This PDF file contains the front matter associated with SPIE Proceedings Volume 12932, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Ultrasound tomography (UST) is an emerging medical imaging modality that has found its way into clinical practice after its recent approval by the Food and Drug Administration (FDA) for breast cancer screening and diagnostics. As an active area of research, UST also shows promise for applications in brain, prostate, limb and even whole-body imaging. The historical development of ultrasound tomography is rooted in the idea of “seeing with sound” and the concept borrows heavily from diverse disciplines, including oceanography, geophysics and astrophysics. A brief history of the field is provided, followed by a review of current reconstruction methods and imaging examples. Unlike other imaging modalities, ultrasound tomography in medicine is computationally bounded. Its future advancement is discussed from the perspective of ever-increasing computational power and Moore's Law.

Focused cardiac ultrasound (FoCUS) is a medical technique that uses ultrasound to examine the heart at the bedside. To pave the way for an automatic analysis of FoCUS videos, the first step is to identify the imaged cardiac view. Previously developed methods have considered each frame separately and disregarded any temporal information, which is key due to FoCUS’ characteristic instability. Hereto, different strategies for video-level FoCUS view classification were investigated, all built on top of a ResNet-like network. In total, twelve networks were compared, from those sharing a spatial feature extractor to an architecture employing a spatiotemporal feature extractor. Overall, the latter led to the best performing model, highlighting a better ability to identify moving structures like heart valves, with a Matthews correlation coefficient of 0.9569.

Image segmentation of ultrasound videos is not an easy task because of the vagueness of ultrasound images, yet it is a helpful application for intra-surgery tumor detection and key organ protection. Previous ultrasound segmentation methods, such as UNet and Swin-UNetR focus on single image segmentation, and natural image video segmentation methods such as VisTR requires a large volume of GPU memory which is not easily trained and applied with limited calculation resources. In this paper, we put forward an ultrasound video segmentation framework called Ultra-TransUNet, which makes use of both temporal and spatial context, to segment ultrasound videos. Based on the Dice metric, our method improved the Dice and sensitivity performance from baseline methods for about 2.6% and 14.1%, respectively, averaged for 3 datasets. In this paper we evaluate our methods on phantom, animal and cadaver labs for segmenting two types of clinically relevant targets: tumors through phantom studies and the ureter in animal and cadaver labs. We hope our algorithm could provide with surgeons real-time support to locate key structures with ultrasound during surgeries, and thus protect patients and improve surgical outcomes.

Early image quality metrics were often designed with clinicians in mind, and ideally better metrics would correlate with the subjective opinion of clinically better images. Over time, adaptive beamformers and other post-processing methods have become more common, and these newer methods often violate assumptions of earlier image quality metrics, making the meaning of these metrics inaccurate at best. The result is the possibility of beamformers that can “manipulate” metrics to be better, while not producing clinically better images. In this work, Smith et al.’s SNR metric for lesion detectability¹ is considered, and a more robust version, here called generalized SNR (gSNR), is proposed that uses gCNR^{2, 3} as a core, and therefore is more robust to transformations and manipulations. SNR differs from gCNR in that it uses lesion size and spatial resolution as components of it’s calculation, which gCNR does not. It is analytically shown that for Rayleigh distributed data, gCNR can be written in terms of Smith et al.’s C_ψ (and therefore can be used as a substitution), and more robust methods for estimating the resolution cell size are considered. This allows for a robust estimate of lesion detectability based on estimated gCNR that may correlate with clinical assessments of image quality.

Partial flexor tendon tears are common, but their diagnosis presents a few challenges. The degree of a partial flexor tendon tear necessitating surgical intervention remains under debate. This is primarily due to the lack of a sensitive and accessible imaging modality to assess the depth of a tendon laceration. We suggest the use of three-dimensional ultrasound (3DUS) to provide a more accurate grading of partial flexor tendon tears by providing surgeons with a decision-making tool to identify when surgical intervention is necessary. As a proof of concept, we dissected 6 digits from 2 fresh-frozen cadaveric specimens. Each digit was imaged using 3DUS and MRI to determine if the tendons and tendon lacerations are identifiable in the 3DUS images. This will be performed in preparation for a larger trial where each digit will be randomly assigned to an intact, low-grade laceration, or high-grade laceration group. 3DUS images were collected from each digit. MR images were also collected from each digit to compare the sensitivity of each imaging modality. Further analysis of the images will require a trained radiologist to grade the lacerations in each image, while being blinded to the actual grade of each laceration and comparing the radiologist’s grading of each image to each laceration’s actual grading to assess the sensitivity of 3DUS and MR in detecting partial tendon lacerations. We anticipate that 3DUS imaging to be comparable to MR imaging in detecting partial tendon tears. Future work includes applying load to the affected fingers to investigate whether that will improve partial tendon laceration detection. This work expands on the applications of 3DUS in musculoskeletal imaging and provides clinicians with an accessible tool to accurately detect and grade partial tendon lacerations.

Prostate cancer ranks as the second most prevalent cancer among men globally. Accurate segmentation of prostate and the central gland plays a pivotal role in detecting abnormalities within the prostate, paving the way for early detection of prostate cancer, quantitative analysis and subsequent treatment planning. Micro-ultrasound (MUS) imaging is a novel ultrasound technique that operates at frequencies above 20MHz and offers superior resolution compared to conventional ultrasound, making it particularly effective for visualizing fine anatomical structures and pathological changes. In this paper, we leverage deep learning (DL) techniques for the segmentation of prostate and its central gland on micro-ultrasound images, investigating their potential in prostate cancer detection. We trained our DL model on MUS images from 80 patients, utilizing a five-fold cross-validation. We achieved Dice similarity coefficient (DSC) scores of 0.918 and 0.833, and an average surface-to-surface distance (SSD) of 1.176mm and 1.795mm for the prostate and the central gland, respectively. We further evaluated our method on a publicly available MUS dataset, achieving a DSC score of 0.957 and a Hausdorff Distance (HD) of 1.922mm for prostate segmentation. These results outperform the current state-of-the- art (SOTA).

The alignment of MRI and ultrasound images of the prostate is crucial in detecting prostate cancer during biopsies, directly affecting the accuracy of prostate cancer diagnosis. However, due to the low signal-to-noise ratio of ultrasound images and the varied imaging properties of the prostate between MRI and ultrasound, it’s challenging to efficiently and accurately align MRI and ultrasound images of the prostate. This study aims to present an effective affine transformation method that can automatically register prostate MRI and ultrasound images. In real-world clinical practice, it may increase the effectiveness of prostate cancer biopsies and the accuracy of prostate cancer diagnosis.

Acoustic Radiation Force Impulse (ARFI) Variance of Acceleration (evaluated as its decadic log and denoted, ‘log(VoA)’) imaging is a noninvasive ultrasound elastography method that differentiates tissue according to both its mechanical and acoustic properties. In application to stroke risk prediction, ARFI log(VoA) imaging has been demonstrated in pilot in vivo clinical studies for delineating carotid atherosclerotic plaque features that convey vulnerability to rupture. This presentation will review the diagnostic performance of ARFI log(VoA) imaging and highlight recent advancements to ARFI log(VoA) imaging, including its high-framerate implementation for efficient integration with vector Doppler data acquisition throughout the cardiac cycle, its combination with wall shear stress (WSS) measurement for a more complete assessment of plaque rupture potential, and its execution using a swept aperture for volumetric interrogation of plaque composition, structure, and WSS. Finally, a new framework for ground truth validation of experimental WSS measurement in human stenosed carotid arteries will be presented.

