KEYWORDS: Ultrasonography, Elastography, Tissues, Signal to noise ratio, Detection and tracking algorithms, Deformation, Data acquisition, Computer simulations, Visualization, Sampling rates
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.
KEYWORDS: Ultrasonography, Statistical analysis, Education and training, Deep learning, Data modeling, Tissues, Point spread functions, Measurement uncertainty, Backscatter, Scattering
Quantitative ultrasound (QUS) aims to find properties of scatterers which are related to the tissue microstructure. Among different QUS parameters, scatterer number density has been found to be a reliable biomarker to detect different abnormalities. The homodyned K-distribution (HK-distribution) is a model for the probability density function of the ultrasound echo amplitude that can model different scattering scenarios but requires a large number of samples to be estimated reliably. Parametric images of HK-distribution parameters can be formed by dividing the envelope data into small overlapping patches and estimating parameters within the patches independently. This approach imposes two limiting constraints: the HK-distribution parameters are assumed to be constant within each patch, and each patch requires enough independent samples. In order to mitigate those problems, we employ a deep learning approach to estimate parametric images of scatterer number density (related to HK-distribution shape parameter) without patching. Furthermore, an uncertainty map of the network’s prediction is quantified to provide insight about the confidence of the network about the estimated HK parameter values.
KEYWORDS: Ultrasonography, Education and training, Signal to noise ratio, Denoising, Data modeling, Signal attenuation, Liver, Data analysis, Tissues, Breast
On one hand, the transmitted ultrasound beam gets attenuated as propagates through the tissue. On the other hand, the received Radio-Frequency (RF) data contains an additive Gaussian noise which is brought about by the acquisition card and the sensor noise. These two factors lead to a decreasing Signal to Noise Ratio (SNR) in the RF data with depth, effectively rendering deep regions of B-Mode images highly unreliable. There are three common approaches to mitigate this problem. First, increasing the power of transmitted beam which is limited by safety threshold. Averaging consecutive frames is the second option which not only reduces the framerate but also is not applicable for moving targets. And third, reducing the transmission frequency, which deteriorates spatial resolution. Many deep denoising techniques have been developed, but they often require clean data for training the model, which is usually only available in simulated images. Herein, a deep noise reduction approach is proposed which does not need clean training target. The model is constructed between noisy input-output pairs, and the training process interestingly converges to the clean image that is the average of noisy pairs. Experimental results on real phantom as well as ex vivo data confirm the efficacy of the proposed method for noise cancellation.
Anatomical landmark identification is crucial in the registration of medical images in a wide range of clinical applications. Various machine and deep learning (DL) techniques have been proposed to annotate anatomical landmarks automatically. However, very few have taken advantage of the more recent Transformer models that do not suffer from inductive bias of convolutional neural networks as well as incorporating uncertainty assessments. This paper proposes a novel technique based on the Swin Transformer V2 (Swin-V2) with an uncertainty quantification module for robust anatomical landmark detection in paraspinal muscle MRIs to facilitate muscle morphometric analysis for low back pain (LBP). Specifically, we employ statistical measurements from stochastic sampling using Monte Carlo (MC) dropouts as uncertainty metrics for automatic landmark detection. The proposed method was trained and validated using 350 axial MRI slices of paraspinal muscles at the mid-disc spinal levels commonly associated with LBP. We demonstrate that the selected uncertainty metrics are correlated with automatic landmark detection errors and, in addition, can be used as inputs to grade the quality of the identified landmarks with a lightweight random forest algorithm for a more straightforward interpretation.
Recently, Convolutional Neural Networks (CNNs) have been very successful in optical flow estimation in computer vision. UltraSound Elastography (USE) displacement estimation step can be performed by optical flow CNNs. However, there is a large domain gap between ultrasound Radio-Frequency (RF) data and computer vision images which reduces the overall accuracy of displacement estimation. Some modifications of the network architecture are required to be able to extract reliable information from RF data. Modified Pyramidal Network (MPWC-Net) which is based on the well-known PWC-Net was among the first attempts that adopts the optical flow CNNs to USE displacement estimation. However, MPWC-Net suffers from several shortcomings that limit its application especially for unsupervised training. In this paper, we propose additional modifications to substantially improve MPWC-Net. We also publicly released the network’s trained weights.
