Neural networks for faster laser ultrasound tomography in tissue phantoms.

Natural variation in sound speed within ocean water can degrade sub-seabed images from 2D, 3D and 4D seismic reflection datasets. Degrading effects include vertical offset of reflections between adjacent or intersecting sail lines and difficulties in suppressing multiples from water layer reverberations. Here we investigate water layer variability in Rockall Trough, offshore Ireland, in water depths ranging from 200 m to 3.5 km. A compilation of vertical sound speed profiles, calculated from temperature and salinity profiles obtained by probes lowered from ships, shows that the mean sound speed in the water layer mostly varies between 1490 and 1500 m s −1 . Vertical offsets of up to about 15 ms two-way travel time are predicted at seismic line intersections. A significant amount of the total observed sound speed variability can occur along a single seismic sail line. These effects result mainly from spatial and temporal fluctuations in the thicknesses of, and vertical sound speed gradients within, an upper layer of North Atlantic Central Water and a mid-depth layer of Mediterranean Outflow Water. Seismic sections across Rockall Trough commonly show lateral variability in reflectivity within these same two water layers. Some reflective packages contain lens-shaped structures consisting of reflective rims and transparent cores and with diameters between 10 and 50 km. Other reflective packages have abrupt, almost vertical boundaries and no distinct transparent core. We infer that the lateral boundaries of the reflective packages are likely to be associated with significant variations in average water layer sound speed. When processing 2D, 3D and 4D seismic surveys of regions of high oceanic variability, such as Rockall Trough, it is necessary to employ techniques that can solve for and then remove the effect of significant fluctuations in water layer sound speed over time periods as short as a few hours and distances as short as a kilometre.

Full noncontact laser ultrasound (LUS) imaging has several distinct advantages over current medical ultrasound (US) technologies: elimination of the coupling mediums (gel/water), operator-independent image quality, improved repeatability, and volumetric imaging. Current light-based ultrasound utilizing tissue-penetrating photoacoustics (PA) generally uses traditional piezoelectric transducers in contact with the imaged tissue or carries an optical fiber detector close to the imaging site. Unlike PA, the LUS design presented here minimizes the optical penetration and specifically restricts optical-to-acoustic energy transduction at the tissue surface, maximizing the generated acoustic source amplitude. With an appropriate optical design and interferometry, any exposed tissue surfaces can become viable acoustic sources and detectors. LUS operates analogously to conventional ultrasound but uses light instead of piezoelectric elements. Here, we present full noncontact LUS results, imaging targets at ~5 cm depths and at a meter-scale standoff from the target surface. Experimental results demonstrating volumetric imaging and the first LUS images on humans are presented, all at eye- and skin-safe optical exposure levels. The progression of LUS imaging from tissue-mimicking phantoms, to excised animal tissue, to humans in vivo is shown, with validation from conventional ultrasound images. The LUS system design insights and results presented here inspire further LUS development and are a significant step toward the clinical implementation of LUS.

Theoretical and phantom based investigation of the impact of sound speed and backscatter variations on attenuation slope estimation

To show the effect of speed of sound (SOS) aberration on ultrasound guided radiotherapy (US-gRT) as a function of implemented workflow. US systems assume that SOS is constant in human soft tissues (at a value of 1540 m∕s), while its actual nonuniform distribution produces small but systematic errors of up to a few millimeters in the positions of scanned structures. When a coregistered computerized tomography (CT) scan is available, the US image can be corrected for SOS aberration. Typically, image guided radiotherapy workflows implementing US systems only provide a CT scan at the simulation (SIM) stage. If changes occur in geometry or density distribution between SIM and treatment (TX) stage, SOS aberration can change accordingly, with a final impact on the measured position of structures which is dependent on the workflow adopted. Four basic scenarios were considered of possible changes between SIM and TX: (1) No changes, (2) only patient position changes (rigid rotation-translation), (3) only US transducer position changes (constrained on patient's surface), and (4) patient tissues thickness changes. Different SOS aberrations may arise from the different scenarios, according to the specific US-gRT workflow used: intermodality (INTER) where TX US scans are compared to SIM CT scans; intramodality (INTRA) where TX US scans are compared to SIM US scans; and INTERc and INTRAc where all US images are corrected for SOS aberration (using density information provided by SIM CT). For an experimental proof of principle, the effect of tissues thickness change was simulated in the different workflows: a dual layered phantom was filled with layers of sunflower oil (SOS 1478 m∕s), water (SOS 1482 m∕s), and 20% saline solution (SOS 1700 m∕s). The phantom was US scanned, the layer thicknesses were increased and the US scans were repeated. The errors resulting from the different workflows were compared. Theoretical considerations show that workflows implementing SOS correction based on SIM-CT scan (INTERc, INTRAc) give null errors in all scenarios except when tissues thickness changes, where an error proportional to the degree of change in SOS maps between SIM and TX (ΔSOS) occurs. An uncorrected workflow such as INTER produces in all scenarios a pure SOS error, while uncorrected INTRA produces a null error for rotation-translation of the patient, a ΔSOS error for changing tissues thickness and an error proportional to the degree of SOS distribution change along the different lines of view when shifting the transducer. The dual layered phantom demonstrated experimentally that the effect of SOS change between SIM and TX is clinically nonrelevant, being less than the intrinsic resolution of imaging systems, even when a substantial change in thicknesses is applied, provided that a SIM-CT-based SOS aberration correction is applied. Noncorrected workflows produce errors up to 4 mm for INTER and to 3 mm for INTRA in the phantom test. A SOS correction is advantageous for all US-gRT workflows and clinical cases, where the effect of SOS change can be considered a second order effect.