A recent technique named Mechanically-inspired L1-norm-based Second-Order Ultrasound eLastography (L1- MechSOUL) has provided a promising solution to the well-known issue of 2D tracking (both axial and lateral) in ultrasound strain elastography. This technique optimizes a cost function consisting of a data term, a mechanical congruence term, and first- and second-order continuity terms. However, L1-MechSOUL’s second-order regularizer considers only the unmixed second derivatives and disregards the mixed derivatives, which is a simplification that causes suboptimal noise suppression and inaccurate displacement estimation. We propose to address these challenges by formulating and optimizing a novel L1-norm-based second-order regularizer that penalizes both mixed and unmixed displacement derivatives. The unmixed second derivatives in the direction of displacement components (i.e., axial or lateral) regularize the normal strains, whereas we interpret the mixed second derivatives of the axial or lateral displacement to regularize both normal and shear strains. We compared the proposed technique against L1-MechSOUL using simulated and phantom datasets, resulting in improved mean structural similarity, elastographic signal-to-noise ratio, and elastographic contrast-to-noise ratio by up to 14%, 91%, and 132%, respectively. These quantitative improvements collectively highlight our ability to deliver high-accuracy 2D displacement and strain estimations that will advance the state-of-the-art in elastography-guided interventions.

Quasi-static ultrasonic elasticity imaging (QUSE) lacks necessary data to solve inverse 3-D elasticity problems mathematically. The autoprogressive method (AutoP) is a recently developed, data-driven approach to QUSE that leverages finite element physics and machine learning techniques primed for sensing complex relationships with limited data. The process involves propagating measured force and displacement data from ultrasound into stress-strain training data through finite element analysis (FEA) to train artificial neural networks (ANNs) to learn both constitutive behavior and spatial heterogeneity. Internal displacement data provides material information about the entire 3-D contiguous volume that is integrated into the learning process through FEA physics. To use FEA for tissues as a solid, we must assume they are somewhat compressible to avoid numeric instability, however this assumption introduces a bias into the training data that impacts the quality of the learned material properties. The trade-off between bias and numeric instability must be resolved to provide the highest quality training data to the networks. To understand this trade-off, force and displacement data were acquired from quasi-static compression of an incompressible, linear-elastic gelatin phantom with a stiff cylindrical inclusion. AutoP tests were performed using simulated and measured displacement data sets. For each data set, ANNs were initialized with stress-strain data from simulated compression of a uniform material with a Poisson’s ratio of 0.30, 0.40, and 0.45. We found improvement in the resulting Young’s modulus distribution using a Poisson’s ratio of 0.4 over 0.3, however using 0.45 resulted in non-physical behavior. We have developed an approach that balances numeric instability with training data bias and considers additional complications of sampling and meshing decisions to generate high-quality material property images. Discerning these trade-offs for linear-elastic materials is essential for the extension into multi-phasic, non-linear and time-dependent material property imaging with AutoP.

Cardiovascular disease (CVD) is the main cause of death worldwide. Cardiac adipose tissue (CAT) around the heart correlates with CVD risk. MRI and CT scans are preferred for CAT quantification. Ultrasound (US) is limited to linear CAT thickness measurements on the right ventricle. Comprehensive CAT volume and distribution quantification with US could aid CVD assessment. This study uses deep learning to automatically segment the left myocardium (LM), left epicardium (LE), and right epicardium (RE) boundaries in echocardiograms. A U-Net ResNet34 model was trained using 506 expert-traced images. Data was split into 70/10/20 training/validation/test sets. The model achieved Dice scores of 0.94, 0.89, and 0.88 for the three boundaries respectively. Regions-of-interest (ROIs) around the combined left and right epicardium contours were classified as containing CAT or not using spectral analysis of raw RF data. MRI expert-traced images of the same patients provided ground truth labels for CAT. A total of 102 corresponding US-MRI image pairs were matched using visual assessment and anatomical landmarks. ROIs were defined around the predicted epicardium contours, and for each ROI, 9 spectral parameters were computed as a random forest classifier input. A randomized 75/25 training/test split was used. The classifier achieved an average accuracy of 76%, sensitivity of 85%, and specificity of 74%. Results show the feasibility of performing automatic CAT assessment involving echocardiogram segmentation, contour detection, and classification of ROIs around the perimeter of the left and right epicardium in short-axis echocardiograms.

Ultrasound speckle tracking has the potential to quantify shear strain in patients with myofascial pain, leading to previous clinical conclusions (derived from a cross-correlation-based, windowed-search speckle tracking method) that shear strain is altered in patients with myofascial pain when compared to healthy individuals. However, differences in speckle-tracking techniques are known to impact the accuracy of estimated displacement fields, which has the potential to alter clinical conclusions. Therefore, this work aims to assess the reliability of two displacement estimation algorithms in the context of muscle lateral shear strain measurements, specifically Search and a Search-initialized Second-Order Ultrasound eLastography that we name SOUL-Search. Search utilizes a window-based displacement tracking method that determines the optimal displacement based on maximum signal correlation, whereas SOUL-Search fine-tunes the Search initial estimates by optimizing a cost function comprising data and regularization terms. A simulation study was implemented to provide controlled in silico assessments, followed by initial comparisons with in vivo data from the shoulder muscles of three research participants with myofascial post-stroke shoulder pain, acquired with a robot-held ultrasound probe with passive muscle motion induced by a bimanual arm trainer. SOUL-Search improved the in silico root-mean-square error, mean absolute error, and mean structural similarity of lateral displacement estimates by 64%, 43%, and 34%, respectively, relative to Search. When applied to the in vivo data, both Search and SOUL-Search agree with a novel clinical hypothesis that painful paretic shoulders exhibit altered muscle shear strain when compared to each participant’s non-paretic shoulder. However, SOUL-Search produced less-noisy lateral displacement images (enabling better qualitative assessments of clinical observations) and provided a smoother progression of lateral shear strain over time when compared to Search (which aligns with expectations based on the controlled in vivo environment). Results are promising for the creation and validation of new clinical hypotheses surrounding the identification and management of myofascial pain.

In mild hyperthermia (MHTh), targeted tumors are heated to approximately 41 to 44°C, typically to enhance chemo-, radiation, and/or immunotherapy. This study demonstrates the efficacy of a ring-array ultrasound (US) transducer in generating and monitoring MHTh in heterogeneous media. A ring-array provides enhanced focusing and heat localization compared to conventional linear-array US. We simulated scenarios using either one, two, or four 128-element linear-array transducers, or a 256-element ring-array transducer, to generate focused US profiles. The full width half maximum (FWHM) was measured to quantify the results. The ring-array achieved the most accurate and localized focal pressure profile with a focal spot size of 2×2mm. In a simulated heterogeneous breast model, sound speed (SS) compensation with a ring-array resulted in more efficient acoustic focusing than non-compensated fields. The focal point without SS compensation shifted spatially from its target due to acoustic aberrations, highlighting the importance of aberration correction for precise heat localization. Additionally, the ring-array's heat generation capability was evaluated in-vitro using a tissue-mimicking phantom, where the temperature at the focal point was increased by 6°C in 12 minutes and was sustainable. Lastly, the capability of a ring-array to track temperature was evaluated using US tomography (UST) in another in-vitro experiment. Here, a cylindrical inclusion in a tissue-mimicking phantom was filled with preheated water and allowed to passively cool from 45°C to 25°C. A ring-array tracked the temperature changes based on SS images with a mean error of 0.06±0.28°C. In summary, a ring-array transducer (1) achieved the best focal profile compared to standard linear transducers, (2) accomplished superior aberration correction using SS images, resulting in better aberration-free focusing in heterogeneous media, (3) generated sustainable heat at a focal point, and (4) accurately monitored temperature changes with UST.

