Search

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.

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.

Ultrasound transmission tomography promises a high potential and novel imaging method for early breast cancer diagnosis; it can quantitatively characterize tissues or materials by the attenuation and speed of sound (SoS). Reconstruction of ultrasound transmission tomography is an inverse problem that can be solved iteratively based on a paraxial approximation of the Helmholtz equation as forward model, which is highly non-linear and time-consuming. In order to address these problems and reconstruct desired images, we design a dual domain network architecture for ultrasound transmission tomography reconstruction. It can enhance the information of measurement domain and directly reconstruct from pressure field measurements without using any initialization of reconstruction and fully connected layer. We train the network on simulated ImageNet data and transfer it for ultrasound transmission tomography images to avoid overfitting when the amount of ultrasound transmission tomography images is limited. Our experimental results demonstrate that a dual domain network produces significant improvements over state-of-the-art methods. It improves the measured structural similarity measure (SSIM) from 0.54 to 0.90 and normalized root mean squared error (nRMSE) from 0.49 to 0.01 on average concerning the SoS reconstruction, and from 0.46 to 0.98 for SSIM, from 353 to 0.03 for nRMSE on average concerning the attenuation reconstruction.

Ultrasound computer tomography (USCT) is a promising modality for breast cancer diagnosis which images the reflectivity, sound speed and attenuation of tissue. Elastic properties of breast tissue, however, cannot directly be imaged although they have shown to be applicable as a discriminator between different tissue types. In this work we propose a novel approach combining USCT with the principles of strain elastography. Socalled USCT-SE makes use of imaging the breast in two deformation states, estimating the deformation field based on reconstructed images and thereby allows localizing and distinguishing soft and hard masses. We use a biomechanical model of the breast to realistically simulate both deformation states of the breast. The analysis of the strain is performed by estimating the deformation field from the deformed to the undeformed image by a non-rigid registration. In two experiments the non-rigid registration is applied to ground truth sound speed images and simulated SAFT images. Results of the strain analysis show that for both cases soft and hard lesions can be distinguished visually in the elastograms. This paper provides a first approach to obtain mechanical information based on external mechanical excitation of breast tissue in a USCT system.

For image registration of breast MRI and X-ray mammography we apply detailed biomechanical models. Synthesizing X-ray mammograms from these models is an important processing step for optimizing registration parameters and deriving images for multi-modal diagnosis. A fast computation time for creating synthetic images is essential to enable a clinically relevant application. In this paper we present a method to create synthetic X-ray attenuation images with an hardware-optimized ray tracing algorithm on recent graphics processing units’ (GPU) ray tracing (RT) cores. The ray tracing algorithm is able to calculate the attenuation of the X-rays by tracing through a triangular polygon-mesh. We use the Vulkan API, which enables access to RT cores. One frame for a triangle mesh with over 5 million triangles in the mesh and a detector resolution of \(1080\times 1080\) can be calculated and transferred to and from the GPU in about 0.76 s on NVidia RTX 2070 Super GPU. Calculation duration of an interactive application without the transfer overhead allows real time application with more than 30 frames per second (fps) even for very large polygon models. The presented method is able to calculate synthetic X-ray images in a short time and has the potential for real-time applications. Also it is the very first implementation using RT cores for this purpose. The toolbox will be available as an open source.

Conventional iterative methods for image reconstruction in ultrasound transmission tomography need to perform many iterations where at each iteration one has to compute the complex forward model of ultrasound wave propagation, and hence they are time-consuming. We use a U-net neural network to accelerate the reconstruction, by training the network to map from an initial reconstruction obtained via a few iterations of L-BFGS method to the target ground truth image. Since the computation of a forward pass of the neural network is very fast, we can expect a significant acceleration using the trained network for image reconstruction. Experiments show that our trained network can replace 40 L-BFGS iterations to generate equivalent reconstructions with slightly better quantitative quality in terms of normalized root mean square error and better visual quality due to the network's denoising effect. It can achieve up to 283× speedup compared with L-BFGS method for reconstructing small-size sound speed images with 80×80 pixels. This implies that we can expect even greater acceleration effects when applying such approach to reconstruct large-size 3D images.

