This PDF file contains the front matter associated with SPIE Proceedings Volume10955, including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Super-resolution ultrasound imaging (SR-US) is a new technique which breaks the diffraction limit and can help visualize microvascularity at a resolution of tens of microns. However, image processing methods for spatiotemporal filtering needed in SR-US for microvascular delineation, such as singular value decomposition (SVD), are computationally burdensome and must be performed off-line. The goal of this study was to evaluate a novel and fast method for spatiotemporal filtering to segment the microbubble (MB) contrast agent from the tissue signal with a trained 3D convolutional neural network (3DCNN). In vitro data was collected using a programmable ultrasound (US) imaging system (Vantage 256, Verasonics Inc, Kirkland, WA) equipped with an L11-4v linear array transducer and obtained from a tissue-mimicking vascular flow phantom at flow rates representative of microvascular conditions. SVD was used to detect MBs and label the data for training. Network performance was validated with a leave-one-out approach. The 3DCNN demonstrated a 22% higher sensitivity in MB detection than SVD on in vitro data. Further, in vivo 3DCNN results from a cancer-bearing murine model revealed a high level of detail in the SR-US image demonstrating the potential for transfer learning from a neural network trained with in vitro data. The preliminary performance of segmentation with the 3DCNN was encouraging for real-time SR-US imaging with computation time as low as 5 ms per frame.

Non-contrast perfusion ultrasound imaging is difficult, mainly because of tissue clutter interference with blood. We previously developed an adaptive tissue clutter demodulation technique to overcome this problem and showed that power Doppler image quality can be improved when combining adaptive demodulation with improvements in beamforming and tissue filtering, namely angled plane wave beamforming and singular value decomposition filtering. In this work we aim to evaluate an independent component analysis-based filtering method using angled plane wave beamforming and compare it to singular value decomposition filtering with and without adaptive demodulation using single vessel simulations and phantoms. We show that with optimal filter cutoffs, independent component analysis-based filtering consistently improves signal and contrast-to-noise ratios, and it resulted in an 8.4dB average increase in optimal signal-to-noise ratio compared to singular value decomposition filtering in phantoms with 1mm/s flow and a 700ms ensemble.

In echocardiography, blood-flow measurement is important, and several methods of measuring the velocity vector of blood flow have been proposed including echographic speckle tracking. Echographic speckle tracking is typically based on blockmatching algorithms; however, they incur high calculation cost; thus, are time-consuming. To enable real-time blood-flow vector measurement, we applied the Kanade-Lucas-Tomasi (KLT) algorithm to echographic speckle tracking, but the measurement accuracy was low in preliminary trial. This is mainly because echographic speckles deform as speckles move due to the acoustic pressure field of a transmitting beam. The objective of this study was to minimize the estimation error of KLT-based speckle tracking by analyzing error propagation. We analyzed error propagation from the acoustic pressure field to the velocity error by simplifying speckle deformation and formulated the major error factors. From this analysis, we propose a policy of determining the measurement conditions, which are region-of-interest (ROI) size, waveform, and number of ROI divisions, for minimizing estimation error. We verified the proposed policy through numerical simulations. As a result of the analysis and simulations, the gradient of the pressure field, number of ROI divisions, and moving distance of a speckle accounted for most of the estimation error. In addition, optimizing these conditions restricted the mean estimation error to less than 10%. These results indicate that the accuracy of KLT-based speckle tracking can reach a practical level by designing measurement conditions based on the proposed policy.

Diabetes is a major disease and known to impair microvascular recruitment due to insulin resistance. Previous quantifications of the changes in microvascular networks at the capillary level were being performed with either full or manually selected region-of-interests (ROIs) from super-resolution ultrasound (SR-US) images. However, these approaches were imprecise, time-consuming, and unsuitable for automated processes. Here we provided a custom software solution for automated multiscale analysis of SR-US images of tissue microvascularity patterns. An Acuson Sequoia 512 ultrasound (US) scanner equipped with a 15L8-S linear array transducer was used in a nonlinear imaging mode to collect all data. C57BL/6J male mice fed standard chow and studied at age 13-16 wk comprised the lean group (N = 14), and 24-31 wk-old mice who received a high-fat diet provided the obese group (N = 8). After administration of a microbubble (MB) contrast agent, the proximal hindlimb adductor muscle of each animal was imaged (dynamic contrast-enhanced US, DCE-US) for 10 min at baseline and again at 1 h and towards the end of a 2 h hyperinsulinemiceuglycemic clamp. Vascular structures were enhanced with a multiscale vessel enhancement filter and binary vessel segments were delineated using Otsu’s global threshold method. We then computed vessel diameters by employing morphological image processing methods for quantitative analysis. Our custom software enabled automated multiscale image examination by defining a diameter threshold to limit the analysis at the capillary level. Longitudinal changes in AUC, I_PK, and MVD were significant for lean group (p < 0.02 using Full-ROI and p < 0.01 using 150 μm-ROI) and for obese group (p < 0.02 using Full-ROI, p < 0.03 using 150 μm-ROI). By eliminating large vessels from the ROI (above 150 μm in diameter), perfusion parameters were more sensitive to changes exhibited by the smaller vessels, that are known to be more impacted by disease and treatment.

In addition to structural morphology, tissue’s vascular network may provide valuable complementary information on the altered lesions and the tumor angiogenesis. Although ultrafast Doppler ultrasound (UDF) imaging enables ultrasound to image microvessels with high sensitivity, these images still suffer from artifacts. In this study, we addressed small vessel visualization and associated noise problem in ultrasound high framerate plane wave in-vivo imaging. We developed a combination of nonlocal means and morphological filtering on the UDF clutter removed data in order to obtain enhanced vessel images and improved outlining. We tested our algorithm on a flow phantom and in vivo data of the breast masses. The results show that the proposed method added an incremental gain of about 16 dB in terms of signal to noise ratio and has potential to facilitate ultrasound small vessel imaging quantification.

Ultrasound computed tomography (USCT) is an emerging computed imaging modality that holds great promise for breast cancer screening. Waveform inversion-based image reconstruction methods that are based on the wave equation can produce speed of sound (SOS) images with improved spatial resolution over those produced by ray-based methods. However, waveform inversion methods are computationally demanding and the computational burden increases significantly when wave propagation is conducted in the 3D domain. Experimental systems that are carefully designed with elevationally focused transducers allow the reconstruction of SOS over a 3D volume to be estimated as a reconstruction of a stack of 2D slices. This allows us to circumvent the computational burden associated with 3D waveform inversion by applying full-waveform inversion (FWI) algorithms in the computationally attractive 2D domain. In such scenario, there is a model mismatch between the 2D model employed in the reconstruction process, and the 3D model that represents the true physics of wave propagation. The mismatch is more pronounced when the medium properties are inhomogeneous in 3D and can have deleterious effects on the reconstructed FWI images. To overcome this issue, we propose to implement a convolutional neural network that can map a 3D USCT dataset to its equivalent 2D USCT dataset. The transformed data can then be subsequently employed in a 2D waveform inversion algorithm, allowing for mitigation of artifacts due to the 3D-2D model mismatch without significant increase in computational cost. Reconstructed images from realistic numerical breast phantoms are employed to demonstrate the feasibility and effectiveness of the approach.