Conventional reflectivity imaging with Ultrasound Tomography (USCT) reconstructs qualitative images proportional to the magnitude of the impedance gradient. We propose a method to additionally recover the scatter characteristics for each reconstructed voxel, i.e. whether a reflection is diffuse or specular. This is achieved using a modification of 3D Synthetic Aperture Focusing Technique (SAFT). Our novel approach separates the incoming and outgoing ultrasound intensity in each voxel according to the incident direction and the direction to the receiver. To reduce memory requirements, we propose several strategies for selecting a subset of data or aggregating data to predefined directions. The reconstruction leads to five-dimensional data, from which for each voxel in 3D space, a 2D scatter map can be derived. It can be interpreted as the distribution of energy which has been introduced from a certain direction and which has been reflected into a particular direction. We validated our approach by a simulation based on the Phong reflection model and perform first reconstructions of experimental data. Using the 2D scatter maps it is possible to distinguish specular from diffuse reflections visually. Extracting first-order statistics from the 2D scatter maps for each voxel can be a means to break down the 5D information to 3D for traditional slice-based visualization. The multidimensional data provided by our method may be used in future as a biomarker for diagnosis as e.g. a plain surface of a cyst may reflect the ultrasound differently than a rough surface of a spiculated mass.

Most ultrasound transducer designs are driven towards high frequencies by the need to maximize the resolution achievable by B-mode reflectivity imaging. Because full-waveform inversion (FWI) requires low frequencies to overcome cycle skipping, it is often difficult to apply FWI to the same high-frequency transducer hardware. Recent work on time-domain adaptive waveform inversion (AWI) addresses this based on a deconvolution approach to overcome cycle skipping. However, most clinical implementations of ultrasound tomography rely on frequency-domain FWI to quickly reconstruct images on clinically relevant time scales. Although the broadband nature of AWI makes it difficult to translate to the frequency domain, we can approximate the properties of AWI by using a frequency-differencing approach. We develop and describe both a model-domain and a data-domain method for frequency-differencing. Simulations and phantom experiments show that each frequency-differencing approach helps overcome cycle skipping by providing a sufficiently accurate starting model for FWI.

We present a full-waveform inversion (FWI) of an in-vivo data set acquired with a transmission-reflection optoacoustic ultrasound imaging platform containing a cross-sectional slice through a mouse. FWI is a high-resolution reconstruction method that provides quantitative images of tissue properties such as the speed of sound. As an iterative data-fitting procedure, FWI relies on the ability to accurately predict the physics of wave propagation in heterogeneous media to account for the non-linear relationship between the ultrasonic wavefield and the tissue properties. A key component to accurately predict the ultrasonic field numerically is a precise knowledge of the source characteristics. For realistic problems, however, the source-time function is generally unknown, which necessitates an auxiliary inversion that recovers the time series for each transducer. This study presents an updated sound speed reconstruction of a cross-section through a mouse using source wavelets that are inverted individually per transducer. These source wavelets have been estimated from a set of observed data by application of a source-wavelet correction filter, which is equivalent to a water-level deconvolution. Compared to previous results, the spatial resolution of anatomical features such as the vertebral column is increased whilst artefacts are suppressed.

Ultrasound computed tomography (USCT) is an emerging imaging modality that holds great promise for breast imaging. Full-waveform inversion (FWI)-based image reconstruction methods incorporate accurate wave physics to produce high spatial resolution quantitative images of speed of sound or other acoustic properties of breast tissues from USCT measurement data. However, FWI is computationally burdensome and requires a good initial guess of the speed of sound distribution due to the nonconvex nature of the underlying optimization problem (cycle-skipping). Alternatively, the use of a simplified linear model, such as the Born approximation, allows the image reconstruction problem to be formulated as a convex optimization problem, but sacrifices accuracy. This work proposes utilizing a convolutional neural network (CNN) to correct pressure data and accurately reconstruct images using a simplified forward model, thus combining the benefits of accurate reconstructions from traditional FWI methods with the reduced computational complexity of inversion with simplified models. Furthermore, applying this correction to the measurements before inversion avoids issues inherent to other deep learning reconstruction methods that first invert and then apply correction to the images. Specifically, correction in the measurement domain is well-defined by a mathematical model and avoids hallucinations by an improperly learned image prior. This reconstruction approach was validated with a set of anatomically realistic test images and compared to traditional reconstruction methods (FWI and uncorrected Born inversion), a data-driven learned reconstruction method, and a machine learning method for artifact correction in the image domain after reconstructing using an inaccurate physics model.

Stroke is a significant cause of mortality and disability in America. Due to differences in the treatment of ischemic and hemorrhagic stroke, imaging must be performed before administration of therapeutic medication. Unfortunately, the current standard imaging methods, namely CT and MRI, require specialized locations and staff, which can induce delays in triage, and therefore, treatment time. Recent work suggests that ultrasound tomography (UST) is capable of imaging in vivo tissue properties and may have potential as a diagnostic tool during stroke treatment which could be performed at the point of injury rather than at a local hospital. In this work, we investigate the feasibility of using UST imaging to image the brain via in silico, in vitro, and ex vivo studies. The results of this work indicate some of the challenges which must be overcome to effectively image in vivo stroke patients.

Constructing a physics-augmented digital twin of the skull is imperative for a wide range of transcranial ultrasound applications including ultrasound computed tomography and focused ultrasound therapy. The high impedance contrast as well as the acoustic-elastic coupling observed between soft tissue and bone increase the complexity of the ultrasound wavefield considerably, thus emphasizing the need for waveform-based inversion approaches. This work applies reverse time migration in conjunction with the spectral-element method to an in vitro human skull to obtain a starting model, which can be used for full-waveform inversion and adjoint-based shape optimization. Two distinct brain phantoms are considered where the cranial cavity of the in vitro human skull was filled with (1) homogeneous water and (2) gelatin with two cylindrical inclusions. A 2D slice through the posterior of the skull was collected using a ring-like aperture consisting of 1024 ultrasound transducers with a bandwidth of approximately 1MHz to 3MHz. Waveform-based reverse time migration was then used to resolve the inner and outer contours of the skull from which a conforming hexahedral finite-element mesh was constructed. The synthetically generated measurements which are obtained by solving the coupled acoustic-viscoelastic wave equation are in good agreement with the observed laboratory measurements. It is demonstrated that using this revised wave speed model for recomputing the reverse time migration reconstructions allows for improved localization of the gelatin inclusions within the cranial cavity.

Large apertures capable of real-time acquisition can offer increased field of view, enhanced imaging performances and reduced operator dependence. In this work, we are trying to bridge the gap between standard ultrasound and tomography by developing large apertures capable of real time imaging and blow flow characterization (Doppler). In the context of breast imaging, we are pursuing the development of a 2.5 MHz 1024-element half-ring geometry (200 mm diameter) to scan the entire breast in a few seconds with dry coupling (i.e. no water bath). The array is interfaced with a 1024 channels ultrasound platform (Verasonics) with a dedicated GPU processing pipeline.