A 3D ultrasound computer tomography (USCT) device with a nearly isotropic and spatially invariant 3D point spread function has been constructed at Institute for Data Processing and Electronic (IPE), Karlsruhe Institute of Technology (KIT). This device is currently applied in clinical studies for breast cancer screening. In this paper, a new method to develop an automated segmentation algorithm for USCT acquired images is proposed. The method employs distance regularized level set evolutionary (DRLSE) active contours along with surface fitting extrapolation and 3D binary mask generation for fully automatic segmentation outcome. In the first stage of the proposed algorithm, DRLSE is applied to those 3D USCT slice images which contain breast and are less affected by noise and ring artifacts named as Cat2. The DRLSE segmentation results are employed to extrapolate the rest of slice images known as Cat1. To overcome defectively segmented slice images, a 3D binary mask is generated out of USCT attenuation images. The 3D binary mask is multiplied by the DRLSE-based segmentation results to form finally segmented 3D USCT images. The method was tested on 12 clinical dataset images. According to F-measure criterion, the proposed method shows higher performance than the previously proposed semiautomatic segmentation one.

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.

To accelerate the process of 3D ultrasound computed tomography, we parallelize the most time-consuming part of a paraxial forward model on GPU, where massive complex multiplications and 2D Fourier transforms have to be performed iteratively. We test our GPU implementation on a synthesized symmetric breast phantom with different sizes. In the best case, for only one emitter position, the speedup of a desktop GPU reaches 23 times when the data transfer time is included, and 100 times when only GPU parallel computing time is considered. In the worst case, the speedup of a less powerful laptop GPU is still 2.5 times over a six-core desktop CPU, when the data transfer time is included. For the correctness of the values computed on GPU, the maximum percent deviation of L2 norm is only 0.014%.

In ultrasound transmission tomography, image reconstruction is an inverse problem which is solved iteratively based on a forward model that simulates the wave propagation of ultrasound. A commonly used forward model is paraxial approximation of the Helmholtz equation, which is time-consuming. Hence developing optimizers that minimize the number of forward solutions is crucial to achieve clinically acceptable reconstruction time, while the state-of-the-art methods in this field such as Gauss-Newton conjugate gradient (CG) and nonlinear CG are not capable of reaching this goal. To that end, we focus on Jacobian-free optimizers or accelerators in this paper, since the computation of the Jacobian is expensive. We investigate the limited memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) algorithm as a preconditioning technique due to its ability to efficiently approximate inverse Hessian without performing forward model or its adjoint. We show L-BFGS can reach a speedup of more than one order of magnitude for the noise-free case, while the method still halves the reconstruction time in presence of noise in the data. The performance drop is explained by perturbed gradients due to noise in the data. We also show when used alone as a quasi-Newton method, L-BFGS is competitive with the accelerated CG based methods regarding the number of iterations, and outperforms them regarding reconstruction time.

The purpose of this paper is to investigate Nesterov’s accelerated gradient (NAG) method for the reconstruction of speed of sound and attenuation images in ultrasound computed tomography. The inverse problem of reconstruction is tackled via minimizing the deviation between exact measurements and the predicted measurements based on a paraxial approximation of the Helmholtz equation which simulates the ultrasound wave forward propagation. To solve this optimization problem, NAG is performed and compared with other algorithms. Also, a line search method is used to compute the step size for each iteration since finding proper step sizes is crucial for the convergence of such optimization algorithms. The strong Wolfe conditions are adopted as the termination condition for line search. We have compared five algorithms, namely Gauss-Newton conjugate gradient, gradient descent, NAG, gradient descent with line search, and NAG with line search. On one hand, NAG with line search has the fastest convergence rate in respect to the number of used iterations compared to the other methods. However, due to the increased computational complexity of line search for each iteration, it requires extra computational time. On the other hand, NAG with a fixed step size for all iterations is the fastest method among all the tested methods regarding computational time.

Ultrasound transmission tomography can offer quantitative characterization of breast tissue by reconstructing its sound speed and attenuation images. The image reconstruction process is a nonlinear inverse problem which we iteratively solve using the paraxial approximation of wave equation. The problem is tackled via the Gauss-Newton method yielding a set of linear systems and iteratively solving these linear systems. In this paper, we study multigrid methods for solving these linear systems. We test three multigrid schemes including V-cycle, W-cycle, and full multigrid (FMG), with up to four resolution levels. At each grid level, we use the conjugate gradient (CG) method as a standard solver. Our interest is at first by how far these schemes allow us to reduce the computations alone. For performance evaluation, we compare the multigrid methods with fixed-grid CG method where we directly apply CG on the finest grid. Results show that all tested multigrid methods have an accelerating effect in terms of needing fewer CG iterations on the finest grid. The best-case reduction is 32% for V-cycle, 33% for W-cycle, and 27% for FMG. This means that our multigrid scheme has the potential for significantly reducing reconstruction time for large-scale 2D or 3D images, where the computation cost on the finest grid is very high.