This work presents refraction-corrected sound speed reconstruction techniques for transmission-based ultrasound computed tomography using a circular transducer array. Pulse travel times between element pairs can be calculated from slowness (the reciprocal of sound speed) using the eikonal equation. Slowness reconstruction is posed as a nonlinear least squares problem where the objective is to minimize the error between measured and forward-modeled pulse travel times. The Gauss-Newton method is used to convert this problem into a sequence of linear least-squares problems, each of which can be efficiently solved using conjugate gradients. However, the sparsity of ray-pixel intersection leads to ill-conditioned linear systems and hinders stable convergence of the reconstruction. This work considers three approaches for resolving the ill-conditioning in this sequence of linear inverse problems: 1) Laplacian regularization, 2) Bayesian formulation, and 3) resolution-filling gradients. The goal of this work is to provide an open-source example and implementation of the algorithms used to perform sound speed reconstruction, which is currently being maintained on Github: https://github.com/ rehmanali1994/refractionCorrectedUSCT.github.io

The Distorted Born Iterative (DBI) method is used for ultrasound tomography in order to localize and identify malignant breast tissues. This approach begins with the Born approximation to generate an initial prediction of the scattering function. Then, iteratively solves the forward problem for the total field and the inhomogeneous Green’s function, and the inverse problem for the scattering function. The drawback of this method is that the associated inverse scattering problem is ill-posed. We are proposing the Truncated General Singular Value Decomposition (TGSVD) approach as a regularization method for the ill posed inverse problem Xy = b in DBI and comparing it to the well known Truncated Singular Value Decomposition (TSVD). The TGSVD employs generalized SVD (GSVD) of matrix pair (X,L) and is neglecting the smallest, contaminated with noise, generalized singular values, while regularization matrix L (we used the first order derivative operator) is responsible for smoothing the solution. This results in better image quality. We compared the performances of these two methods on simulated phantom and proved that TGSVD produces lower relative error and better reconstructed image.

For breast cancer imaging by ultrasound computed tomography (CT) without dependence on patient breast size, we previously developed a high-sensitivity scan method in which a virtual fan-beam (vfan-beam) is generated from ultrasound waves emitted from 128 sources with unique delay times. Full waveform inversion (FWI) with multiple sound sources has not been previously applied to ultrasound CT using a ring transducer array. We have now developed a FWI calculation process that enables a vfan-beam to generate accurate sound speed images. A vfan-beam is accurately modeled by positioning the 128 sources and considering the delay times. The performance of the FWI calculation process with a vfan-beam was evaluated using a prototype ultrasound CT. For a circular phantom, the spatial resolution of a FWI image obtained with a vfan-beam was better than that of a filtered back projection (FBP) image. The image contrast of the FWI calculation process with a vfan-beam was comparable to that of the process with a conventional fanbeam generated from a single source. For a high-attenuation ellipse phantom, the sound speed image obtained with a conventional fan-beam had severe artifacts due to the low signal to noise ratio (SNR). Using a vfan-beam reduced the number of artifacts in the images due to the higher SNR. The FWI calculation process with a vfan-beam visualized a 3 mm inclusion more clearly than the FBP process. A measurement study demonstrated that the FWI process with a vfanbeam with a high SNR reduced the number of artifacts in the sound speed images and improved the spatial resolution for a high-attenuation breast.

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.

Purpose: This study aims to investigate correlation of speed of sound (SoS) map with T2-weighted (T2w) MRI and pathology in an ex vivo human prostate tissue with cancer, as an early proof of concept towards cost effective augmented ultrasound diagnosis of prostate cancer. Method: A commercial breast full angle ultrasound tomography scanner was used to generate US tomography images. Prostate-specific Echolucent mold was fabricated to allow MRI and UST to be spatially correlated. Similarly, a 3D printed mold was developed to align the histology slices with the UST and MRI. The resulting slices of prostate tissue were H and E stained. A radiologist with 10 years of experience in using multi parametric MRI for prostate cancer diagnosis labeled and contoured the suspicious ROIs in both MRI and UST. For all tissue blocks (N=10 slices with 6 mm thickness), H and E slides were prepared and labeled by an expert pathologist. Results: The radiologist found two slices with prominent cancer in each modality (i.e. MR and UST) in the peripheral zone. These two pairs of slices correlated with each other and with slices #5 and #7 in pathology. The cancer ROIs were found at similar locations in all modalities, although MR and UST underestimated the size of lesions (Sørensen–Dice coefficients, with respect to pathology, for T2w and UST were 0.11 and 0.20 respectively for first ROI, and 0.33 and 0.27 for second ROI). The SoS was 1580.4±17.7 m/s and 1571.4±9.2 m/s for normal and cancer tissues in first ROI, and 1577.7±17.7 m/s and 1574.5±10.1 m/s for second ROI. Conclusions: SoS map can correlate with MRI and pathology findings in prostate cancer. Further ex vivo validation with fresh prostate tissue is warranted.

Many soft tissues, including skeletal muscle and kidney, can be modeled astransversely isotropic (TI) materials defined by an axis of symmetry (AoS) perpendicular to a plane of isotropy. In such materials, mechanical properties differ along versus across the AoS. The degree of mechanical anisotropy in TI materials was previously assessed as the ratio of peak displacement (PD) achieved when the long axis of an asymmetric acoustic radiation force (ARF) excitation point spread function (PSF) was aligned along versus across the material’s AoS, but the measurement was qualitative. The objectives of this work were: (i) to derive an empirical model describing the relationship between the PD ratio and shear moduli ratio; (ii) to investigate the impact of ARF excitation PSF aberration due to speed of sound (c), attenuation (α), and dimension of ARF excitation PSF on the empirical model; and (iii) to estimate mechanical anisotropy in excised pig biceps femoris muscles and in vivo pig kidney using the empirical model. The empirical model was derived by simulating ARF Impulse (ARFI) imaging of ‘train’ TI materials (shear moduli ratios varying from 1.0 to 10 in steps of 0.75) and validated on ‘test’ materials (shear moduli ratios varying from 1.25 to 8.75 in steps of 0.75) using finite element method (FEM) models. Siemens VF73 transducer parameter with lateral F/1.5 was simulated for ARFI imaging. The speed of sound and attenuation was set to1540 ms^-1 and 0.5 dB/cm/MHZ, respectively for train materials and was set to1540 or 1620 ms^-1 and 0.5 or 1.0 dB/cm/MHZ, respectively for test materials. To find the impact of ARF excitation PSF dimension, a matrix array was simulated and the lateral and elevational F/# was set to 2.0 and 3.4, respectively for train materials and was set to 2.0 or 3.0 and 3.4 or 5.1, respectively for test materials. Ultrasound tracking of FEM displacements was performed in Field II with an SNR of 30 dB. The average absolute percent error in predicting shear moduli ratio of all ‘test’ materials was 1.6%. The empirical model was not impacted by the deviation from expected attenuation, sound speed, and ARF excitation PSF dimension. Shear moduli ratios derived using the empirical model matched those derived from shear wave elasticity imaging (SWEI) in pig muscle (model: 4.44 ± 0.47 and SWEI: 4.38 ± 0.27), renal medulla (model: 1.31 ± 0.07 and SWEI: 1.32 ± 0.04) and renal cortex (model: 2.0 ± 0.19 and SWEI: 1.99 ± 0.06). These results suggest the feasibility of using the PD empirical model to quantify mechanical anisotropy in biological tissues.

