute error, root mean square error, structural similarity and peak signal noise ratio of 0.006, 0.079, 0.982 and 38.424 dB, respectively. Additionally, compared with the situation specific to a single patient, the proposed method greatly increased the reconstruction speed and save 70% of model training time. Conclusion The neural network model established in the study can reconstruct the corresponding 3D CT images using single-view 2D image of different patients. The study may play an important role in simplifying clinical imaging devices and image guidance in radiotherapy.