Increasing interest in multimodal breast cancer diagnosis has led to the development of methods for MRI to X-ray mammography registration. The severe breast deformation in X-ray mammography is often tackled by biomechanical models, yet there is no common consensus in literature about the required complexity of the deformation model and the simulation strategy. We present for the first time an automated patient-specific biomechanical model based image registration of MRI to digital breast tomosynthesis (DBT). DBT provides three-dimensional information of the compressed breast and as such drives the registration by a volume similarity metric. We compare different simulation strategies and propose a patient-specific optimization of simulation and model parameters. The average three-dimensional breast overlap measured by Dice coefficient of DBT and registered MRI improves for four analyzed subjects by including the estimation of unloaded state, simulation of gravity, and a concentrated pull force that mimics manual positioning of the breast on the plates from 88.1% for a mere compression simulation to 93.1% when including all our proposed simulation steps, whereas additional parameter optimization further increased the value to 94.4%.

Increasing interest in multimodal breast cancer diagnosis has led to the development of methods for MRI to X-ray mammography registration to provide direct correlation of modalities. The severe breast deformation in X-ray mammography is often tackled by biomechanical models, which however have not yet brought the registration accuracy to a clinically applicable level. We present a novel registration approach of MRI to X-ray tomosynthesis. Tomosynthesis provides three-dimensional information of the compressed breast and as such has the ability to open new possibilities in the registration of MRI and X-ray data. By bundling the 3D information from the tomosynthesis volume with the 2D projection images acquired at different measuring angles, we provide a correlation between the registration error in 3D and 2D and evaluate different 3D- and 2D-based similarity metrics to drive the optimization of the automated patient-specific registration approach. From the preliminary study of four analysed patients we found that the projected registration error is in general larger than the 3D error in case of small registration errors in the cranio-caudal direction. Although both image shape and intensitybased 2D similarity metrics showed a clear correlation with the 2D registration error at different projection angles, metrics that relied on the combined 2D and 3D information yielded in most of the cases the minimal registration error and as such had better performance than similarity metrics that rely only on the shape similarity of volumes.

Ultrasound Computer Tomography (USCT) is a promising new imaging system for breast cancer diagnosis. An essential step before further processing is to remove the water background from the reconstructed images. In this paper we present a fully-automated image segmentation method based on three-dimensional active contours. The active contour method is extended by applying gradient vector flow and encoding the USCT aperture characteristics as additional weighting terms. A surface detection algorithm based on a ray model is developed to initialize the active contour, which is iteratively deformed to capture the breast outline in USCT reflection images. The evaluation with synthetic data showed that the method is able to cope with noisy images, and is not influenced by the position of the breast and the presence of scattering objects within the breast. The proposed method was applied to 14 in-vivo images resulting in an average surface deviation from a manual segmentation of 2.7 mm. We conclude that automated segmentation of USCT reflection images is feasible and produces results comparable to a manual segmentation. By applying the proposed method, reproducible segmentation results can be obtained without manual interaction by an expert.

Synthetic Aperture Focusing Technique (SAFT) allows fast data acquisition and optimally focused images. The computational burden for 3D imaging is large as for each voxel the delay for each acquired A-scan has to be calculated, e.g. O(N5) for N3 voxels and N2 A-scans. For 3D reconstruction of objects which are large in terms of the wavelength, e.g. ≥ (100 λ)3, the computation of one volume takes several days on a current multicore PC. If the 3D distribution of the speed of sound is applied to correct the delays, the computation time increases further. In this work a time of flight interpolation based GPU implementation (TOFI-SAFT) is presented which accelerates our previous GPU implementation of speed of sound corrected SAFT by a factor of 7 to 16 min. with only minor reduction of image quality.

Due to their different physical origin, X-ray mammography and Magnetic Resonance Imaging (MRI) provide complementary diagnostic information. However, the correlation of their images is challenging due to differences in dimensionality, patient positioning and compression state of the breast. Our automated registration takes over part of the correlation task. The registration method is based on a biomechanical finite element model, which is used to simulate mammographic compression. The deformed MRI volume can be compared directly with the corresponding mammogram. The registration accuracy is determined by a number of patient-specific parameters. We optimize these parameters--e.g. breast rotation--using image similarity measures. The method was evaluated on 79 datasets from clinical routine. The mean target registration error was 13.2mm in a fully automated setting. On basis of our results, we conclude that a completely automated registration of volume images with 2D mammograms is feasible. The registration accuracy is within the clinically relevant range and thus beneficial for multimodal diagnosis.