Ultrasound (US) is a low-cost, portable, and safe tool for breast cancer screening. However, automatic classification of invasive ductal carcinoma (IDC) in US is a difficult classification task due to their similar appearance to fibroadenoma (FA) (a type of benign tumor). Another challenge is the limited availability of US data with ground truth labels, further complicating the adoption of deep learning techniques for IDC detection. It has been shown that deep classification networks perform better when they simultaneously learn multiple correlated tasks. However, most previous studies on breast US classifications focused on the binary classification of benign versus malignant tumors. To this end, we propose a multi-class classification deep learning-based strategy mainly focusing on the classification of IDC. Inspired by multi-task learning (MTL), we adopt a novel scheme in adding the background tissue as an additional class and show substantial improvements in IDC detection.
Ultrasound elastography entails imaging mechanical properties of tissue and is therefore of significant clinical importance. In elastography, two frames of radio-frequency (RF) ultrasound data that are obtained while the tissue is undergoing deformation, and the time-delay estimate (TDE) between the two frames is used to infer mechanical properties of tissue. TDE is a critical step in elastography, and is challenging due to noise and signal decorrelation. This paper presents a novel and robust technique TDE using all samples of RF data simultaneously. We assume tissue deformation can be approximated by an affine transformation, and hence call our method ATME (Affine Transformation Model Elastography). The affine transformation model is utilized to obtain initial estimates of axial and lateral displacement fields. The affine transformation only has six degrees of freedom (DOF), and as such, can be efficiently estimated. A nonlinear cost function that incorporates similarity of RF data intensity and prior information of displacement continuity is formulated to fine-tune the initial affine deformation field. Optimization of this function involves searching for TDE of all samples of the RF data. The optimization problem is converted to a sparse linear system of equations, which can be solved in real-time. Results on simulation are presented for validation. We further collect RF data from in-vivo patellar tendon and medial collateral ligament (MCL), and show that ATME can be used to accurately track tissue displacement.
KEYWORDS: Statistical analysis, Ultrasonography, Elastography, Computer programming, Tissues, In vivo imaging, Image analysis, Real time image processing, Lithium, Data acquisition, Liver cancer
Ultrasound Elastography is an emerging imaging technique that allows estimation of the mechanical characteristics of tissue. Two issues that need to be addressed before widespread use of elastography in clinical environments are real time constraints and deteriorating effects of signal decorrelation between pre- and post-compression images. Previous work has used Dynamic Programming (DP) to estimate tissue deformation. However, in case of large signal decorrelation, DP can fail. In this paper we, have proposed a novel solution to this problem by solving DP on a tree instead of a single Radio-Frequency line. Formulation of DP on a tree allows exploiting significantly more information, and as such, is more robust and accurate. Our results on phantom and in-vivo human data show that DP on tree significantly outperforms traditional DP in ultrasound elastography.
Ultrasound (US) tissue characterization provides valuable information for the initialization of automatic segmentation algorithms, and can further provide complementary information for diagnosis of pathologies. US tissue characterization is challenging due to the presence of various types of image artifacts and dependence on the sonographer’s skills. One way of overcoming this challenge is by characterizing images based on the distribution of the backscatter data derived from the interaction between US waves and tissue. The goal of this work is to classify liver versus kidney tissue in 3D volumetric US data using the distribution of backscatter US data recovered from end-user displayed Bmode image available in clinical systems. To this end, we first propose the computation of a large set of features based on the homodyned-K distribution of the speckle as well as the correlation coefficients between small patches in 3D images. We then utilize the random forests framework to select the most important features for classification. Experiments on in-vivo 3D US data from nine pediatric patients with hydronephrosis showed an average accuracy of 94% for the classification of liver and kidney tissues showing a good potential of this work to assist in the classification and segmentation of abdominal soft tissue.
Ultrasound elastography is an imaging technology which can detect differences in tissue stiffness based on tissue
deformation. For successful clinical use in cancer diagnosis and monitoring the method should be robust to sources
of decorrelation between ultrasound images. A regularized Dynamic Programming (DP) approach was used for
displacement estimation in compressed tissue. In the Analytic Minimization (AM) extension of DP, integer
displacements are calculated just for one RF-line, and later propagated laterally throughout the entire image.
This makes the seed RF-line very important; faulty seed lines could propagate erroneous displacement values
throughout the image resulting in the appearance of false "lesions". In this paper we analyze the robustness of
this method in free-hand palpation of laboratory tissue phantoms. We are proposing an update to the algorithm
which includes a random search for the most robust seed RF-line. Axial integer displacements are obtained
on each random seed line individually with DP optimization. For each random axial RF-line, multiple random
values for decorrelation compensation are used in the displacement estimation. The displacement values are then
compared and several metrics of stability and consistency are considered. A ranking is established and the line
deemed most robust will become the seed line for displacement propagation, while also selecting the most stable
value for decorrelation compensation. The random search can be achieved at no additional computational cost
in a parallel implementation. The results indicate significant improvement in the robustness of the DP approach,
while maintaining real-time computation of strain images.
