Objective. The image reconstruction of ultrasound computed tomography is computationally expensive with conventional iterative methods. The fully learned direct deep learning reconstruction is promising to speed up image reconstruction significantly. However, for direct reconstruction from measurement data, due to the lack of real labeled data, the neural network is usually trained on a simulation dataset and shows poor performance on real data because of the simulation-to-real gap.Approach. To improve the simulation-to-real generalization of neural networks, a series of strategies are developed including a Fourier-transform-integrated neural network, measurement-domain data augmentation methods, and a self-supervised-learning-based patch-wise preprocessing neural network. Our strategies are evaluated on both the simulation dataset and real measurement datasets from two different prototype machines.Main results. The experimental results show that our deep learning methods help to improve the neural networks' robustness against noise and the generalizability to real measurement data.Significance. Our methods prove that it is possible for neural networks to achieve superior performance to traditional iterative reconstruction algorithms in imaging quality and allow for real-time 2D-image reconstruction. This study helps pave the path for the application of deep learning methods to practical ultrasound tomography image reconstruction based on simulation datasets.

@en

Deep neural networks have proven to excel classical medical image reconstruction techniques. Some networks are based on fully connected (FC) layers to achieve domain transformation such as from the data acquisition domain to the image domain. However, FC layers result in huge numbers of parameters which take a lot of GPU memory. Hence, they do not scale well, and the overall performance is limited. For ultrasound computed tomography (USCT) application, we propose a memory-efficient convolutional network that reconstructs images from the frequency domain to image domain with much less parameters compared with multilayer perceptron, by using data-driven learning. Extensive experiments demonstrate that our method achieves high reconstruction quality. It improves the structural similarity measure (SSIM) from 0.73 to 0.99 when compared with state-of-the-art reconstruction methods in this field while reduces 2/3 parameters when compared with deep neural network with FC layers to reconstruct images from frequency domain to image domain.

@en

Image reconstruction of ultrasound computed tomography based on the wave equation is able to show much more structural details than simpler ray-based image reconstruction methods. However, to invert the wave-based forward model is computationally demanding. To address this problem, we develop an efficient fully learned image reconstruction method based on a convolutional neural network. The image is reconstructed via one forward propagation of the network given input sensor data, which is much faster than the reconstruction using conventional iterative optimization methods. To transform the ultrasound measured data in the sensor domain into the reconstructed image in the image domain, we apply multiple down-scaling and up-scaling convolutional units to efficiently increase the number of hidden layers with a large receptive and projective field that can cover all elements in inputs and outputs, respectively. For dataset generation, a paraxial approximation forward model is used to simulate ultrasound measurement data. The neural network is trained with a dataset derived from natural images in ImageNet and tested with a dataset derived from medical images in OA-Breast Phantom dataset. Test results show the superior efficiency of the proposed neural network to other reconstruction algorithms including popular neural networks. When compared with conventional iterative optimization algorithms, our neural network can reconstruct a 110 × 86 image more than 20 times faster on a CPU and 1000 times faster on a GPU with comparable image quality and is also more robust to noise.

@en

Whole breast ultrasound scanning systems are used to screen a women’s breast for suspicious lesions. Typically, the transducers are located at fixed positions at relatively large distances from the breast to avoid any contact with the breast. Unfortunately, these large distances give rise to large spatial domains to be imaged. These large domains hamper the applicability of imaging by inversion. To reduce the size of the spatial computational domain, we present a two-dimensional redatuming method based on Hankel decomposition of the measured field. With this method, the field measured over an arbitrary-shaped closed curve can be redatumed to a new curve enclosing a smaller spatial domain. Additional advantages of the proposed method are that it allows to account for the finite size and orientation of a transducer and that it is robust to noise. The proposed method is successfully validated using synthetic and measured data and the results show that the recorded field can be redatumed to any position in the embedding.@en

To design breast ultrasound scanning systems or to test new imaging methods, various computer models are used to simulate the acoustic wave field propagation through a breast. The computer models vary in complexity depending on the applied approximations. The objective of this paper is to investigate how the applied approximations affect the resulting wave field. In particular, we investigate the importance of taking three-dimensional (3-D) spatial variations in the compressibility, volume density of mass, and attenuation into account. In addition, we compare four 3-D solution methods: a full-wave method, a Born approximation method, a parabolic approximation method, and a ray-based method. Results show that, for frequencies below 1 MHz, the amplitude of the fields scattering off the compressibility or density contrasts are at least 24 dB higher than the amplitude of the fields scattering off the attenuation contrasts. The results also show that considering only speed of sound as a contrast is a valid approximation. In addition, it is shown that the pressure field modeled with the full-wave method is more accurate than the fields modeled using the other three methods. Finally, the accuracy of the full-wave method is location independent whereas the accuracy of the other methods strongly depends on the point of observation.