Computer aided diagnosis (CAD) of breast cancer is mainly focused on monomodal applications. Here we present a fully automated multimodal CAD, which uses patient-specific image registration of MRI and two-view X-ray mammography. The image registration estimates the spatial correspondence between each voxel in the MRI and each pixel in cranio-caudal and mediolateral-oblique mammograms. Thereby we can combine features from both modalities. As a proof of concept we classify fixed regions of interest (ROI) into normal and suspect tissue. We investigate the classification performance of the multimodal classification in several setups against a classification with MRI features only. The average sensitivity of detecting suspect ROIs improves by approximately 2% when combining MRI with both mammographic views compared to MRI-only detection, while the specificity stays at a constant level. We conclude that automatically combining MRI and X-ray can enhance the result of a breast CAD system.

In our first clinical study with a full 3D Ultrasound Computer Tomography (USCT) system patient data was acquired in eight minutes for one breast. In this paper the patient movement during the acquisition was analyzed quantitatively and as far as possible corrected in the resulting images. The movement was tracked in ten successive reflectivity reconstructions of full breast volumes acquired during 10 s intervals at different aperture positions, which were separated by 41 s intervals. The mean distance between initial and final position was 2.2 mm (standard deviation (STD) ± 0.9 mm, max. 4.1 mm, min. 0.8 mm) and the average sum of all moved distances was 4.9 mm (STD ± 1.9 mm, max. 8.8 mm, min. 2.7 mm). The tracked movement was corrected by summing successive images, which were transformed according to the detected movement. The contrast of these images increased and additional image content became visible.

Ultrasound Computer Tomography (USCT) is a promising new imaging method for breast cancer diagnosis. We developed a 3D USCT system and tested it in a pilot study with encouraging results: 3D USCT was able to depict two carcinomas, which were present in contrast enhanced MRI volumes serving as ground truth. To overcome severe differences in the breast shape, an image registration was applied. We analyzed the correlation between average sound speed in the breast and the breast density estimated from segmented MRIs and found a positive correlation with R=0.70. Based on the results of the pilot study we now carry out a successive clinical study with 200 patients. For this we integrated our reconstruction methods and image post-processing into a comprehensive workflow. It includes a dedicated DICOM viewer for interactive assessment of fused USCT images. A new preview mode now allows intuitive and faster patient positioning. We updated the USCT system to decrease the data acquisition time by approximately factor two and to increase the penetration depth of the breast into the USCT aperture by 1 cm . Furthermore the compute-intensive reflectivity reconstruction was considerably accelerated, now allowing a sub-millimeter volume reconstruction in approximately 16 minutes. The updates made it possible to successfully image first patients in our ongoing clinical study.

Ultrasound Computer Tomography (USCT), developed at KIT, is a promising new imaging system for breast cancer diagnosis, and was successfully tested in a pilot study. The 3D USCT II prototype consists of several hundreds of ultrasound (US) transducers on a semi-ellipsoidal aperture. Spherical waves are sequentially emitted by individual transducers and received in parallel by many transducers. Reflectivity volumes are reconstructed by synthetic aperture focusing (SAFT). However, straight forward SAFT imaging leads to blurred images due to system imperfections. We present an extension of a previously proposed approach to enhance the images. This approach includes additional a priori information and system characteristics. Now spatial phase compensation was included. The approach was evaluated with a simulation and clinical data sets. An increase in the image quality was observed and quantitatively measured by SNR and other metrics.

Computer aided detection CADe of breast cancer is mainly focused on monomodal applications. We propose an automated multimodal CADe approach, which uses patient-specific image registration of MRI and X-ray mammography to estimate the spatial correspondence of tissue structures. Then, based on the spatial correspondence, features are extracted from both MRI and X-ray mammography. As proof of principle, distinct regions of interest ROI were classified into normal and suspect tissue. We investigated the performance of different classifiers, compare our combined approach against a classification with MRI features only and evaluate the influence of the registration error. Using the multimodal information, the sensitivity for detecting suspect ROIs improved by 7i¾?% compared to MRI-only detection. The registration error influences the results: using only datasets with a registration error below $$10\,mm$$, the sensitivity for the multimodal detection increases by 10i¾?% to a maximum of 88i¾?%, while the specificity remains constant. We conclude that automatically combining MRI and X-ray can enhance the result of a CADe system.