We have fabricated a cylindrical intravascular ultrasound (IVUS) transducer array prototype capable of generating an acoustic radiation force impulse (ARFI) for shear wave elasticity imaging (SWEI). The prototype array was a 4-mm long, 2.5-mm diameter, 4 MHz PZT-8 tube, axially segmented into 12 elements on a 334 µm pitch. This transducer array was used in custom vessel phantoms and in ex vivo porcine artery experiments to investigate the potential for IVUS SWEI to distinguish soft lipid cores from stiffer surrounding tissues. By using this array transducer to generate a radially-directed ARFI “push”, and a Verasonics linear array probe to track displacements in planes parallel to the “push”, SWEI images of a vessel phantom with hard vessel walls and a soft inclusion were obtained. In tissue-mimicking phantoms, focusing the transducer array to a range of 5 mm generated ARFI displacements up to 1.36 and 1.76 times greater than unfocused excitations in the soft and stiff regions, respectively. The measured shear wave speed in the soft inclusion and stiff vessel wall was 0.97±0.59 m/s and 1.66±0.91 m/s, respectively, and was close to the calibrated measurements of 1.21±0.05 m/s and 1.56±0.05 m/s, respectively. A SWEI image of an ex vivo porcine renal artery was obtained using the prototype transducer and external tracking array, and showed an average shear wave speed of 3.97±1.12 m/s. These results demonstrate the potential of this IVUS array to enable SWEI, to quantifiably assess vulnerable vascular plaques.

Cardiac Adipose Tissue (CAT) is a type of visceral fat that is deposited between the myocardium and pericardium. An increased volume of CAT has been recognized as a crucial contributor to cardiovascular and coronary artery diseases. This tissue is a metabolically active organ that affects the cardiac functioning by secreting inflammatory adipokines making it a hazard when present in excess amounts. Quantifying CAT, therefore, can be an important factor in understanding the level of cardiovascular risk. The study presented in this paper investigates the use of frequency content from echocardiography and spectral analysis techniques in differentiating three different cardiac tissue types, including the adipose tissue. Thirteen spectral parameters were computed from the power spectrum of the radio frequency data in three different bandwidth ranges, including 3, 6 and 20 dB. Autoregressive models of order 4 were used as they provide effective estimates of the power spectrum for short-time data. The derived spectral parameters were used in generating random forests for tissue classification. Out of the total 175 ROIs available, 70% of the data was divided into training data and the remaining used as test data. The random forest classifier with 50 classification trees resulted in an overall accuracy of 92.4%, sensitivity of 91.1%, specificity of 93.9%, and Youden’s index of 0.85 for a 20dB bandwidth. This result demonstrates the potential of echocardiography and spectral analysis techniques in differentiating CAT, myocardium, and blood.

Reliable blood flow measurements in the neonatal brain are difficult to obtain with conventional Power Doppler (PD) due to small vessel size, slow flow, and strong reverberation from the cranium. Under such imaging conditions, it is important to use long ensemble lengths and to reduce the acoustic noise in order to separate the slow-flow signal from the stationary-tissue clutter. We have recently developed the short-lag angular coherence (SLAC) beamforming method to reduce noise in the Doppler data, and used it to track blood-flow changes in the brains of neonates. SLAC suppresses the incoherent portion of the beam-summed signals and utilizes Fourier beamforming for fast processing of large Doppler ensembles. To remove stationary tissue signal from the data, we have also utilized spatiotemporal filtering prior to the SLAC processing step. The matching frames of SLAC-based PD and conventional PD were reconstructed from the same Doppler data captured on the neonatal brain vasculature over 4 cm depth. To achieve a fair comparison, the Doppler signal of each modality was normalized by its respective noise profile measured as a function of depth from a stationary speckle phantom. The SLAC images showed better delineation of small vessels, and the vessel SNR was measured to be up to 2 dB higher in SLAC images than in matching PD images. To demonstrate the quantitative aspect of SLAC-based PD, we have also created matched conventional PD and SLAC-based PD videos from the ten-second Doppler scans of neonatal brains. For the vasculature of interest, integrated pixel intensity was computed as a function of time. SLAC-based PD was able to capture changes in the cortical flow, and it closely followed the corresponding conventional PD signal for the duration of the acquisition. No external stimuli were applied during the scans. Normalized cross-correlation between the two signals was 0.991.

Ultrasonic flow imaging remains susceptible to cluttered imaging environments, which often results in degraded image quality. Coherent Flow Power Doppler (CFPD)–a beamforming technique–has demonstrated efficacy in addressing sources of diffuse clutter. CFPD depicts the normalized spatial coherence of the backscattered echo, which is described by the van Cittert-Zernike theorem. However, the use of a normalized coherence metric in CFPD uncouples the image intensity from the magnitude of the underlying blood echo. As a result, CFPD is not a robust approach to study gradation in blood echo energy, which depicts the fractional moving blood volume. We have developed a modified beamforming scheme, termed power-preserving Coherent Flow Power Doppler (ppCFPD), which employs a measure of signal covariance across the aperture, rather than normalized coherence. As shown via Field II simulations, this approach retains the clutter suppression capability of CFPD, while preserving the underlying signal energy, similar to standard power Doppler (PD). Furthermore, we describe ongoing work, in which we have proposed a thresholding scheme derived from a statistical analysis of additive noise, to further improve perception of flow. Overall, this adaptive approach shows promise as an alternative technique to depict flow gradation in cluttered imaging environments.

Ultrasound Computed Tomography is a very promising medical imaging technology to be used to discover breast cancer early. The conventional ultrasound emission method (fan beam), which utilizes a single element for one emission, might result in a signal-to-noise ratio (SNR) too low for measuring dense breasts. This research proposes a virtual fan emission method that can maintain high accuracy, a large field of view, and a high SNR at the same time, using multiple elements while mimicking the wave field of single element emission. We experimentally proved its effectiveness in improving SNR by imaging a phantom with high attenuation to mimic a dense breast. Imaging of excised human breast tissues also suggested that the proposed virtual fan beam emission is more effective than conventional fan beam emission to screen for breast cancer correctly.