@en

Ultrasound Computer Tomography (USCT) medical imaging is a promising approach for early detection of breast cancer. At Karlsruhe Institute of Technology (KIT) a 3-D USCT system is developed. The region-of-interest (ROI) of 10 × 10 × 10 cm³ volume is surrounded by a aperture of 2014 semi-spherical positioned ultrasound transducers. Results from a first patient study reveals the requirement of a significantly increased ROI to cover bodily variations. Design considerations and simulations show a demand for circular transducers with a diameter of ca. 500 µm, increasing the opening angle of the transducers to ca. 60°. Piezofiber composite technology is predestinated to simply provide circular transducers of the required dimensions. Moreover, piezocomposites based on single PZT (lead zirconate titanate Pb[Zr_xTi_1-x]O₃) fibers enable a cost-effective and series-production alternative to currently used dice-and-fill composites. A transducer design is presented which utilizes individually arranged single piezoceramic fibers with 460 μm in diameter within piezocomposite discs. As a result, fibers are independently addressable as single transducer elements allowing for the desired transducer arrangement. The electrical performance of each piezoceramic fiber is determined proofing a strong dependence both of the coupling coefficient and the resonance frequency from the transducer thickness. In further processing, the piezocomposite discs are connected to printed circuits, integrated into a cylindrical housing, and backfilled with polyurethane. Ultrasound characteristics such as sound pressure and opening angle are evaluated quantitatively. The results show that the transducer opening angles lie in the expected range, that the desired center frequency is achieved and that the bandwidth could be preserved compared to former dice-and-fill transducers.

@en

At the Karlsruhe Institue of Technology (KIT), a 3D-Ultrasound Computer Tomography (3D-USCT) medical imaging system for early breast cancer detection is currently developed. With the next generation of 3D-USCT 2.5, the current region of interest (ROI) of 10 x 10 x 10 cm³ shall be increased to 20 x 20 x 20 cm³ to allow reliable imaging results also for bigger female breasts. Therefore, the opening angle (OA) of the future transducers should be increased to approx. 60 at 3 dB while other characteristics such as bandwidth (BW) and resonance frequency should be preserved or even improved. Based on Fraunhofer IKTS Piezofibre composites utilized for transducer production, an optimization is performed on piezoelectric sensor geometry and size, type and structure of matching and backing layer and interconnection technology of the several parts of the transducer@en

Ultrasound Computer Tomography is an exciting new technology mostly aimed at breast cancer imaging. Due to the complex interaction of ultrasound with human tissue, the large amount of raw data, and the large volumes of interest, both image acquisition and image reconstruction are challenging. Following the idea of open science, the long term goal of the USCT reference database is establishing open and easy to use data and code interfaces and stimulating the exchange of available reconstruction algorithms and raw data sets of different USCT devices. The database was established with freely available and open licensed USCT data for comparison of reconstruction algorithms, and will be maintained and updated. Additionally, the feedback about data and system architecture of the scientists working on reconstruction methods will be published to help to drive further development of the various measurement setups.@en

In the past years we have perceived within the USCT research community a demand for freely available USCT data sets. Inspired by the idea of Open Science, this collection of data sets could stimulate the collaboration and the exchange of ideas and experiences between USCT researchers. In addition, it may lead to comprehensive comparison of different reconstruction algorithms and their results. Finally, by collecting feedback from the users about data and system architecture, valuable information is gathered for further development of measurement setups. For the above reasons, we have initiated a digital portal with several reference data sets and access scripts under free licenses. To kick off this initiative, we organized a USCT data challenge event at SPIE Medical Imaging 2017.

@en