3D Ultrasound Computer Tomography (USCT) is a new imaging method for breast cancer diagnosis. In the current state of development it is essential to correlate USCT with a known imaging modality like MRI to evaluate how different tissue types are depicted. Due to different imaging conditions, e.g. with the breast subject to buoyancy in USCT, a direct correlation is demanding. We present a 3D image registration method to reduce positioning differences and allow direct side-by-side comparison of USCT and MRI volumes. It is based on a two-step approach including a buoyancy simulation with a biomechanical model and free form deformations using cubic B-Splines for a surface refinement. Simulation parameters are optimized patient-specifically in a simulated annealing scheme. The method was evaluated with in-vivo datasets resulting in an average registration error below 5mm. Correlating tissue structures can thereby be located in the same or nearby slices in both modalities and three-dimensional non-linear deformations due to the buoyancy are reduced. Image fusion of MRI volumes and USCT sound speed volumes was performed for intuitive display. By applying the registration to data of our first in-vivo study with the KIT 3D USCT, we could correlate several tissue structures in MRI and USCT images and learn how connective tissue, carcinomas and breast implants observed in the MRI are depicted in the USCT imaging modes.

3D Ultrasound Computer Tomography (3D USCT) promises reproducible high-resolution images for early detection of breast tumors. The KIT prototype provides three different modalities: reflectivity, speed of sound, and attenuation. The reflectivity images are reconstructed using a Synthetic Aperture Focusing Technique (SAFT) algorithm. For high-resolution re ectivity images, with spatially homogeneous reflectivity, attenuation correction is necessary. In this paper we present a GPU accelerated attenuation correction for 3D USCT and evaluate the method by means of image quality metrics; i.e. absolute error, contrast and spatially homogeneous reflectivity. A threshold for attenuation correction was introduced to preserve a high contrast. Simulated and in-vivo data were used for analysis of the image quality. Attenuation correction increases the image quality by improving spatially homogeneous reflectivity by 25 %. This leads to a factor 2.8 higher contrast for in-vivo data.

In a first pilot study with 3D Ultrasound Computer Tomography especially the speed of sound images of the breast showed promising results for breast cancer detection. Yet, the resolution of transmission tomography in our system is limited in comparison to the reflectivity volumes. In this paper we describe a setup for an improved device overcoming this limitation. An optimized geometry of the transducers in a sparse distribution improves both the contrast in reflection and contrast and resolution in transmission tomography by nearly an order of magnitude. Furthermore data acquisition is accelerated from 10 to less than 2 minutes. The derived methods, simulation, and measuring results are described.

An essential processing step for comparison of Ultrasound Computer Tomography images to other modalities, as well as for the use in further image processing, is to segment the breast from the background. In this work we present a (semi-) automated 3D segmentation method which is based on the detection of the breast boundary in coronal slice images and a subsequent surface fitting. The method was evaluated using a software phantom and in-vivo data. The fully automatically processed phantom results showed that a segmentation of approx. 10% of the slices of a dataset is sufficient to recover the overall breast shape. Application to 16 in-vivo datasets was performed successfully using semi-automated processing, i.e. using a graphical user interface for manual corrections of the automated breast boundary detection. The processing time for the segmentation of an in-vivo dataset could be significantly reduced by a factor of four compared to a fully manual segmentation. Comparison to manually segmented images identified a smoother surface for the semi-automated segmentation with an average of 11% of differing voxels and an average surface deviation of 2mm. Limitations of the edge detection may be overcome by future updates of the KIT USCT system, allowing a fully-automated usage of our segmentation approach.

A promising candidate for improved imaging of breast cancer is ultrasound computer tomography (USCT). The aim of this work was to design a new aperture for our full 3D USCT which extends the properties of the current aperture to a larger ROI fitting the buoyant breast in water and decreasing artifacts in transmission tomography. The optimization resulted in a larger opening angle of the transducers, a larger diameter of the aperture and an approximately homogeneous distribution of the transducers, with locally random distances. The developed optimization methods allow us to automatically generate an optimized aperture for given diameters of apertures and transducer arrays, as well as quantitative comparison to other arbitrary apertures. Thus, during the design phase of the next generation KIT 3D USCT, the image quality can be balanced against the specification parameters and given hardware and cost limitations. The methods can be applied for general aperture optimization, only limited by the assumptions of a hemispherical aperture and circular transducer arrays.