For Ultrasound Tomography reflectivity imaging Synthetic Aperture Focusing Technique (SAFT) is often applied. Phase aberration correction is required to achieve images with high resolution and high contrast, for which a sound speed map is required. For USCT systems these sound speed maps are usually reconstructed using the transmission data from the raw data set, which is also used for reflection tomography. We compare straight and bent ray phase aberration correction SAFT algorithms with respect to different reconstruction algorithms to derive the sound speed map. Evaluations are carried out based on a simulated phantom and measured data from the Multimodal Ultrasound Breast Imaging System (MUBI). Phase aberration correction enables recovering the contrast of the image, while without SAFT results in considerably unfocused inner structures. By applying a reconstructed sound speed map however the local contrast cannot be fully recovered compared to the ideal case. Introducing bent ray transmission reconstruction approaches based on the Fast Marching or B´ezier curve method in all cases improves the results over the straight ray transmission tomography.

The challenge of ultrasound tomography in the presence of high impedance contrast is well known. We have successfully used full 3D transmission inverse scattering and refraction corrected reflection tomography to create 3D high-resolution images of the human breast. However, these tissues do not encompass the high contrast that occurs in orthopaedics scenarios, such as the human knee, where cranial and trabecular bone are present. Even though the high contrast of the bone is problematic for model based iterative reconstruction methods, we successfully image the tissue near, and in, the Femur-Tibia (F-T) space using an adapted QT Ultrasound Scanner and adapted inverse scattering algorithm. We show preliminary reconstructions of a cadaver knee that indicates that we can quantitatively and accurately image proximal soft tissue structures. We give correlations between MR images and QT Ultrasound transmission images that show correlation with known structures: besides the femur, tibia, and fibula, we see the condyle structures (medial and lateral), medial and lateral menisci internal to the F-T space, collateral ligaments, infrapatellar fat pad (Hoffa’s pad), patellar ligament, and various ligaments, tendons and musculature in the leg above and below the knee. We establish that a substantially different reconstruction protocol (than that of the breast) for 3D inverse scattering is required to obtain these images and we discuss the implications of these changes. These preliminary results show that high resolution of clinically relevant tissue is feasible with ultrasound tomography even within the F-T space.

Ultrasound computed tomography his paper designs and implements a high throughout, extensible and flexible ultrasound excitation and data acquisition system that transmits sustained high-speed ultrasound data to the server by Ethernet technology. The system is mainly used for the second-generation ultrasound computed tomography system designed in the medical ultrasound lab, but can also be utilized by other types of ultrasound imaging systems. The system consists of one or more ultrasonic excitation and acquisition boards. Each board includes multiplexing switches, pulse generators with T/R switches, analog front ends, analog-to-digital converters, and an FPGA, and can be used to excite a 256-element probe to transmit and receive ultrasound signals. The peak and the average bandwidth of one single board are 44.8Gbps and 4Gbps, respectively. Potential users can combine several excitation and acquisition boards to build high-end ultrasound imaging systems. The system has been applied to upgrade our ultrasound computed tomography system.

The potential of ultrasound tomography has been noticed to quantify the tissue acoustic properties for advanced clinical diagnosis. However, the location of most of the human anatomies limits the tomography for a few angles that leads the reconstruction as a more challenging problem. In this work, a deep convolutional neural networks- based technique is presented to estimate the speed of sound of tissue from a limited angle projection data. The underlying concept is based on filtered back projection technique, where the convolutional neural network is used to model the high-pass filter before the back projection. Moreover, we use a post convolutional neural network to suppress the artifacts generated due to the limited angle tomography. We train the network from a set of simulation experiments; on the test set consisting of 1,750 simulation experiments, we achieve an average mean absolute error of 2.1% in predicting the speed of sound map.

Ultrasound computed tomography (USCT) is a 3D imaging tool, especially for breast screening. Sound-speed tomography as one imaging modal of USCT is widely studied by researchers because of its great clinical potential for early breast cancer detection. Sound-speed reconstruction methods include ray-based methods and wave-based methods. In this study, a ray-based method for sound speed reconstruction: Fresnel volume tomography (FVT) is implemented. We use Limitedmemory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) optimization algorithm to solve the large and sparse equation for the inversion step. Considering the great computation burden in the L-BFGS inversion process, two kinds of acceleration schemes: CPU parallel and GPU parallel schemes are used and evaluated by in vitro experiment. The corresponding acceleration ratios are 5.3 and 18.6 for the 512×512 sound speed image reconstruction, compared to CPU serial computation.

In order to improve resolution, transducers with larger aperture size are desirable in ultrasound imaging. However, the practical aperture size is limited. Inhomogeneities and aberrating layers cause phase errors restricting the improvements provided by large arrays. Recently, we have shown the feasibility of a fully coherent multitransducer ultrasound imaging system (CMTUS), formed by several ultrasound transducers that are synchronized and freely located in space. The transducer locations along with the average speed of sound in the medium are deduced by maximizing the coherence between backscattered echoes from targeted point-like scattereres in the common field of view of the transducers. An improved image is obtained through coherent combination of the multiple transducers acting as a single larger effective aperture. In this study, the behavior of CMTUS in the presence of aberration is further investigated with simulations. A parametric study is presented, in which the geometry of the system, defined by two linear arrays, and the presence of acoustic clutter are investigated. In this framework, typical image quality metrics - resolution, contrast and contrast-to-noise ratio - are evaluated. Results suggest that the imaging enhancement made by the CMTUS is limited by the location of the transducers in space. Based on image metrics, an optimal spatial location is proposed for a CMTUS formed by two linear arrays. In addition, results show that, in the presence of clutter, image quality diminishes at larger aperture sizes. Nevertheless, CMTUS successfully corrects the aberration effects, without degrading the gains made by the large effective aperture.

We investigated using deep neural networks (DNNs) to beamform ultrasound images with high dynamic range targets. The DNNs operated on frequency domain data, the inputs consisted of the separated in-phase and quadrature components observed across the aperture of the array, and the outputs of the DNNs had the same structure as the inputs. We compared several methods for generating training data, including training with hypoechoic and anechoic cysts. All training data was generated using a linear ultrasound simulation tool. The results demonstrate the potential for using DNN beamformers to extend the dynamic range of ultrasound beamforming.

A delay-and-sum beamformer implementation for 3D imaging with row-column arrays is presented. It is written entirely in the MATLAB programming language for flexible use and fast modifications for research use, and all parts can run on either the CPU or GPU. Dynamic apodization with row-column arrays is presented and is supported in both transmit and receive. Delay calculations are simplified compared to previous beamformers, and 3D delay and apodization calculations are reduced to 2D problems for faster calculations. The performance is evaluated on an Intel Xeon E5-2630 v4 CPU with 64 GB RAM and a NVIDIA GeForce GTX 1080 Ti GPU with 11 GB RAM. A 192+192 array is simulated to image a volume of 96-by-96-by-45 wavelengths sampled at 0.3 wavelength in the axial direction and 0.5 wavelength in the lateral and elevation directions giving 5.53 million sample points. A single-element synthetic aperture sequence with 192 emissions is used. The 192 volumes are beamformed in approximately 1 hour on the CPU and 5 minutes on the GPU corresponding to a speed-up of up to 12.2 times. For a smaller beamforming problem consisting of the three center planes in the volume, a speed-up of 4.6 times is found from 109 to 24 seconds. The GPU utilization is around 5.0% of the possible floating point calculations indicating a trade-off between the easy programming approach and high performance.