Our previous work on estimating the local speed of sound from average sound speed assumes a perfectly layered medium where sound speed is only allowed to vary axially away from the transducer surface. This layered-medium approach relies on inverting the relationship between the local interval sound speeds in each layer and the effective average sound speed up to a particular imaging depth. The primary limitation of this approach is that local sound speed estimation can become inaccurate in the presence of lateral variations in sound speed or a curved transducer surface. To better estimate sound speed in the presence of these non-idealities, we propose a travel-time tomographic approach that accounts for propagation paths from the scattering volume to each transducer element.

In ultrasound elastography, tissue strains are determined by localizing changes in ultrasound echoes during mechanical loading. The technique has been proposed for arthroscopic quantification of the mechanical properties of cartilage. The accuracy of ultrasound elastography depends on the invariability of sound speed in loaded tissue. In unconfined geometry, mechanical compression has been shown to induce variation in sound speed, leading to errors in the determined mechanical properties. This phenomenon has not been confirmed in indentation geometry, the only loading geometry applicable in situ or in vivo. In the present study, ultrasound speed during indentation of articular cartilage was characterized and the effect of variable sound speed on the strain measurements was investigated. Osteochondral samples (n = 7, diameter = 25.4 mm), prepared from visually intact bovine patellae (n = 7), were indented with a plane-ended ultrasound transducer (diameter = 5.6 mm, peak frequency: 8.1 MHz). A sequence of three compression tests (strain-rate = 10%/s, 2700-s relaxation) was applied using the mean strains of 2.2%, 4.5%, and 6.4%. Then, ultrasound speed during the ramp and stress-relaxation phases was determined using the time-of- flight technique. To investigate the role of cartilage structure and composition for sound speed in loaded articular cartilage, a sample-specific fibril-reinforced poroviscoelastic (FRPVE) finite element model was constructed and fitted to experimental mechanical data. Ultrasound speed in articular cartilage decreased significantly during dynamic indentation (p <; 0.05). The magnitude of the decrease in speed during indentation was related to the applied strain. However, the relative error in acoustically determined tissue strain was inversely related to the magnitude of true strain. The modeling results suggested that the compression-related variation in sound speed is controlled by changes in the collagen architecture during dynamic indentation. To conclude, variation in sound speed during dynamic indentation of articular cartilage may lead to significant errors in the values of measured mechanical parameters. Because the relative errors are inversely proportional to applied strain, higher strains should be used to minimize the errors in, e.g., in vivo measurements.

The refined resilient model for underwater acoustic positioning

An image reconstruction method for endoscopic photoacoustic tomography in tissues with heterogeneous sound speed

Radial Anatomic Variation of Ultrasonic Velocity in Human Cortical Bone

With the rapid development of deep learning technology, deep neural networks can effectively enhance the performance of computed tomography (CT) reconstructions. One kind of commonly used method to construct CT reconstruction networks is to unroll the conventional iterative reconstruction (IR) methods to convolutional neural networks (CNNs). However, most unrolling methods primarily unroll the fidelity term of IR methods to CNNs, without unrolling the prior terms. The prior terms are always directly replaced by neural networks. In conventional IR methods, the prior terms play a vital role in improving the visual quality of reconstructed images. Unrolling the hand-crafted prior terms to CNNs may provide a more specialized unrolling approach to further improve the performance of CT reconstruction. In this work, a primal-dual network (PD-Net) was proposed by unrolling both the data fidelity term and the total variation (TV) prior term, which effectively preserves the image edges and textures in the reconstructed images. By further deriving the Chambolle-Pock (CP) algorithm instance for CT reconstruction, we discovered that the TV prior updates the reconstructed images with its divergences in each iteration of the solution process. Based on this discovery, CNNs were applied to yield the divergences of the feature maps for the reconstructed image generated in each iteration. Additionally, a loss function was applied to the predicted divergences of the reconstructed image to guarantee that the CNNs' results were the divergences of the corresponding feature maps in the iteration. In this manner, the proposed CNNs seem to play the same roles in the PD-Net as the TV prior in the IR methods. Thus, the TV prior in the CP algorithm instance can be directly unrolled to CNNs. The datasets from the Low-Dose CT Image and Projection Data and the Piglet dataset were employed to assess the effectiveness of our proposed PD-Net. Compared with conventional CT reconstruction methods, our proposed method effectively preserves the structural and textural information in reference to ground truth. The experimental results show that our proposed PD-Net framework is feasible for the implementation of CT reconstruction tasks. Owing to the promising results yielded by our proposed neural network, this study is intended to inspire further development of unrolling approaches by enabling the direct unrolling of hand-crafted prior terms to CNNs.