Ultrasound Computer Tomography (USCT) is a promising modality for breast imaging. We developed and tested the first full 3D USCT system aimed at in-vivo imaging. It is based on approx. 2000 ultrasound transducers surrounding the breast within a water bath. From the acquired signal data, reflectivity, attenuation and sound speed images are reconstructed. In a first in-vivo study we imaged ten patients and compared them to MRI images. To overcome the considerably different breast positioning in both imaging methods, an image registration and image fusion based on biomechanical modeling of the buoyancy effect and surface-based refinement was applied. The resulting images are promising: compared with the MRI ground truth, similar tissue structures can be identified. While reflection images seem to image even small structures, sound speed imaging seems to be the best modality for detecting cancer. The registration of both imaging methods allows browsing the volume images side by side and enables recognition of correlating tissue structures. The first in-vivo study was successfully completed and encourages for a second in-vivo study with a considerably larger number of patients, which is currently ongoing.

The KIT 3D USCT was tested in a pilot study on ten patients. The primary goals of the pilot study were to test the USCT device, the data acquisition protocols, the image reconstruction methods and the image fusion techniques in a clinical environment. The study was conducted successfully; the data acquisition could be carried out for all patients with an average imaging time of six minutes per breast. First reconstructions provide promising images. Overlaid volumes of the modalities show qualitative and quantitative information at a glance. The results led to further optimization of the system and the data acquisition protocol.

Ultrasound Computer Tomography (USCT) is a promising 3D modality for early breast cancer detection, which is expected to provide quantitative imaging. The aim of this paper is to evaluate the quantitative diagnostic value of the USCT images, i.e. sound speed and attenuation images, using X-ray mammograms as ground truth. For this purpose we applied our 2D/3D registration method, which is based on biomechanical modeling of the breast. Mammograms were segmented into fatty, glandular and tumorous tissue. For each tissue, the average sound speed and attenuation in the corresponding USCT images was calculated. Tumorous tissue could be separated from fatty and glandular tissue using a fixed absolute sound speed threshold in all regarded datasets. By combining sound speed and attenuation, the separation between fatty and glandular tissue could be improved. By overlaying sound speed and attenuation information on the mammogram, quantitative and morphological information can be combined for multimodal diagnosis. This may benefit early breast cancer detection in future.

Ultrasound Computer Tomography is an upcoming imaging modality for early breast cancer detection. For evaluation of the method, comparison with the standard method X-ray mammography is of strongest interest. To overcome the significant differences in dimensionality and compression state of the breast, in earlier work a registration method based on biomechanical modeling of the breast was proposed. However only homogeneous models could be applied, i.e. inner structures of the breast were neglected. In this work we extend the biomechanical modeling of the breast by estimating patient-specific tissue parameters automatically from the speed of sound volume. Two heterogeneous models are proposed modeling a quadratic and an exponential relationship between speed of sound and tissue stiffness. The models were evaluated using phantom images and clinical data. The size of all lesions is better preserved using heterogeneous models, especially using an exponential relationship. The presented approach yields promising results and gives a physical justification to our registration method. It can be considered as a first step towards a realistic modeling of the breast.

We designed and built a 3D ultrasound computer tomography (USCT) device with a nearly isotropic and spatially invariant 3D point spread function, to be tested in a clinical study. The objective of this work was to image two healthy volunteers and to evaluate the USCT volumes in comparison to corresponding Magnetic Resonance Images (MRI). The here presented volumes are reflectivity images generated with 3D synthetic aperture focusing technique. The volunteers were imaged with different parameterizations of the data acquisition. The data acquisition time was between four and twelve minutes. For both volunteers we found that the breast surface and inner structures are clearly shown in the USCT volume and fit the structures given by the MRI.

Ultrasound Computer Tomography (USCT) is an upcoming modality for early breast cancer diagnosis, which provides morphological as well as quantitative imaging. In order to compare USCT images to the standard modality X-ray mammography, a 2D/3D registration has to be applied. To analyze the relevance of sound speed images as a quantitative imaging method for tissue characterization, the aim of this paper is to quantify sound speed values of different types of tissue using mammograms as ground truth. Mammograms are segmented and classified into fat, glandular and tumorous tissue. For each tissue class the average sound speed in the registered sound speed image is calculated. The mean absolute sound speed was 1457 m/s for regions segmented as fatty tissue, 1470 m/s for glandular and 1509 m/s for tumorous tissue. For all ten datasets, the sound speed in tumorous tissue was significantly higher than in glandular and fatty tissue. Also glandular and fatty tissue could be separated easily by the absolute sound speed values. By color-coding sound speed, quantitative information from USCT and morphological information from X-ray mammography are fused for combined diagnosis. We believe this method will help radiologists in gaining experience in the reading of USCT images. The combination of diagnostic information is likely to be beneficial to early breast cancer detection.