When using ultrasound to image heterogeneous media, echoes from multiple and off-axis scattering can overwrite the recorded ballistic wavefronts of interest. This reduces the coherence of signals across the aperture and causes clutter in the final image. Therefore, separating those unwanted events from the signal of interest is necessary to improve the visibility of structures in a B-mode image, and also to enable other processing methods that require coherent channel signals, such as various phase-aberration-correction techniques and sound-speed estimators. We used prediction-error filters (PEFs) to model the signal and the assumed additive noise in the data acquired through a 10 mm thick layer of beef tissue placed above a speckle region of a phantom. The PEF coefficients used to model the signal were first computed from the phantom data collected without tissue and subsequently employed to deconvolve the tissue data and find the PEF associated with the noise. These two filters were then used in a joint-inversion framework to separate the signal and noise components recorded within the original tissue data. In order to be able to apply our method in scenarios where direct measurements of the signal proxy are not available, we also evaluated the signal-PEF coefficients from the theoretical model of the signal from diffuse targets as provided by the van-Cittert Zernike (VCZ) theorem. To evaluate the quality of the separation of signal from the noise, we compared the original channel data acquired through the tissue with its estimated ballistic-wave component, as well as their corresponding spectra. We also compared performance of the proposed technique to F-X filter, which is a popular linear-predictionbased filter used to suppress noise in channel data. After the removal of acoustic noise from the channel data, coherence across the aperture increases. The average nearest-neighbor cross-correlation computed on the original data is 0.47, while the nearest-neighbor cross-correlation of the estimated ballistic-wave component is 0.81 or 0.97, depending whether the experimental or theoretical signal-PEFs are used in the estimation process.

Kidney stones are often poorly visualized with ultrasound despite the fact that they have a large impedance mismatch. In previous kidney stone studies conducted by our group, we demonstrated that the Mid-Lag Spatial Coherence (MLSC) beamforming method suppresses the incoherent background speckle while enhancing coherent scatterers. This allowed kidney stones to be highlighted. To study this approach in more detail Field-II simulations and in-house phantoms containing kidney stones were used to test the effectiveness of MLSC with different parameters. The number of lags used during beamforming and the brightness of the point target were altered. Then, the CNR, SNR, CR, and PSNR of the phantoms and simulations were compared. The CNR experienced little change between lag ranges, but the SNR and PSNR increased with the start lag. SNR increased by 12.9% ± 2.9% between the lowest and highest lag range while PSNR increased by 27.9% ± 4.6% between the lowest and highest lag range. CR did not change in a regular pattern but remained consistently higher than delay and sum beamforming. We also compare MLSC against short-lag spatial coherence (SLSC) and show that we also see improvements over this method including an increase of MLSC over SLSC ranging between 250% and 401% for PSNR and between 414% and 879% for CR.

Calculating left ventricular ejection fraction (LVEF) accurately is crucial for the clinical diagnosis of cardiac disease, patient management, or other therapeutic treatment decisions. The measure of a patient's LVEF often affects their candidacy for cardiovascular intervention. Ultrasound (US) is one of the imaging modalities used to non-invasively assess LVEF, and it is the most common and least expensive. Despite the advances in 3D US transducer technology, only limited US machines are equipped with such transducer to enable true 3D US image acquisition. Thus, 2D US images remain to be widely used by cardiologists to image the heart and their interpretation is inherently based on two dimensional information immediately available in the US images. Past knowledge indicates that visual estimation of the LVEF based on the area changes of the left ventricle blood pool between systole and diastole (as depicted in 2D ultrasound images) may significantly underestimate the ejection fraction, rendering some patients as suitable candidates for potentially unnecessary interventions or implantation of assistive devices. True LVEF should be calculated based on changes in LV volumes, but equipment and time constraint limit the current technique to assess 3D LV geometry. The estimation of the systolic and diastolic blood pool volumes requires additional work beyond a simple visual assessment of the blood pool area changed in the 2D US images. Specifically, following the manual segmentation of the endocardial LV border, 3D volume would be assessed by reconstructing a LV volume from multiple tomographic views. In this work, we leverage on two idealized mathematical models of the left ventricle | a truncated prolate spheroid (TPS) and a paraboloid geometric model to characterize the LV shape according to the range of possible dimensions gathered from our patient-specific multi-plane US imaging data. The objective of this work is to reveal the necessity of calculating LVEFs based on volumes by showing that LVEF estimated using area changes underestimate the LVEF computed using volume changes. Additionally, we present a method to reconstruct the LV volume from 2D blood pool representations identified in the multi-plane 2D US images and use the reconstructed 3D volume throughout the cardiac cycle to estimate the LVEF. Our preliminary results show that the area-based LVEF significantly underestimates the true volume-based LVEF across both the theoretical simulations using idealized geometric models of the LV shape, as well as the patient-specific US imaging data. Specifically, both the TPS and paraboloid model showed an area-based LVEF of 41:3±4:7% and a volume-based LVEF of 55:4±5:7%, while the US image data showed an area-based LVEF of 34:7 ± 11:9% and a volume-based LVEF of 48:0 ± 14:0%. In summary, the area-based LVEF estimations using both the idealized TPS and paraboloid models was 14.1% lower than volume- based LVEF calculations using corresponding models. Furthermore, the area-based LVEF based on reconstructed LV volumes are 13.3% lower than volume-based estimates. Evidently, there is a need to further investigate a method to enable practical volume-based LVEF calculations to avoid the need for clinicians to estimate LVEF based on visual, holistic assessment of the blood pool area changes that improperly infer volumetric blood pool changes.

We created a new high resolution (50-MHz) three-dimensional ultrasound biomicroscopy (3D-UBM) imaging system and applied it to the measurement of iridoconeal angle, an important biomarker for glaucoma patients. Glaucoma, a leading cause of blindness, often results from poor drainage of the fluid from the eye through structures located at the iridiocorneal angle. Measurement of the angle has important implications for predicting the course of the disease and determining treatment strategies. An angle measured at a particular location with conventional 2D-UBM can be biased due to tilt in the hand-held probe. We created a 3D-UBM system by automatically scanning a 2D UBM with a precision translating stage. Using 3D-UBM, we typically acqure several hundred 2D images to create a high-resolution volume of the anterior chamber of the eye. Image pre-processing included intensity based frame-to-frame alignment to reduce effects of eye motion, 3D noise reduction, and multi-planar reformatting to create rotational views along the optic-axis with the pupil at the center, thereby giving views suitable for measurement of the iridiocorneal angle. Anterior chambers were segmented using a semantic-segmentation convolutional neural network, which gave folded “leave-one-eye-out” accuracy of 98.04%±0.01%, sensitivity of 90.97%±0.02%, specificity of 98.91%±0.01%, and Dice coefficient of 0.91±0.04. Using segmentations, iridiocorneal angles were automatically estimated using a modification of the semi-automated trabecular- iris-angle method (TIA) for each of ∼360 rotational views. Automated measurements were compared to those made by four ophthalmologist readers in eight images from two eyes. In these images, an insignificant difference (p = 0.996) was shown between readers and automated results.

