With the goal of developing an accurate and fast image reconstruction algorithm for ultrasound computed tomography, we combine elements of model- and data-driven approaches and propose a learned method which addresses the disadvantages of both approaches. We design a deep neural network which accounts for a nonlinear forward operator and primal-dual algorithm by its inherent network architecture. The network is trained end-to-end, with ultrasound pressure field data as input to get directly an optimized reconstruction of speed of sound and attenuation images. The training and test data are based on a set of Optical and Acoustic Breast Phantom Database, where we use the image as ground truth and simulate pressure field data according to our forward model. Extensive experiments show that our method achieves significant improvements over state-of-the-art reconstruction methods in this field. Experiments show that the proposed algorithm improves the measures structural similarity measure (SSIM) from 0.74 to 0.95 and root mean squared error (RMSE) from 0.13 to 0.09 on average concerning the speed of sound reconstruction, while it improves the SSIM from 0.60 to 0.94 and RMSE from 0.24 to 0.10 on average in attenuation reconstruction.

为了开发用于超声计算机断层扫描的准确且快速的图像重建算法，我们结合了模型驱动和数据驱动方法的元素，并提出了一种解决两种方法缺点的学习方法。我们设计了一个深度神经网络，它通过其固有的网络架构来考虑非线性前向算子和原始对偶算法。该网络是端到端训练的，以超声压力场数据作为输入，直接获得声速和衰减图像的优化重建。训练和测试数据基于一组光学和声学乳房模型数据库，我们使用图像作为地面实况并根据我们的前向模型模拟压力场数据。大量实验表明，我们的方法比该领域最先进的重建方法取得了显着的改进。实验表明，该算法在声音重建速度方面将结构相似性度量（SSIM）从0.74提高到0.95，均方根误差（RMSE）从0.13平均提高到0.09，同时将SSIM从0.60提高到0.94，衰减重建中的 RMSE 平均从 0.24 到 0.10。