Breast cancer is the most common cancer among women. The established screening method to detect breast cancer in an early state is X-ray mammography. However, X-ray frequently provides limited contrast of tumors located within glandular tissue. A new imaging approach is Ultrasound Computer Tomography generating threedimensional volumes of the breast. Three different images are available: reflectivity, attenuation and speed of sound. The correlation of USCT volumes with X-ray mammograms is of interest for evaluation of the new imaging modality as well as for a multimodal diagnosis. Yet, both modalities differ in image dimensionality, patient positioning and deformation state of the breast. In earlier work we proposed a methodology based on Finite Element Method to register speed of sound images with the according mammogram. In this work, we enhanced the methodology to register all three image types provided by USCT. Furthermore, the methodology is now completely automated using image similarity measures to estimate rotations in datasets. A fusion methodology is proposed which combines the information of the three USCT image types with the X-ray mammogram via semitransparent overlay images. The evaluation was done using 13 datasets from a clinical study. The registration accuracy was measured by the displacement of the center of a lesion marked in both modalities. Using the automated rotation estimation, a mean displacement of 10.4 mm was achieved. Due to the clinically relevant registration accuracy, the methodology provides a basis for evaluation of the new imaging device USCT as well as for multimodal diagnosis.

Grid infrastructures provide large computing resources for grand challenge applications. The Interactive European Grid project developed mechanisms for the interactive and MPI-parallel usage of these resources. This chapter presents in detail an interactive approach that allows API style communication with grid resources and execution of parallel applications using MPI. The RPC API of GridSolve and the middleware gLite are integrated by the software Giggle for resource allocation. Furthermore, we describe the adaption of Giggle and GridSolve to allow the remote execution of MPI applications on RPC basis. Performance measurements are carried out to benchmark the solution. We can show an overhead of only few seconds in contrast to a job submitted via gLite, which can take minutes. We find that our approach is mainly limited by the data transfer rates to the different computing sites.

Breast cancer is the most common cancer among women. The established screening method to detect breast cancer is X-ray mammography. However, X-ray frequently provides poor contrast of tumors located within glandular tissue. In this case, additional modalities like MRI are used for diagnosis in clinical routine. A new imaging approach is Ultrasound Computer Tomography, generating three-dimensional speed of sound images. High speed of sound values are expected to be an indicator of cancerous structures. Therefore, the combination of speed of sound images and X-ray mammograms may benefit early breast cancer diagnosis. In previous work, we proposed a method based on Finite Elements to automatically register speed of sound images with the according mammograms. The FEM simulation overcomes the challenge that X-ray mammograms show two-dimensional projections of a deformed breast whereas speed of sound images render a three-dimensional undeformed breast in prone position. In this work, 15 datasets from a clinical study were used for further evaluation of the registration quality. The quality of the registration was measured by the displacement of the center of a lesion marked in both modalities. We found a mean displacement of 7.1 mm. For visualization, an overlay technique was developed, which displays speed of sound information directly on the mammogram. Hence, the methodology provides a good basis for multimodal diagnosis using mammograms and speed of sound images. It proposes a guidance tool for radiologists who may benefit from the combined information.

Breast cancer is the most common type of cancer among women in Europe and North America. The established screening method to detect breast cancer is X-ray mammography, although X-ray frequently provides poor contrast for tumors located within glandular tissue. A new imaging approach is Ultrasound Tomography generating three-dimensional speed of sound images. This paper describes a method to evaluate the clinical applicability of three-dimensional speed of sound images by automatically registering the images with the corresponding X-ray mammograms. The challenge is that X-ray mammograms show two-dimensional projections of a deformed breast whereas speed of sound images render a three-dimensional undeformed breast in prone position. This conflict requires estimating the relation between deformed and undeformed breast and applying the deformation to the three-dimensional speed of sound image. The deformation is simulated based on a biomechanical model using the finite element method. After simulation of the compression, the contours of the X-ray mammogram and the projected speed of sound image overlap congruently. The quality of the matching process was evaluated by measuring the overlap of a lesion marked in both modalities. Using four test datasets, the evaluation of the registration resulted in an average tumor overlap of 97%. The developed registration provides a basis for systematic evaluation of the new modality of three-dimensional speed of sound images, e.g. allows a greater understanding of tumor depiction in these new images.