We propose a method to automatically segment prostate from TRUS image based on multi-derivate deeply supervised network and multi-directional contour refinement. 3D multi-derivate V-Net is introduced to enable end-to-end segmentation. Deep supervision mechanism is integrated into the hidden layers to cope with the optimization difficulties when training such a network with limited training data. The probability map of new prostate contour is generated by the well-trained network and fused to reconstruct the prostate contour by multi-directional contour refinement. This proposed algorithm was evaluated using 30 patients’ data with TRUS image and manual contours. The mean Dice similarity coefficient (DSC) and mean surface distance (MSD) were 0.92 and 0.60 mm, which demonstrate the high accuracy of the proposed segmentation method. We have developed a novel deep learning-based method demonstrated that this method could significantly improve contour accuracy especially around the apex and base region. This segmentation technique could be a useful tool in ultrasound-guided interventions for prostate-cancer diagnosis and treatment.

Fusion-based ultrasound (US)-guided biopsy in a breast is challenging due to the high deformability of the tissue combined with the fact that the breast is usually differently deformed in CT, MR, and US acquisition which makes registration difficult. With this phantom study, we demonstrate the feasibility of a fusion-based ultrasound-guided method for breast biopsy. 3D US and 3D CT data were acquired using dedicated imaging setups of a breast phantom freely hanging in prone position with lesions. The 3D breast CT set up was provided by Koning (Koning Corp., West Henrietta, NY). For US imaging, a dedicated breast scanning set up was developed consisting of a cone-shaped revolving water tank with a 152- mm-sized US transducer mounted in its wall and an aperture for needle insertion. With this setup, volumetric breast US data (0.5×0.5×0.5 mm³ voxel size) can be collected and reconstructed within 3 minutes. The position of the lesion as detected with breast CT was localized in the US data by rigid registration. After lesion localization, the tank rotates the transducer until the lesion is in the US plane. Since the lesion was visible on ultrasound, the performance of the registration was validated. To facilitate guided biopsy, the lesion motion, induced by needle insertion, is estimated using cross-correlation-based speckle tracking and the tracked lesion visualized in the US image at an update frequency of 10 Hz. Thus, in conclusion a fusion-based ultrasound-guided method was introduced which enables ultrasound-guided biopsy in breast that is applicable also for ultrasound occult lesions.

Despite its high success rate, transcatheter aortic valve implantation (TAVI) is still associated with numerous complications, among which paravalvular regurgitation (PVR, which has been associated to long-term outcome. Assessment of PVR remains challenging in clinical routine; it lacks a solid reference method and the clinically used echocardiographic-based parameters have numerous limitations. Moreover, the development/validation of novel automated PVR quantification methods is hampered by the absence of ground truth data. In this sense, this study proposes the use of an ultrasound simulation-based pipeline to generate synthetic 3D color Doppler ultrasound images that mimic retrograde blood flow in cases of PVR in TAVI patients. These synthetic volumes are created using flow fields obtained from computational fluid dynamics (CFD) simulations, in which the underlying regurgitant volume (RV) is known. Besides the CFD-based flow field, the pipeline requires also an anatomical model of the aortic tract wall to simulate both B-mode and color Doppler volumes. The presented pipeline was used to generate twenty simulated volumes mimicking transesophageal echocardiographic images of cases with distinct levels of PVR severity, showing a visually similar appearance to clinical images. Interestingly, PVR severity scores estimated from the synthetic volumes correlated well with the known CFD-based RV, as well as against post-implantation angiography-based severity scores. Overall, these results demonstrate the pipeline’s potential to generate synthetic images to be used for the validation of automated PVR quantification software and/or other clinical PVR-related studies.

For computer-assisted interventions in orthopedic surgery, automatic bone surface delineation can be of great value. For instance, given such a method, an automatically extracted bone surface from intraoperative imaging modalities can be registered to the bone surfaces from preoperative images, allowing for enhanced visualization and/or surgical guidance. Ultrasound (US) is ideal for imaging bone surfaces intraoperatively, being real-time, non-ionizing, and cost-effective. However, due to its low signal-to-noise ratio and imaging artifacts, extracting bone surfaces automatically from such images remains challenging. In this work, we examine the suitability of deep learning for automatic bone surface extraction from US. Given 1800 manually annotated US frames, we examine the performance of two popular neural networks used for segmentation. Furthermore, we investigate the effect of different preprocessing methods used for manual annotations in training on the final segmentation quality, and demonstrate excellent qualitative and quantitative segmentation results.

The wave equation is linear, and it scales in time and space. As a consequence, wave phenomena that occur during fractions of a millisecond in human tissue often have a close correspondence in waves travelling for hours through the interior of the Earth. The scale invariance of the wave equation is the foundation for collaboration and technology transfer between medical ultrasound and seismic imaging - the promotion of which is the main goal of this contribution. In the first part of our presentation, we review the current state of the art in seismic imaging, with a focus on regional to global scales. Special emphasis will be on (1) high-performance modelling of seismic wave propagation through a heterogeneous, attenuating and anisotropic Earth, (2) the nature of seismic data and the resulting characteristics of the inverse problem, (3) recent images of 3D deep-Earth structure, and (4) future challenges in the field. In the second part, we highlight efforts to translate techniques from seismic imaging to medical ultrasound. This includes optimal design to position transducers, finite-frequency traveltime tomography to image out of plane, reverse-time migration, and 3D multi-parameter full-waveform inversion. Finally, we discuss several non-mathematical challenges that still impede technology transfer, and that remain to be addressed. These include the acceptable time to solution, and the ability of well-trained radiologists and seismic interpreters to handle entirely new types of images (and artifacts).

There is a need to provide better imaging methods for infants as there are few good options. CT can provide reasonable image quality with limited soft tissue contrast at a cost of large radiation dose. MRI can provide better soft tissue contrast, but the small size of an infant produces poor signal to noise and thus long scan times. Both types require anesthesia, which carries a substantial mortality risk for young patients and especially sick ones. Ultrasound imaging has been principally relegated to relatively simple applications in in orthopedics and diagnostics due to the inability to achieve high resolution at depth in complex structures. Quantitative Transmission (QT) Ultrasound relies on low frequency information which has greater penetrating power and 3D Inverse Scattering to produce high resolution and contrast at substantial depth. We built a prototype device for imaging small animals and tested the performance on 7-10lb piglets to simulate the conditions necessary to scan a newborn infant human. Image acquisition was entirely conventional with the currently available QT ultrasound breast imagers, but reconstruction required significant modification to deal with the additional complexity. We report on the changes in methods as well as the preliminary performance of the system in this configuration.