We previously showed that domain adaptive deep neural networks (DNNs) can outperform delay-and-sum (DAS) beamforming in the context of abdominal imaging. We hypothesize the ability of our domain adaptive DNN framework to be applied to transthoracic echocardiography (TTE). We also propose architectural improvements, such as leveraging an encoder-decoder structure and skip connections, to further improve ultrasound image quality for echocardiography tasks such as the detection of thrombi in the left atrial appendage (LAA). DNN training data utilized simulated and in vivo cardiac data. Simulated anechoic and hypoechoic cysts with various amounts of clutter were generated through Field II and in vivo data was collected by scanning patients at Vanderbilt University Medical Center. Fundamental frequency TTE data from five separate cases were processed with DAS, ADMIRE, the baseline model, and multiple models with modified architectures. We found that even when varying the amount of training data, the DNNs consistently achieved higher generalized contrast-to-noise (gCNR) and contrast ratio (CR) but lower contrast-to-noise ratio when compared to DAS. The best-performing beamformer was one DNN with our architectural improvements, achieving higher average gCNR and CR values of .907 and 48.30 dB compared to the baseline DNN values of .788 and 39.45dB, and DAS values of .717 and 14.08dB. Our results demonstrate that our domain adaptive DNN can effectively be applied in the context of transthoracic cardiology, and an encoder-decoder architecture with skip connections can result in even more improvements. Further advancements may improve image quality even more.

Multi-line transmit (MLT) imaging enables the acquisition of high frame rate (HFR) data in ultrasound imaging, especially in echocardiography where capturing rapid events associated with heart motion can provide valuable information for disease diagnosis. MLT beams are generated by simultaneously transmitting multiple focused beams in different spatial directions within a single pulse-echo event. The main drawback of MLT imaging is the generation of crosstalk artifacts due to the interferences between multiple beams and targets, which will in turn decorrelate the backscattered echoes and will reduce the spatial coherence significantly, leading to poor image quality. In this study, we have investigated the effects of synthetic focusing on the overall coherence of the received signals. We have shown that achieving more accurate focusing due to the implementation of synthetic aperture beamformers could be less susceptible to the artifacts introduced with MLT, and this will in turn improve the coherence of the backscattered signals, resulting in an improved image quality. Simulation, phantom, and in-vivo experiments have been conducted to demonstrate that spatial coherence enhances as a result of synthetic focusing in MLT imaging (especially away from the transmit focus). Furthermore, we have implemented synthetic aperture methods together with coherence-based techniques to investigate their synergistic performance in further suppressing the incoherent backscattered echoes and improving target detectability. The results demonstrate that this provides considerable benefits in rejecting MLT crosstalk artifacts compared to the conventional dynamic receive focusing.

Pancreatic cancer is one of the deadliest cancer types with less than a 10% 3-year survival rate (SR). Radiotherapy (RT) is critical for achieving local control and enabling dose escalation, which can double and triple the 2-year and 3-year SR. However, intrafraction organs at risk (OAR) and target motion cause considerable uncertainty in dose delivery. Motion management, such as via beam gating triggered by ultrasound (US) guidance, can mitigate these uncertainties without the need for large safety margins that increase OAR dose and toxicity. We propose a novel motion management technique for pancreatic cancer radiotherapy called FLEX-RT, which uses real-time US images acquired by the flexible array transducer (FLEX). We created a motion phantom for simulating the pancreas' motion. A realistic model of the pancreas and duodenum was floated in a water tank over two straps that were connected to a respiratory motion platform with alternating motion. To evaluate the head of the pancreas, we compared the measured motion from the generic L7-4 probe (reference) with the FLEX. We then virtually placed FLEX on the patient's body in the CT scans of ten pancreatic cancer patients treated at our institution. To evaluate the feasibility of FLEX for motion tracking, we compared the dose distribution for (i) without FLEX, (ii) with FLEX and beam avoidance, and (iii) with FLEX and without beam avoidance. Both L7-4 and FLEX could successfully track the induced motion, with similar amplitude and cross-correlation of 0.76. The dosimetric analysis showed that FLEX can be used directly during treatment, with minimal impact on dose distribution. Not accounting for the probe caused a minor PTV coverage drop ranging from 1% to 3%. FLEX-RT, RT with FLEX-enabled motion management, is a novel and feasible solution for the longstanding problem of high motion uncertainty of pancreatic cancer RT.

The development of a dual-modality bioluminescence/ultrasound imaging system marks a significant advancement in preclinical research, particularly for creating accurate 3D tumor models. This system combines the long-term tumor tracking capabilities of bioluminescence imaging with the soft tissue examination strengths of ultrasound imaging. Its effectiveness in delineating tumor boundaries makes it an invaluable tool in oncology and related medical fields, enhancing our understanding of tissue structures and anomalies. The homemade 7.8MHz ultrasound transducers were used to obtain structural images of nude mice, while an EMCCD camera captured functional bioluminescence images. This combination allowed for the reconstruction of 3D bioluminescence emissions, effectively integrating 3D bioluminescence tomography with 3D ultrasound. This dual-modality approach resulted in highly accurate tumor localization and morphology details, as indicated by a high DICE coefficient of 88.5% and minimal localization error of 0.4mm. This demonstrates the system's capability in precise tumor mapping.

Osteoarthritis (OA) is a prevalent, degenerative disease that affects the bone and soft tissue structures of the joint. The basal thumb joint is a common site of OA and is the most important joint in the thumb. Thumb OA causes pain, stiffness, and swelling. Inflammation is recognized as an important aspect of OA, contributing to disease pathogenesis and symptoms. Blood vessel growth in the joint lining, known as the synovium, has also been associated with inflammation. Ultrasound (US) imaging provides joint visualization and Doppler US technologies can detect and visualize blood flow which indicates active joint inflammation. Despite the ability to visualize the vasculature within the synovium with Doppler technologies, its function and the role it plays in disease progression are not well understood in thumb OA. Physical therapy programs for thumb OA have shown improved patient pain and function, and changes in Doppler ultrasound signal in rheumatoid arthritis. This paper investigates synovial blood flow changes with exercise in thumb OA patients using three-dimensional (3D) US with Doppler technologies. Ten thumb OA patients were imaged with 3D US before and after resistance thumb exercises. Synovial volumes and synovial Doppler signals were determined and quantified. Changes in synovial blood flow measures with exercise were investigated to evaluate the role of synovial blood flow and the effect of exercise on the vasculature. This work aims to improve the understanding of synovial microvasculature in thumb OA.

Prostate cancer is the second-most lethal cancer in men. Since early diagnosis and treatment can drastically increase the 5-year survival rate of patients to >99%, magnetic resonance imaging (MRI) has been utilized due to its high sensitivity of 88%. However, due to lack of access to MRI, transrectal b-mode ultrasound (TRUS)-guided systematic prostate biopsy remains the standard of care for 93% of patients. While ubiquitous, TRUS-guided prostate biopsy suffers from the lack of lesion targeting, resulting in a sensitivity of 48%. To address this gap, we perform a preliminary study to assess the feasibility of localizing clinically significant cancer on b-mode ultrasound images of the prostate as input and propose a deep learning framework that learns to distinguish cancer at the pixel level. The proposed deep learning framework consists of a convolutional network with deep supervision at various scales and a clinical decision module that simultaneously learns to reduce false positive lesion predictions. We evaluated our deep learning framework using b-mode TRUS data with pathology confirmation from 330 patients, including 123 patients with pathology-confirmed cancer. Our results demonstrate the feasibility of using b-mode ultrasound images to localize prostate cancer lesions with a patient-level sensitivity and specificity of 68% and 91% respectively, compared to the reported clinical standard of 48% and 99%. The outcomes of this study show the promise of using a deep learning framework to localize prostate cancer lesions on the universally available b-mode ultrasound images; eventually improving the prostate biopsy procedures and enhancing the clinical outcomes for prostate cancer patients.