KEYWORDS: Computer programming, Signal to noise ratio, Tissues, Ultrasonography, Elastography, Medical imaging, Interference (communication), Image compression, Signal analyzers, Tumors
Elastography, computation of elasticity modulus of tissue is one of medical imaging methods with
applications such as tumor detection and ablation therapy. Phase-based time delay estimation methods exploit
the frequency information of the RF data to obtain strain estimates [1]. Although iterative Phase zero
estimation is more computationally efficient in comparison to methods that seek for the absolute maximum
cross-correlation between precompression and postcompression echo signals, it is quite sensitive to noise. The
reason for this sensitivity is that for this iterative method an initial guess for the time shift is needed for each
pixel. To estimate time shifts for the sample k, the time shift resulted from iterative phase zero method applied
on sample k-1 is used as an initial value. This makes the method sensitive to noise because the error is
propagating sample by sample and if the method gets unstable for any pixel, it will give unstable result for the
following pixels in image line. Proposed strategy in this work to overcome this problem is to first estimate the
displacement using Dynamic Programming [2] and use the results from DP as an initial guess of displacement
for each pixel in iterative Phase zero method. Recently, regularized methods that incorporate the prior of
tissue continuity in time delay estimation have been shown to produce low-noise and high contrast strain
images [3,5]. In this work, we also incorporate the prior of tissue motion continuity in the phase zero method
to make the zero-phase method more robust to signal decorrelation.
Out-of-plane motion in freehand 3D ultrasound can be estimated using the correlation of corresponding patches,
leading to sensorless freehand 3D ultrasound systems. The correlation between two images is related to their
distance by calibrating the ultrasound probe: the probe is moved with an accurate stage (or with a robot in
this work) and images of a phantom are collected, such that the position of each image is known. Since parts
of the calibration curve with higher derivative gives lower displacement estimation error, previous work limits
displacement estimation to parts with maximum derivative. In this paper, we first propose a novel method for
exploiting the entire calibration curve by using a maximum likelihood estimator (MLE). We then propose for
the first time using constrains inside the image to enhance the accuracy of out-of-plane motion estimation. We
specifically use continuity constraint of a needle to reduce the variance of the estimated out-of-plane motion.
Simulation and real tissue experimental results are presented.
Monitoring the ablation process in order to document the adequacy of margins during treatment is of significant
importance. Observing that the ablation lesion is harder than normal tissue, it has been proposed to monitor
the ablation using ultrasound elastography. Furthermore, it has been reported that the ablated cancer tumor is
harder than ablated normal tissue. In this paper we propose an ultrasound elastography technique for visualizing
the ablation lesion and the ablated cancerous tumor in Hepatocellular carcinoma (HCC). This work focuses on
devising techniques to generate elasticity images which distinguish the ablated cancerous tumor and the ablated
normal lesion. We first calculate the displacement field between two ultrasound images acquired before and after
some compression. We then use the displacement field to calculate the correlation coefficient between the two
images. Parts of the tissue that undergo large deformation give small correlation coefficient due to decorrelation
within each window, and parts of the tissue that undergo small deformation give large correlation coefficient.
Simulating phantoms with two lesions, a harder tumor inside a hard lesion, using finite element and Field II, we
show that this method enables delineating the tumor from the lesion.
Volumetric ultrasound imaging has not gained wide recognition, despite the availability of real-time 3D ultrasound
scanners and the anticipated potential of 3D ultrasound imaging in diagnostic and interventional radiology. Their use,
however, has been hindered by the lack of real-time visualization methods that are capable of producing high quality 3D
rendering of the target/surface of interest. Volume rendering is a known visualization method, which can display clear
surfaces out of the acquired volumetric data, and has an increasing number of applications utilizing CT and MRI data.
The key element of any volume rendering pipeline is the ability to classify the target/surface of interest by setting an
appropriate opacity function. Practical and successful real-time 3D ultrasound volume rendering can be achieved in
Obstetrics and Angio applications where setting these opacity functions can be done rapidly, and reliably. Unfortunately,
3D ultrasound volume rendering of soft tissues is a challenging task due to the presence of significant amount of noise
and speckle. Recently, several research groups have shown the feasibility of producing 3D elasticity volume from two
consecutive 3D ultrasound scans. This report describes a novel volume rendering pipeline utilizing elasticity
information. The basic idea is to compute B-mode voxel opacity from the rapidly calculated strain values, which can
also be mixed with conventional gradient based opacity function. We have implemented the volume renderer using GPU
unit, which gives an update rate of 40 volume/sec.