In neurosurgeries, brain shift and tumor removal may render the preoperative MRI diffusion tensor scan irrelevant. There is a need for real time and accurate mapping of the brain white matter fibers. Towards solving this problem, for the first time, we demonstrate the feasibility of ultrasound backscatter tensor imaging (BTI) in locating and measuring the corpus callosum (CC), the largest white matter fibers, in an ex vivo formalin-fixed rat brain. BTI analyzes the coherence in the backscatter signal at different transducer-to-fiber orientations to estimate regions of high anisotropy and thus the presence of fibers. We collected ultrasound radio frequency signals for 180° with a step-size of 10° . At each step, we used focused ultrasound beams to scan the central axis of the rat transverse plane. We then calculated and mapped the coherence factors (CF) to infer the size and location of the CC at two locations of high and low anisotropy, respectively. Lastly, we compared our results in the high anisotropy plane to a rat brain MRI Diffusion Tensor Imaging (DTI) atlas. The CC thickness in the measured plane was 0.87 mm (atlas) vs. 1.0±0.3 mm (CF map), while the distance to the rat brain medial dorsal surface was at 1.53 mm (atlas) vs. 1.7±0.3 mm (CF map). This is an ongoing study with limitations in the axial and lateral resolution, speed of acquisition, and signal to noise ratio. To our best knowledge, this is a first study in demonstrating the potential of BTI in detecting the corpus callosum with promising results to warrant further efforts towards clinical translation.

In our experiments, a technique has been developed to simultaneously acquire Electroacoustic (EA) signal captured by a single channel ultrasonic traducer in a linear array structure. The system utilizes micro- to nano-second pulsed electric field applied by an excitation source that can be used for clinical purposes (i.e. electroporation applications), and a conventional ultrasound transducer to acquire pulse electric field-induced acoustic signals. In this research, for the first time, we present a new real-time imaging-based technique when applying different electric field distributions that disturb electrically charged particles in the media that leads to a change of temperature which increases on the order of mK per a single high intensive, μ𝑠 −𝑛𝑠 short-pulse. We demonstrated this new technique by acquiring real-time acoustic signals induce by electric field distribution inside an agar-conductive based phantom. We used low-noise-amplifiers with a maximum gain of 60dB at 500 kHz with a linear scanning structure within less than 20sec, in 500 steps, and delay time of 500 ms to stabilize the transducer, and establish a linear scanning with a single element transducer. The corresponding EA images are reconstructed with a multi-step line back-projection algorithm. The approach can effectively reduce the artifacts associated with a conventional filter back projection algorithm used in other ultrasonic imaging by linear scanning structure because it is able to take information at multiple points to deliver the best possible image. This EAT technique provides a new and unique imaging approach for realtime, in-situ electrotherapy-based clinical practical applications.

Ultrasound Computed Tomography (USCT) of soft biological tissues today provides images with a high-level of resolution. The signal acquisition system using multichannel and/or multifrequency arrays performs in circular mode and the main (linear) inversion algorithms are based on compression wave propagation modeling. The main limits of these methods for bone imaging are due to the large impedance contrast between tissues, and to propagative phenomena generated through periosteal interfaces (mode conversion, attenuation). The linear inversion methods fail to provide high-level resolution images. Despite their performance and robustness, the non-linear methods are still today unsuitable for clinical applications because of the high computation time required. However, in the special case of children bone imaging, acquisition steps must be as fast as possible, with short-time exposure and low-intensity sonication. In this context, we have developed a fast-acquisition setup (1 sec.) based on a cylindrical-focusing ring antenna, and a protocol (< 5 sec.) using classical Born approximation and spatial Fourier transform. Unfortunately, the result today is a poor contrast-to-noise ratio (CNR) image. Previous work done to improve CNR used signal and image processing. This work focuses on this last point, and an automatic edge detection procedure, using Haar wavelet 2D-decompositon, combining k-means and Ostu algorithms. Results will be presented on ex vivo real bone samples and on geometrical mimicking bone phantom (SawbonesTM). An example of bone defect imaging will be presented and discussed.

One of the most powerful approach in ultrasound tomography (UT) is making use of distorted Born iterative (DBI) method to reconstruct high quality image in order to help locate and identify tumors more precisely. Due to its iterative nature, it begins with Born approximation as the initial guess. Then, it makes use of the inhomogeneous Greens function, as the kernel function, to alternatively calculate the total field for the forward problem and the scattering function for the inverse problem. One principal computational problem involved is that inverse problem is ill-posed, which will result in divergence of the DBI method if inappropriate regularization is used. This paper presents the regularization with truncated total least square (TTLS) where the adaptive algorithm is used to choose the regularization parameter in each iteration of DBI instead of using a fixed truncated value in all the iterations. In order to prevent the solution from being contaminated by noise, adaptive algorithm truncates the smallest singular values while minimizing the loss of signal obtained from transducers. Numerical simulations demonstrate that the proposed adaptive algorithm in conjunction with TTLS outperform TTLS with fixed truncation parameter by effectively reducing the noise and minimizing the relative error.

Problem: The gold standard for prostate cancer diagnosis is B-mode transrectal ultrasound-guided systematic core needle biopsy. However, cancer is indistinguishable under ultrasound and thus additional costly imaging methods are necessary to perform targeted biopsies. Speed of sound is a potential biomarker for prostate cancer and has the potential to be measured using ultrasound tomography. Given the physical constraints of the prostate’s anatomy, this work explores a simulation study using deep learning for limited-angle ultrasound tomography to reconstruct speed of sound. Methods: A deep learning-based image reconstruction framework is used to address the limited-angle ultrasound tomography problem. The training data is generated using the k-wave acoustic simulation package. The general network structure is composed of a series of dense fully-connected layers followed by an encoder and a decoder network. The basic idea behind this neural network is to encode a time of flight map into a lower dimension representation that can then be decoded into a speed of sound image. Results and Conclusions: We show that limited-angle UST is feasible in simulation using an auto-encoder-like DL framework. There was a mean absolute error of 7.5 ± 8.1 m/s with a maximum absolute error of 139.3 m/s. Future validation on experimental data will further assess their ability in improving limited-angle ultrasound tomography.

Presentation of images similar to a new unknown lesion as a reference can be helpful in medical image diagnosis and treatment planning. We have been investigating a method to determine similarity of breast masses as an image retrieval index for an intelligent image analytic system that may support radiologists’ efficient image interpretation. In order to retrieve perceptually similar images, we have obtained subjective similarity ratings from expert radiologists, which were then used in similarity space modeling and training deep neural networks. In this study, we investigated the use of convolutional neural network to model the similarity space for retrieval of diagnostically relevant reference images and also to directly estimate similarity ratings for pairs of images. The preliminary results show that retrieval performance was slightly better in similarity space modeling method than direct estimation method. These results indicate the potential usefulness of the proposed methods for retrieval of reference images as diagnostic assistance.