Vascular invasion is a crucial parameter for assessing the respectability of pancreatic cancer, making accurate diagnosis essential for effective treatment decisions. Deep learning techniques have shown promise in various applications with pancreatic endoscopic ultrasound (EUS) images. However, no previous study has focused on vascular invasion in the context of pancreatic cancer. There exist domain differences among ultrasound imaging equipment and also variations in preferred imaging settings among hospitals and medical professionals. Utilizing images obtained from different domains for training deep learning models can potentially result in poor diagnostic performance. Thus, it is required to overcome domain variations in EUS images to enhance diagnostic accuracy of the deep learning model for invasive pancreatic cancer. A total of 459 EUS images were acquired from 76 patients with pancreatic cancer diagnosed from two different hospitals. To overcome the differences between domains, pre-processing methods, such as alpha-blending and multi-operator transformation, were applied and their data representations were visualized. To reduce the domain variations in the deep learning model, domain adaptation was applied. The images were categorized into source and target groups based on their respective hospitals. The performance of the proposed model was evaluated and compared with well-established models including VGG16, EfficientNetB5, and ViT-L-32. Evaluation metrics were the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. Proposed model with pre-processing and domain adaptation shows significant improvement in the diagnosis of invasive pancreatic cancer. For the source domain, the proposed model achieves an impressive 92.3% accuracy, 96.5% AUC score, 85.7% sensitivity, and 96.0% specificity. For the target domain, the model exhibits performance with 73.2% accuracy, 87.8% AUC score, 92.8% sensitivity, and 66.7% specificity. The findings validate the effectiveness of the proposed model in diagnosing invasive pancreatic cancer across domain variations.

Ultrasound imaging is a powerful imaging modality for diagnosing breast tumors due to its non-invasive nature, real-time imaging capabilities, and lack of ionizing radiation. Ultrasound imaging has certain limitations that can make it demanding to detect masses compared to other imaging modalities. Therefore, breast ultrasound image segmentation is a crucial and challenging task in computer-aided diagnosis (CAD) systems. Deep learning (DL) has revolutionized medical image segmentation. Among DL models, UNet architecture is widely used for its exceptional performance. This study assesses the effectiveness of sharpening filters and attention mechanisms between the decoder and encoder in UNet models for breast ultrasound segmentation. Combining Sharp UNet and Attention UNet, we propose a novel approach called Parallel Sharp Attention UNet (PSA_UNet). A public dataset of 780 cases was utilized in this study. The results are promising for the proposed method, with the Dice coefficient and F1 score of 0.93 and 0.94, respectively. McNemar's results show that our proposed model outperforms the earlier designs upon which our model is based. In addition to introducing a new network, this study highlights the importance of optimization and finetuning in improving UNet-based segmentation models. The results offer potential improvements in breast cancer diagnosis and treatment planning through more accurate and efficient medical image segmentation techniques.

Breast cancer is one of the most common cancers among women worldwide, with early detection significantly increasing survival rates. Ultrasound imaging is a critical diagnostic tool that aids in early detection by providing real-time imaging of breast tissue. We conducted a thorough investigation of the Segment Anything Model (SAM) for the task of interactive segmentation of breast tumors in ultrasound images. We explored three pre-trained model variants: ViT_h, ViT_l, and ViT_b, among which ViT_l demonstrated superior performance in terms of mean pixel accuracy, Dice score, and IoU score. The significance of prompt interaction in improving the model's segmentation performance was also highlighted, with substantial improvements in performance metrics when prompts were incorporated. The study further evaluated the model's differential performance in segmenting malignant and benign breast tumors, with the model showing exceptional proficiency in both categories, albeit with slightly better performance for benign tumors. Furthermore, we analyzed the impacts of various breast tumor characteristics--size, contrast, aspect ratio, and complexity--on segmentation performance. Our findings reveal that tumor contrast and size positively impact the segmentation result, while complex boundaries pose challenges. The study provides valuable insights for using SAM as a robust and effective algorithm for breast tumor segmentation in ultrasound images.

Photoacoustic imaging (PAI) has emerged as a promising technique for various image guidance procedures. While convolutional neural networks (CNNs) trained on simulated radiofrequency (RF) data have been employed for point source reconstruction, their performance on real data remains a challenge. This paper addresses this limitation by introducing a novel deep learning-based method that utilizes a limited amount of experimental laser-diode-based data for the reconstruction of multiple point sources. The proposed approach employs a dual generative adversarial network (Dual-GAN) trained on experimental RF data from a combination of point source images. The Dual-GAN exhibits superior performance compared to the conventional delay-and-sum (DAS) method, demonstrating enhanced image contrast and a reduced full width at half maximum (FWHM). Notably, the axial and lateral localization errors of the Dual-GAN predictions surpass previous studies, measuring 0.028±0.018mm and 0.087±0.096mm, respectively. Additionally, the model demonstrates generalization capability by successfully reconstructing multiple point sources imaged using a different Nd:YAG laser system. This innovative method marks a significant advancement, offering improved accuracy and versatility in PAI applications involving multiple point sources.

Current ultrasound imaging techniques face challenges in producing clear brain images, primarily due to the contrast in sound velocity between the skull and brain tissues, and the difficulty in effectively coupling large probes with skulls. The reverse time migration (RTM) technique, known for its effectiveness in the geophysics community, is utilized to address this coupling issue. In addition, we propose the use of smaller probes capable of generating limited ultrasound brain image fragments from various angles. Subsequently, we have developed a new brain imaging method, termed BrainPuzzle, to restore brain images from these limited fragments. Unlike traditional transformer-based image generation models, BrainPuzzle not only uses a transformer to recognize and rearrange the fragments into their correct positions but also integrates a graph convolutional network (GCN) to automatically capture the spatial relationships among the fragments, thereby enhancing the model’s capabilities. Furthermore, we introduce the concept of using the RTM method to generate these ultrasound brain image fragments. The experimental results based on two distinct sets of generated datasets, demonstrate the exceptional performance of the proposed method in reconstructing the complete brain images from the fragments of ultrasound brain images.

High dose-rate brachytherapy is a typical part of the treatment process for cervical cancer. During this procedure, radioactive sources are placed locally to the malignancy using specialized applicators or interstitial needles. To ensure accurate dose delivery and positive patient outcomes, medical imaging is utilized intra-procedurally to ensure precise placement of the applicator. Previously, the fusion of three-dimensional ultrasound images has been investigated as an alternative volumetric imaging technique during cervical brachytherapy treatments. However, the need to manually register the two three-dimensional ultrasound images offline resulted in excessively large registration errors. To overcome this limitation, we have designed and developed a tracked, automated mechatronic system to inherently register three-dimensional ultrasound images in real-time. We perform a system calibration using an external coordinate system transform and validate the system tracking using a commercial optical tracker. The results of both experiments indicated sub-millimeter system accuracy, indicating the superior performance of our device. Future work for this study includes performing phantom validation experiments and translating our device into clinical work.