Breast irradiation significantly reduces the risk of recurrence of cancer. There is growing evidence suggesting that
irradiation of only the involved area of the breast, partial breast irradiation (PBI), is as effective as whole breast
irradiation. Benefits of PBI include shortened treatment time, and perhaps fewer side effects as less tissue is
treated. However, these benefits cannot be realized without precise and accurate localization of the lumpectomy
cavity. Several studies have shown that accurate delineation of the cavity in CT scans is very challenging and
the delineated volumes differ dramatically over time and among users.
In this paper, we propose utilizing 3D ultrasound (3D-US) and tracked strain images as complementary
modalities to reduce uncertainties associated with current CT planning workflow. We present the early version
of an integrated system that fuses 3D-US and real-time strain images. For the first time, we employ tracking
information to reduce the noise in calculation of strain image by choosing the properly compressed frames and
to position the strain image within the ultrasound volume. Using this system, we provide the tools to retrieve
additional information from 3D-US and strain image alongside the CT scan. We have preliminarily evaluated
our proposed system in a step-by-step fashion using a breast phantom and clinical experiments.
Radical prostatectomy using the laparoscopic and robot-assisted approach lacks tactile feedback. Without palpation,
the surgeon needs an affordable imaging technology which can be easily incorporated into the laparoscopic
surgical procedure, allowing for precise real time intraoperative tumor localization that will guide the extent
of surgical resection. Ultrasound elastography (USE) is a novel ultrasound imaging technology that can detect
differences in tissue density or stiffness based on tissue deformation. USE was evaluated here as an enabling
technology for image guided laparoscopic prostatectomy. USE using a 2D Dynamic Programming (DP) algorithm
was applied on data from ex vivo human prostate specimens. It proved consistent in identification of
lesions; hard and soft, malignant and benign, located in the prostate's central gland or in the peripheral zone.
We noticed the 2D DP method was able to generate low-noise elastograms using two frames belonging to the
same compression or relaxation part of the palpation excitation, even at compression rates up to 10%. Good
preliminary results were validated by pathology findings, and also by in vivo and ex vivo MR imaging. We also
evaluated the use of ultrasound elastography for imaging cavernous nerves; here we present data from animal
model experiments.
Electronic colon cleansing (ECC) aims to remove the contrast agent from the CT abdominal images so that a virtual
model of the colon can be constructed. Virtual colonoscopy requires either liquid or solid preparation of the colon before
CT imaging. This paper has two parts to address ECC in both preparation methods. In the first part, meniscus removal in
the liquid preparation is studied. The meniscus is the curve seen at the top of a liquid in response to its container. Left on
the colon wall, the meniscus can decrease the sensitivity and specificity of virtual colonoscopy. We state the differential
equation that governs the profile of the meniscus and propose an algorithm for calculating the boundary of the contrast
agent. We compute the surface tension of the liquid-colon wall contact using in-vivo CT data. Our results show that the
surface tension can be estimated with an acceptable degree of uncertainty. Such an estimate, along with the meniscus
profile differential equation will be used as an a priori knowledge to aid meniscus segmentation. In the second part, we
study ECC in solid preparation of colon. Since the colon is pressurized with air before acquisition of the CT images, a
prior on the shape of the colon wall can be obtained. We present such prior and investigate it using patient data. We
show the shape prior is held in certain parts of the colon and propose a method that uses this prior to ease pseudoenhancement
correction.
The elevational distance between two ultrasound images can be obtained from the correlation between the two images, leading to sensorless freehand 3D ultrasound systems. Most of these systems rely on the correlation between patches of fully developed speckles (FDS). Previous work has compared different FDS detectors and concluded that the elevational distance measurement limited to the FDS patches obtained by low order moment test yields significantly more accurate results than other FDS detectors. However, small coherent and FDS regions are spread throughout a typical ultrasound image of real tissue. This makes it extremely unlikely to find a regularly shaped (conventionally a rectangle) FDS patch, making it infeasible to estimate elevational distance accurately1. In this work, first we propose a simple and fast algorithm which is capable of detecting arbitrarily irregular FDS regions in an ultrasound image. In vitro experiments on beef liver, beef steak and chicken breast indicates that the proposed algorithm generates remarkably more FDS patches than the current methods. Preliminary results show that the FDS patches obtained by this algorithm generate more accurate elevational distance measurement. Second, we propose a new calibration scheme to generate decorrelation curves. At a particular location in the image, conventional methods acquire one decorrelation curve. We create multiple curves, as a function of particular statistical properties of the patch. The results reveal a theoretically expected relation between the decorrelation curve and the statistical properties of the patch. As a result of this calibration based on the patch statistical properties, improvement in the out of plane motion estimation is expected.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.