Ultrasound Computer Tomography (USCT) is a novel and a promising low-cost technology for breast cancer imaging. USCT can provide speed-of-sound, attenuation and reflectivity information about the analyzed area. Among USCT methods, the ray-based approach (transmission tomography) is faster and it can offer a-priori information as initialization for more complex USCT methods. To contribute to the generation of such initializations, this work presents, on simulated media, how the use of a spatial filter jointly with refraction can improve the image reconstructed by transmission USCT. The following were used: k-wave toolbox for data generation on a heterogeneous medium; a numerical phantom with objects whose speed of sound are between [1460−1570]m/s; all objects have equal density and attenuation; a Refraction Tomography as transmission reconstruction algorithm (with and without regularization), where the forward problem was solved by Dijkstra’s shortest-path algorithm and the inverse problem with Simultaneous Iterative Reconstruction Technique; an alternative of the Modified Median Filter (m-MMF) as a spatial filter. Two sets of data were generated using 100 and 192 transducers (400kHz) respectively. All transducers were uniformly distributed around the medium. Evaluations were based on Relative-Residual-Error (RRE), normalized-root-mean-square-error (NRMSE) and structural similarity (Q). Comparing the reconstructions, better performance (NRMSE < 0.01) was found when the filter was applied with the Refraction Tomography, regardless of the use or not of the regularization. As expected, the reconstructions improved their performance as the number of transducers was increased. The results suggest that it is possible to improve and obtain satisfactory reconstructions via Transmission reconstruction algorithm in conjunction with a spatial filter.

In order to improve the efficacy of cancer treatment, many new therapy methods have been proposed and tested. The purpose of this study is to investigate the feasibility and potential advantages of using a low-cost, portable and easy-touse ultrasound imaging modality to quantitatively assess treatment efficacy and/or identify optimal treatment methods. For this purpose, we developed a new interactive computer-aided detection (CAD) scheme based image segmentation and feature analysis scheme, which extracts quantitative image features from ultrasound images of athymic nude mice embedded with tumors. Twenty-three mice were involved in this study and treated using 7 different thermal therapy methods. The longitudinal ultrasound images of mice were taken pre- and post-treatment after 3-days of tumor embedment. A graphic user interface (GUI) of the CAD scheme allows manual segmentation of the tumor regions depicting on the images. Two CAD-computed tumor image feature pools were then established including the features computed from (1) pre-treatment images only and (2) difference between post- and pre-treatment images. Through data analysis, a number of top image features were identified to predict the effectiveness of treatment methods. Pearson Correlation coefficients between two top features selected from above two feature pools versus tumor size increase ratio were 0.373 and 0.552, respectively. Using an equally weighted fusion method of the top two features computed from pre- and post-treatment images, correlation coefficient increased to 0.679. Study results demonstrated the feasibility of extracting a new quantitative imaging marker from ultrasound images to assist in the evaluation of treatment efficacy or tumor response to the treatment.

The new technique for the imaging guidance to real-time monitoring of electroporation-based medical interventions could be based on the electroacoustic tomography (EAT), where the electric field applied for the electroporation process leads to induced acoustic signals based on the flow of electrical current inside the target conductive tissue. A microsecond to nanosecond electric-pulse (𝜇𝑠 − 𝑛𝑠𝐸𝑃) excitation source is an essential part of this new imaging guided to real-time monitoring electroporation process. This paper presents the design, configuration, and measurement of a compact, low-cost high voltage MOSFET-based pulsed excitation source and the simple structure of the EAT system with the single channel ultrasonic transducer to acquire acoustic signals and complemented by experimentation of its function based on agar-conductive phantom studies. The high-voltage pulsed excitation source has variable pulse widths ranging from 100 𝑛𝑠 𝑡𝑜 10 𝜇𝑠 electric pulse with a variable pulse potential magnitude of up to 1200 Volts (V). The high-voltage 𝜇𝑠 − 𝑛𝑠𝐸𝑃 is powered from a variable input source of 11.3 to 16 V in direct current (DC) and a power controller using a 0 V to 5 V in DC power line so that it is able to provide 0 to full output in potential magnitude. The high-intensity, ultra-short pulsed electric field is then connected to two electrodes separated by a distance (d) where 𝑑 = 1500𝜇𝑚 𝑡𝑜 3400 𝜇 mounted into the conductive media. An unfocused ultrasound transducer with central frequency of 500 kHz is used to acquire real-time acoustic signals. Various conductive media, including two agar-based phantoms with conductivity of 1 𝜇𝑆/𝑐𝑚 , 34 𝑚𝑆/𝑐𝑚 were studied using this pulsed excitation source to induce corresponding acoustic signals. Results indicate feasibility of the enhancing the EAT system that used up to 8 kV/cm, 𝜇𝑠 𝑡𝑜 𝑛𝑠 pulsed excitation source used in the electroporation-based clinical processes enhancing the EAT system as an imaging guidance to real-time, in-situ monitoring for the electroporation-based techniques.

The methods for nondestructive subcellular imaging of cancer cells have attracted lots of interests. Scanning probe acoustic microscope (SPAM) has broad potential for such imaging, which could acquire information of the morphology as well as the internal structure signals. Neighborhood resonance imaging (NRI) could provide depth information in a high spatial resolution at nano-scale. However, the observations of breast cancer MDA-MB-231 cell via NRI are few. In this paper, we utilized SPAM to perform NRI for mapping MDA-MB-231 cells for specified characterization of the morphology and the internal structures. We verified the feasibility of applying NRI on cell imaging theoretically and experimentally. The simulation experiment demonstrated that NRI could succeed to image the cells with high-resolution. The experimental results illustrated that images acquired by NRI showed clear cell edges and complicated internal structure, compared to the traditional scanning acoustic imaging mode. NRI would build an important and solid basis for studying the morphology and internal structures of the cancer cells in a non-destructive way. In addition, our proposed method could be used to obtain the morphology and internal information in both solid and soft material wafers with the nano-resolution.

Multispectral photoacoustic (MPA) specimen imaging modality is proven successful in differentiating photoacoustic (PA) signal characteristics from a cancer and normal region. The oxy and de-oxy hemoglobin content in a human tissue captured in the MPA data are the key features for cancer detection. In this study, we propose to use deep 3D convolution neural network trained on the thyroid MPA dataset and tested on the prostate MPA dataset to evaluate this potential. The proposed algorithm first extracts the spatial, spectral, and temporal features from the thyroid MPA image data using 3D convolutional layers and detects cancer tissue using the logistic function, the last layer of the network. The model achieved an AUC (area under the curve) of the ROC (receiver operating characteristic) curve of 0.72 on the prostate MPA dataset.