This research presents an innovative mirror-based ultrasound system designed for hand gesture classification using Convolutional Neural Network (CNN) and Vision Transformer (ViT) architectures. Hand gesture recognition using ultrasound has garnered significant interest due to its potential applications in various fields such as prosthetic control and human-machine interfacing. Traditionally, ultrasound probes are placed perpendicular to the forearm causing discomfort and interference with natural arm movements due to the center of mass of the wearable ultrasound system being distanced from the body. To address this challenge, a novel approach utilizing the advantages of acoustic reflection is proposed. A convex ultrasound probe is strategically aligned with the forearm, and ultrasound waves are transmitted to the forearm, and received back using a mirror placed at 45 degrees to the imaging region and the forearm. By aligning the probe parallel to the arm, the center of mass is brought closer to the body, ensuring enhanced stability and reduced strain on the user's arm during data collection. A dataset comprising 5 hand gestures was collected to train and evaluate the performance with Support Vector Machines with linear kernel, CNN, and ViT based approaches. It was observed that the performance of the mirror-based ultrasound system is comparable to the traditional perpendicular approach for hand gesture classification. The experimental results demonstrate the potential of the system in assisting with data acquisition and device development for hand gesture recognition using ultrasound in the field of human-machine interfacing, prosthetic control, human-computer interaction, and beyond.

Ultrasound computed tomography (UCT) is a new revolution imaging technique in clinic, which was first proposed by Greenleaf from Mayo Clinic in 1980’s, and become active in recent 10 years. In this study, we present our research advances in UCT, both techniques and clinic applications. First, an ultrasound computed tomography imaging technique to solve the small breast problem is introduced, which a set of moldings made of solid gel coupling agent (SGCA) are designed for shaping the small breast in UCT, such that a high quality and complete breast can be acquired; secondly, a new UCT technique based on sparse representation and AI technology is developed, which can greatly speed up the imaging speed in UCT by sparsely emission instead of full complete emission; finally many clinic application examples are described to demonstrate the possible potential application in clinic. Furthermore, the future works in UCT in HUST are given.

We fabricated an agar phantom to assess the detection capabilities of transabdominal ultrasonography after the consumption of various beverages. Sausages and bamboo skewers, measuring 5- and 3-mm in diameter, respectively, were embedded as signals. The thickness of the beverage-filled mock stomach was set at 3cm. We analyzed the detectability of the signal at depths of 1.5, 4.5, 7.5, and 10.5cm from the stomach. The liquid area, representing the stomach, was alternately filled with water, coffee, and Japanese, black, and milk teas to compare their signal detectability. We found no difference in detectability when the phantom was filled with water, Japanese tea, or coffee. However, using milk tea resulted in a marked improvement in signal detection. While it has been long understood that filling the stomach with water enhances signal detection during abdominal ultrasonography, our study found that filling the stomach with milk tea could lead to even more accurate results. Contrary to the belief that drinking milk prior to abdominal ultrasonography might interfere with the examination, our study demonstrated that milk tea improves detectability more than black tea.

The newly released Segment Anything Model (SAM) is a popular tool used in image processing due to its superior segmentation accuracy, variety of input prompts, training capabilities, and efficient model design. However, its current model is trained on a diverse dataset not tailored to medical images, particularly ultrasound images. Ultrasound images tend to have a lot of noise, making it difficult to segment out important structures. In this project, we developed ClickSAM, which fine-tunes the Segment Anything Model using click prompts for ultrasound images. ClickSAM has two stages of training: the first stage is trained on single-click prompts centered in the ground-truth contours, and the second stage focuses on improving the model performance through additional positive and negative click prompts. By comparing the first stage’s predictions to the ground-truth masks, true positive, false positive, and false negative segments are calculated. Positive clicks are generated using the true positive and false negative segments, and negative clicks are generated using the false positive segments. The Centroidal Voronoi Tessellation algorithm is then employed to collect positive and negative click prompts in each segment that are used to enhance the model performance during the second stage of training. With click-train methods, ClickSAM exhibits superior performance compared to other existing models for ultrasound image segmentation.

In patients deemed at risk of ischemic or embolic stroke, ultrasound elastography can be used to quantify plaque stability and assess a patient’s risk for plaque rupture. Localized plaque and vessel segmentation can help improve the processing time for performing high spatial resolution strain estimation. To provide this segmentation, we present a method for automatically detecting carotid lumen segmentations using a Mask R-CNN network. A previous study using this network found that a single-channel input of RF-derived B-mode images produced the most favorable results for bounding box detection and segmentation of the carotid lumen. However, prediction accuracy was low when the jugular vein was the most prominently visualized vessel in a given B-mode image. In this investigation, we present an optimized version of our Mask R-CNN network following the removal of the image resizing step and the addition of an improved validation technique. The new single-channel B-mode Mask R-CNN model produces a mean bounding box intersection over union (IoU) of 0.84 and a mean lumen segmentation IoU of 0.79 and overall producing cleaner results including those with a prominently visualized jugular vein.

The goal of this study was the development of a phantom for the determination of the image quality of ultrasound (US) based on the Linear System Theory. Modular transfer Function (MTF) and noise power spectrum (NPS) were determined on two US phantoms. One contained a cylinder filled with water, which appears as a circle in the US images, the second was completely homogeneous. The base material of the phantom was Poly(vinyl alcohol) which was mixed with water in a 1:9 ratio. Additionally, micro-plastic spheres and starch, respectively, were included to increase echogenicity. An algorithm was developed that calculates a radial MTF from the circular structure representing spatial resolution averaged across all directions. Noise power spectrum was determined as described by Fredenberg et al., image quality was evaluated by means of a detectability index for different diameters. Two transducers with different bandwidths (4 to 13MHz and 3 to 8MHz) were used to show the dependence of the index on the main frequency of the US wave. In addition, three penetration depths, which also require different frequencies, were used. Detectability was higher with the transducer of higher frequency for all measurements i.e. for all depths and all diameters (paired t-tests, all p < 0.01). There was also a decrease of detectability with increasing depth for both transducers. The dependence of the index on the axial distance of the ROIs was highly significant (two-sided, paired Wilcoxon test, p < 0.00001). With respect to the comparison for the different phantom materials (PVA with starch and PVA with micro-spheres), the null hypothesis (equality of variances; unpaired, two-sided Wilcoxon test) could not be rejected (p=0.14). The results suggest that the concept of the detectability index can also be applied to US images with some reservations.

We developed a novel high resolution 3D ultrasound B-scan (3D-UBS) imaging system that provides automated 3D acquisition and easily interpretable, interactive 3D visualization, including en-face and oblique views of the whole eye. Early and accurate diagnosis of ocular trauma and other associated injuries is essential for future prevention of complications. Conventional 2D ultrasound is limited in its use due to lack of trained ultrasonographer at point-of-care, difficulty in finding the optimal imaging plane and lack of anatomical context for easy interpretation. 2D hand-held ultrasound is also limited in case of perforated globes. Computed tomography (CT) is expensive and cannot be utilized if perforating intra-ocular foreign bodies (IOFBs) are small, non-metallic, or organic in nature. This study aimed to address the unmet clinical need for advanced 3D visualization of IOFBs and ocular injuries with 3D-UBS. We imaged porcine eye models for IOFBs. 3D-UBS enabled easily-obtained, informative images of ocular injuries, without an expert as required in conventional 2D ultrasound. En face and oblique views provided by multiplanar reformatting allows selection of optimal planes after acquisition. Size and shape of the IOFBs can be detected more accurately with 3D-UBS. 3D-UBS also provides information on location of IOFBs with respect to other important ocular structures. 3D-UBS shows 2.4 times contrast improvement compared to CT in wooden IOFB visualization. Our study demonstrated that novel 3D-UBS can be used for assessing ocular injuries (i.e., identifying the location, size, and shape of IOFBs) which can guide the treatment process.