gLite is one of the largest distributed computing infrastructures in operation. It provides access to hundreds of different clusters - all installed and maintained in different ways. This paper analyses the difficulties which users typically experience when moving from their own workstation via clusters or supercomputers to the grid. Based on that analysis, this paper presents tools, which helps to overcome this gap and introduces an advanced command line interface to the grid.

At Forschungszentrum Karlsruhe, a new imaging system for early diagnosis of breast cancer is currently developed. The 3D Ultrasound Computer Tomography (USCT) is based on approx. 2000 ultrasound transducers which produce 20 GB of raw data consisting of 3.5 mio. A-scans (amplitude scans) for one image. The large number of A-scans, the large amount of data and the complex relationship between raw data and reconstructed image makes analysis, understanding and further development difficult for the scientists and especially for new employees and students. For this reason, an interactive graphical user interface (GUI) was developed using MATLAB. It integrates existent analysis methods and is easily extendable with new functionality via a plugin concept. The software provides several visualization functions for the raw data, the reconstructed 3D images, the USCT aperture and the relationships between them. The software shows the capability of MATLAB as programming language for numerical solutions as well as GUIs and complex software systems. It has a large benefit to the working group by integrating analysis methods and sharing them. The GUI and the visualization of the complex relationships of the USCT reduces the training period for new employees and students.

At Forschungszentrum Karlsruhe a 3D ultrasound computer tomograph (USCT) for breast cancer diagnosis is currently under development. For many applications, i.e. reconstruction of speed of sound maps, signal denoising,signal compression and calibration, it is necessary to detect pulses and their parameters (center frequency, bandwidth factor, time of arrival, phase and amplitude) accurately and robustly. The pulse detection is demanding due to low signal-to-noise ratios (SNR) caused by unfocused pulses from single emitters. Besides that angle dependent pulse shapes result in ultrasonic echoes which are similar in their center frequency but vary strongly in their bandwidth factor. Additionally, in sum 3.5 million A-scans (20 GB) are acquired, so that the pulse detection has to be fast and efficient. This is achieved by using a fast pre-classifler in order to separate the ultrasonic echoes from the non-white system noise of our system. The detected echoes are then passed individually to a parameter estimation method to determine pulse parameters accurately. An analysis of several classifiers resulted in an alternating decision tree which is both fast and accurate. Classification performance of 95% could be achieved as well as robust parameter estimation with non-white system noise if the SNR is larger than 3 dB. A comparative image reconstruction resulted in significantly sharper images.

The KIT 3D USCT surrounds the breast with ultrasound transducers on a 3D aperture and emits and receives nearly spherical wave fronts for synthetic aperture focusing. This full 3D system achieves isotropic 3D resolution, has a nearly spatial invariant point spread function, and allows fast data acquisition. The 3D USCT device is equipped with 2041 ultrasound transducers. The acquisition is carried out by sequentially selecting a single emitter, sending a chirp at 2.5 MHz center frequency and recording the transmitted and reflected waves with all receivers. Rotational and translational movement of the aperture is applied to enhance the image contrast. Up to 40 GB of raw data is acquired with 480 parallel channels for digitization at 12 bit and 20 MHz sampling frequency. In a first pilot study, ten patients with different lesions were imaged. Speed of sound, attenuation, and reflection images of each patient were derived from the raw data. Overlaid volumes of the modalities show qualitative and quantitative information at a glance. The results are promising because the breasts' tissue structures and cancerous lesions could be identified in the USCT images.

The computer gaming industry is traditionally the moving power and spirit in the development of computer visualization hardware and software. This year, affordable and high quality virtual reality headsets became available and the science community is eager to get benefit from it. This paper describes first experiences in adapting the new hardware for three different visualization use cases. In all three examples existing visualization pipelines were extended by virtual reality technology. We describe our approach, based on the HTC Vive VR headset, the open source software Blender and the Unreal Engine 4 game engine. The use cases are from three different fields: large-scale particle physics research, X-ray-imaging for entomology research and medical imaging with ultrasound computer tomography. Finally we discuss benefits and limits of the current virtual reality technology and present an outlook to future developments.