In this work we study the coupled response of zero-group-velocity (ZGV) modes and test liquids. In particular, we assess the viability of exploiting the spatiotemporal constraint of ZGV modes to derive techniques for non-contact, non-invasive liquid characterization to support process optimization. The results demonstrate that the ZGV resonance behaves like a baffled piston source that couples into standing waves in the test liquid. Moreover the particle displacement associated with the liquid resonance extends over macroscopic length scales (∼cm). The resonance frequencies are governed by the acoustic properties and depth, h, of the test liquid. Therefore, the test liquid sound speed, cL , can be determined by analysing the spectral content of the system response. While the technique provides an accurate measure of sound speed, the morphology of immiscible mixtures (e.g. buoyancy separated or emulsion) affects the variation in sound speed with composition. In this case, it is necessary to know the mixture morphology in order to utilize the sound speed as a proxy for composition. The technology is applicable to any thin-walled (mm–cm) structure (pipe, tank, etc) commonly utilized for liquid handling. As a result, the measurement is broadly applicable to many industrial settings that require convenient and accurate methods to determine liquid composition in support of process evaluation and optimization.

Regions in the B-mode image of a cylinder where the location of a point reflector is changed by sound speed variation

Stars slightly more massive than the Sun develop small convective cores during their main sequence phase of evolution. The edges of these convective cores are associated with rapid variations in the sound speed which influence the frequencies of acoustic oscillations. In this work we use an asymptotic analysis that is valid near the turning point of the oscillations, to derive the signature that tiny convective cores, such as those expected in main‐sequence solar‐like pulsators, produce in the oscillation frequencies. Moreover, we present a seismic diagnostic tool that is capable of isolating essential properties of that signature. The analysis presented here is an extension of that published by Cunha &amp; Metcalfe (2007). In particular, in the current work the functional form adopted in the modelling of the sharp variation in sound speed at the edge of the core is more realistic. That results in an improved capability to extract the size of the sharp variation in sound speed at the edge of the core, a property that depends on the physical processes of mixing and diffusion taking place in that region of the star (© 2010 WILEY‐VCH Verlag GmbH &amp; Co. KGaA, Weinheim)

This paper considers the problem of calculating the reflection coefficient of a plane wave incident on an inhomogeneous elastic solid layer of finite thickness, overlying a semi-infinite, homogeneous solid substrate. The physical properties of the inhomogeneous layer, namely the density, compressional (sound) speed, shear speed, and attenuation, are all assumed to vary with depth. It is shown that, provided terms involving the gradient of the shear speed and of the shear modulus are ignored, and volume coupling between compressional and shear waves is neglected, analytical solutions of the resulting equations can be obtained. The assumptions made are justified at most frequencies of practical interest in underwater acoustics. In the case of a solid whose density and shear speed are constant, the results obtained are exact solutions of the full equations of motion, and may usefully be compared with numerical solutions in which the variation in sound speed through the layer is represented by a number of homogeneous sublayers. It is concluded that, with realistic sediment and substrate properties, a surprisingly large number of sublayers can be needed to give accurate results.

Ultrasound computed tomography (USCT) is one of the advanced imaging techniques used in structural health monitoring (SHM) and medical imaging due to its relatively low-cost, rapid data acquisition process. The time-domain full waveform inversion (TDFWI) method, an iterative inversion approach, has shown great promise in USCT. However, such an iterative process can be very time-consuming and computationally expensive but can be greatly accelerated by integrating an AI-based approach (e.g., convolution neural network (CNN)). Once trained, the CNN model takes low-iteration TDFWI images as input and instantaneously predicts material property distribution within the scanned region. Nevertheless, the quality of the reconstruction with the current CNN degrades with the increased complexity of material distributions. Another challenge is the availability of enough experimental data and, in some cases, even synthetic surrogate data. To alleviate these issues, this paper details a systematic study of the enhancement effect of a 2D CNN (U-Net) by improving the quality with limited training data. To achieve this, different augmentation schemes (flipping and mixing existing data) were implemented to increase the amount and complexity of the training datasets without generating a substantial number of samples. The objective was to evaluate the enhancement effect of these augmentation techniques on the performance of the U-Net model at FWI iterations. A thousand numerically built samples with acoustic material properties are used to construct multiple datasets from different FWI iterations. A parallelized, high-performance computing (HPC) based framework has been created to rapidly generate the training data. The prediction results were compared against the ground truth images using standard matrices, such as the structural similarity index measure (SSIM) and average mean square error (MSE). The results show that the increased number of samples from augmentations improves shape imaging of the complex regions even with a low iteration FWI training data.