When using a ring array to perform ultrasound tomography, the most computationally intensive component of frequency-domain full waveform inversion (FWI) is the Helmholtz equation solver. The Helmholtz equation is an elliptic partial differential equation (PDE) whose discretization leads to a large system of equations; in many cases, the solution of this large system is itself the inverse problem and requires an iterative method. Our current solution relies on discretizing the 2D Helmholtz equation based on a 9-point stencil and using the resulting block tridiagonal structure to efficiently compute a block LU factorization. Conceptually, the L and U systems are equivalent to a forward and backward wave propagation along one of the spatial dimensions of the grid, resulting in a direct non-iterative solution to the Helmholtz equation based on a single forward and backward sweep. Based on this observation, the PDE representation of the Helmholtz equation is split into two one-way wave equations prior to discretization. The numerical implementations of these one-way wave equations are highly parallelizable and lend themselves favorably to accelerated GPU implementations. We consider the Fourier split-step and phase-shift-plus-interpolation (PSPI) methods from seismic imaging as numerical solutions to the one-way wave equations. We examine how each scheme affects the numerical accuracy of the final Helmholtz equation solution and present its impact on FWI with breast imaging examples.

Breast cancer is the most common cancer in women worldwide. Two million women are diagnosed annually, resulting in 685,000 annual deaths. Early diagnosis is critical to reducing mortality. Although screening with mammography has been shown to have reduced breast cancer-related mortality through early detection, dense breast tissues reduce mammographic sensitivity, potentially delaying diagnoses, and contributing to poorer outcomes. Therefore, there is a need for more accessible and cost-effective supplemental screening technologies, especially for high-risk populations, especially women with dense breasts. To address these challenges, a promising approach involves combining widely available, cost-effective, and accessible ultrasound-based technologies with economical hardware, software modules, and automated techniques. Among these technologies, Doppler imaging plays a crucial role in the clinical evaluation of breast abnormalities, as intratumoural blood flow has been shown to correlate with the aggressiveness and histological grade of the tumour. The development of a novel automated, portable, and a patient-dedicated 3D automated breast ultrasound (ABUS) system for point-of-care breast cancer supplemental screening holds significant promise. The proposed system has previously demonstrated the capability to generate accurate whole-breast B-mode images, which can aid in the early detection of breast cancer in women with dense breasts. Additionally, it offers the advantage of incorporating Doppler imaging for the assessment of blood flow within suspicious lesions, a capability not commonly available with commercial ABUS systems. By leveraging Doppler imaging in conjunction with 3D Bmode ABUS, this innovative approach could improve breast cancer-related health outcomes and equity in access to healthcare, especially for underserved and vulnerable populations.

Speckle noise is an integral component in medical ultrasound imaging, which have a random granular pattern formation. This noise degrades the visual quality of ultrasound images and complicates image-based interpretation and diagnosis. The removal of interference-induced noise is a primary challenge as ultrasound image studies seek to achieve higher accuracy and characterize more subtle small and low-contrast lesions. In this study, a novel method, which combines the nonlocal-means (NLM) filter with a simple unsupervised deep model named Laplacian Eigenmaps network (LENet), has been proposed for ultrasonic speckle reduction. The proposed method exploits both the global features, redundancy information and self-similarity properties of noisy images, which first extract features from the noisy image by the Laplacian Eigenmaps algorithm, and then apply it to refine the image self-similarities weight for helping the NLM filter to provide better despeckling performance. Specifically, this is a two-stage approach that the first stage is to learn filter banks from a small quantity of training samples by LENet network and the following stage is to utilize the output eigenvectors as similarity metrics of pixels within the NLM filter. The performance of our approach is compared with related state-of-the-art methods on synthetic images, simulated image and real ultrasound images. The results show that our method can provide better noise removal ability over many previously despecking filters.

Ultrasound imaging can be used to visualize the boundaries between two media with different properties, but less information is available about the medium penetrated. Therefore, conventional ultrasound devices used in routine medical imaging usually assume a constant value for the speed of sound of the structures. Thus, current research aims to develop new methods to obtain more information about the medium scanned. To support the research ultrasound phantoms with precisely known properties like speed of sound, attenuation or hardness can be used. We developed polyvinyl alcohol-based phantoms and investigated the correlation between the properties and the freeze-thaw cycles (FTC), as well as the effects of glycerol on the speed of sound. Additionally, we designed 3D printed molds for shaping the phantom and improving the FTC process. Furthermore, a construction was designed to prevent the measurement setup from being contaminated by PVA and glycerol defusing out of the phantom. The physical properties of the phantoms measured were the speed of sound, attenuation, and Shore hardness. The measurements conducted were carried out using the standards of the American Institute for Ultrasound in Medicine (AIUM) and the American Society for Testing and Materials (ASTM). There was an increase in Shore hardness and speed of sound as the number of FTCs increased. However, the differences in the speed of sound were only minimal, whereby higher differences could be achieved by the addition of glycerol. At present, no statements can be made about the attenuation. The measurements of the materials are in an early state and further improvement is needed. The manufacturing process and measurements were improved by 3D printed molds.

Non-invasive ultrasound imaging has been used to assess changes in the arterial wall such as changes in stiffness and the presence of plaque and its progression. Typically, stiffer arteries are thought to contribute to the progression of atherosclerosis. Therefore, it would be beneficial to find tools that assess and monitor changes in arterial stiffness to assist in the identification of cardiovascular diseases (CVD) and to help optimize the treatment of CVD risk factors for these individuals. The objective of this research is to use non-invasive ultrasound-based Lagrangian carotid strain imaging to assess arterial stiffness in determining CVD risk. Utilizing radiofrequency data from the right and left common carotid arteries (CCAs), we calculated the accumulated axial, lateral, and shear strain indices and the peak-to-peak mean strain values for segmented CCA walls. Data were collected from participants with carotid plaque and age-matched controls without plaque. Radiofrequency data from two cardiac cycles were used to calculate all strain measures. The Mann- Whitney U-Test was used to compare strain values between participants with plaque and participants without plaque. All mean strain values, except for the LCCA axial strain, were noted to be lower in the participants with plaque compared to the controls without plaque, although these findings were not statistically significantly different. . These findings suggest that participants with plaque may have stiffer arteries due to the lower strain values. This study needs to be performed in a larger cohort to see if these differences are statistically significantly different.

Evidence for Biot slow wave data in 3D ultrasound tomographic (3D UT) data from an orthopedic scan is shown by segmentation interior to Bone. Previous results have shown the quantitative accuracy of the 3D ultrasound volography method for ligaments, cartilage, tendons, skin, fat, muscle, etc. Interior to bone the SOS values are lower than expected. It is known that a slow compressional wave is predicted by Johnson-Biot theory which has speed of sound (SOS) values independently measured which are similar to our values based on segmentation for trabecular bone matrix interior to bone. Values for marrow are determined from segmentation and are commensurate with literature values as well. Concepts from algebraic topology are applied to the tomographic data (the first homotopy group of the N-torus, where N is the number of receiver elements in the array) and a quantitative comparison of the data redundancy with 2D algorithms is carried out. The use of algebraic topology gives a suitable context in which to understand phase unwrapping issues and leads to constraints on the distance between the data acquisition (DA) data levels and the frequencies used in the reconstruction. The data redundancy comparison applies to any 3D vs 2D comparison, e.g. 3D UT compared with MRI and shows the much larger size of the information contributing to a single voxel in the 3D vs 2D as long as the 3D model for wave propagation is used. The implications of this are discussed. Validation of attenuation variation with frequency is shown.

Power Doppler ultrasound (PD-US) is used for visualizing and diagnosing blood circulation, yet its insufficient frame rate presents a significant challenge for practitioners. To increase the frame rate, we investigate three existing motion-based interpolation methods and find they perform better than linear interpolation but can not provide medically consistent temporal interpolations for PD-US. We address these limitations by proposing a method to detect and correct interpolation error-affected regions in temporal interpolations through differences in extrapolation matchings, motion smoothness and motion variance. We validate our approach using clinical obstetric data.

In this work, we introduce the first realistic digital phantoms for prostate ultrasound and photoacoustic (PA) imaging in the male pelvic region. Our model encompasses surrounding tissues or organs around the prostate, including fat, bones (including femoral heads), muscles, urinary bladder, rectum, anal canal, penile bulb, neurovascular bundles, and seminal vesicles. Each digital phantom set contains five parameters: speed of sound, density, acoustic attenuation, optical absorption, and reduced optical scattering. The anatomical structures were derived from open-source computed tomography (CT) and magnetic resonance imaging (MRI) data from the Gold Atlas Project. The acoustic parameters, including the speed of sound and attenuation of the prostate, were obtained from an ex vivo prostate study utilizing the QTscan ultrasound tomography (UST) platform at the National Institutes of Health. All other parameters were acquired from the literature. By employing atlas data from four pelvises and UST images of 62 ex vivo prostate specimens, we generated 248 sets of digital phantoms. Additionally, we demonstrate a practical application, showcasing the identification of the effective insonification window for prostate UST. The developed digital phantoms are made open source and can be found at https://github.com/ywu115/Prostate-digital-phantoms. They can be leveraged for imaging device design, image formation studies, image reconstruction validation, and diagnostic and treatment planning for numerous prostate studies using ultrasound and PA imaging. Since the digital phantoms cover the entire male pelvic region, they can potentially be extended to other applications, such as bladder cancer imaging, cyst detection in seminal vesicles, hernias, and treatment planning for point-of-care ultrasound.

Ultrasound computed tomography (USCT) is an emerging imaging technique that holds great promise for breast cancer diagnosis and screening. Full-waveform inversion (FWI)-based image reconstruction methods can produce high spatial resolution images that depict the acoustic properties of soft tissues. However, FWI is computationally demanding, especially when wave physics is modeled in three dimensions (3D). A common USCT design employs a circular ring-array comprised of elevation-focused ultrasonic transducers, where the array is translated orthogonally to achieve volumetric (3D) imaging. This design allows efficient two-dimensional (2D) slice-byslice reconstruction methods to estimate a 3D volume by stacking reconstructed cross-sectional images at each ring-array position. However, these 2D methods do not account for the 3D wave propagation physics and the focusing properties of the transducers and thus can result in out-of-plane scattering-based artifacts and inaccuracies. Previous work has investigated a learning-based method, in which a deep neural network was used to map the 3D ring-array USCT data to idealized 2D USCT measurements, from which acoustic properties can be estimated by use of 2D FWI. The presented work extends that previous study in two ways. First, sophisticated anatomically realistic 3D numerical breast phantoms are used to construct clinically relevant training and testing sets. Second, a high-fidelity 3D wave propagation forward imaging model incorporating elevation focusing effects is used to virtually image the phantoms. The results show promise in improving image accuracy compared to both 3D FWI and conventional 2D FWI, achieving both mitigation of out-of-plane scattering and 3D-2D model mismatches and significant reduction of computational cost compared to 3D FWI. Additionally, the use of multiple ring-array measurements from adjacent elevations are explored as concurrent neural network inputs, showing improved accuracy compared to only using single-ring data.

Prostate cancer ranks among the most prevalent types of cancer in males, prompting a demand for early detection and non-invasive diagnostic techniques. This paper explores the potential of ultrasound radiofrequency (RF) data to study different anatomic zones of the prostate. The study leverages RF data's capacity to capture nuanced acoustic information from clinical transducers. The research focuses on the peripheral zone due to its high susceptibility to cancer. The feasibility of utilizing RF data for classification is evaluated using ex-vivo whole prostate specimens from human patients. Ultrasound data, acquired using a phased array transducer, is processed, and correlated with B-mode images. A range filter is applied to highlight the peripheral zone's distinct features, observed in both RF data and 3D plots. Radiomic features were extracted from RF data to enhance tissue characterization and segmentation. The study demonstrated RF data's ability to differentiate tissue structures and emphasizes its potential for prostate tissue classification, addressing the current limitations of ultrasound imaging for prostate management. These findings advocate for the integration of RF data into ultrasound diagnostics, potentially transforming prostate cancer diagnosis and management in the future.

Ultrasound is a commonly used modality for medical imaging. While this modality has great advantages in terms of safety and cost relative to other imaging modalities, it also has several limitations. Signal-to-noise ratio varies greatly depending on the acoustic properties of the tissue being imaged and the depth of the target structures. In this work, we evaluate the use of deep learning based methods to reconstruct 3D surfaces of general objects imaged with ultrasound. We evaluate three variants of the 3D U-Net with different training scenarios. We were able to train networks to reconstruct three distinct categories of objects relatively well when trained on limited data from each category. However, the performance of the networks did not generalize well when testing on categories of objects not included in the training. We also investigated the effects of employing dual-task autoencoding on generalizability. These results provide a baseline for exploring modifications to the U-Net framework to improve generalizability. A generalizable method could improve visualization for a number of ultrasound imaging tasks.

This study aimed to develop a software tool to simulate ultrasound computed tomography (USCT) scans of a realistic anthropomorphic breast phantom. The tool was incorporated in an existing simulation framework for x-ray based imaging, resulting in a multimodality in silico evaluation framework. Digital breast models were developed utilizing advanced mathematical modeling based on procedural generation techniques to closely mimic the complex structural and compositional heterogeneity observed in human breast tissue. The breast’s acoustic properties, including the speed of sound, density, and acoustic attenuation, were stochastically incorporated into the phantom in preparation for USCT simulation. Full waveform inversion was employed to reconstruct the ultrasound images. We were able to image a dense breast model with multiple modalities in silico. In addition to USCT simulation with the newly developed software, digital breast tomosynthesis, mammography, and computed tomography images of the same phantom were simulated using the GPU-accelerated Monte Carlo transport software MC-GPU. The presented in silico methods can be used to compare the performance of imaging technologies based on completely different physical characteristics, such as x-ray attenuation or ultrasound speed of sound. In silico evaluation has the advantage of enabling task-based evaluation in a fixed anatomy with known disease ground-truth, and without exposure of patients to ionizing radiation. With the required level of verification and validation, this framework could enhance our understanding of the performance of USCT as a standalone modality or as an adjunct to x-ray modalities, potentially providing in silico evidence that could be leveraged in regulatory evaluation of new USCT devices.