Quantitative ultrasound medical imaging enhanced by intervening tissue determination

Abstract

To provide an ultrasound diagnostic apparatus capable of quantitative evaluation of soft tissue characteristics.SOLUTION: In one approach, quantitative ultrasound imaging is altered to account for the effects of intervening tissue. The tissue layers between a transducer and a region are measured, and the measurement is input to a machine-learned model to quantify from signals for the region of interest and the measurement. This may provide more accurate quantification. In another approach, the region of interest (ROI) is automatically placed, such as through detection of anatomy (e.g., liver capsule), identification of and / or guidance to the field of view, and / or scoring of imaging of the anatomy. This ROI placement may avoid variability in quantification.SELECTED DRAWING: Figure 1

Description

[Background technology]

[0001] The presently disclosed embodiments relate to ultrasound imaging of soft tissue. In ultrasound imaging of soft tissue, such as liver imaging, soft tissue properties can be diagnostically useful. These properties include elasticity, shear transmission, and / or fat fraction. Quantitative assessment of tissue properties is affected by the type, thickness, and alignment of intervening tissue layers. For example, attenuation in intervening layers can cause miscalculation of backscatter in soft tissue. In the case of ultrasound-derived fat fraction (UDFF), higher fat content values ​​may deviate from values ​​determined using magnetic resonance. At higher fat levels, ultrasound-derived fat fraction may plateau or not increase as quickly as magnetic resonance-derived fat fraction, resulting in less accurate quantification.

[0002] Guidelines recommend that for elastography, the measurement line from the skin surface to the region of interest be placed perpendicular to the liver capsule, with the region of interest located a minimum of 1 cm below the capsule. The workflow is challenging when measuring angles and distances from a region of interest, such as the distance from the region of interest to the liver capsule in fat fraction measurements. Manual measurements or visual approximations of distances and angles are often used, resulting in undesirable variability in quantification. Summary of the Invention

[0003] By way of preamble, the preferred embodiments described below include methods, instructions, and systems for ultrasound imaging using an ultrasound scanner. In one approach, quantitative imaging is improved to account for the influence of intervening tissue. The tissue layer between the transducer and the region is measured, and the measurements are input into a machine learning model that quantifies the signal and measurements for the region of interest. This approach can provide more accurate quantification. In another approach, a region of interest (ROI) is automatically positioned, for example, through detection of an anatomical structure (e.g., liver capsule), identification and / or navigation of a field of view, and / or scoring of an anatomical structure image. This ROI setting can avoid variability in quantification.

[0004] In a first aspect, a method for ultrasound imaging using an ultrasound scanner is provided. The ultrasound scanner measures tissue between the liver and a transducer of the ultrasound scanner and scans a region of interest within the liver. An ultrasound-derived fat fraction (UDFF) of the liver is determined using a first machine learning model configured to receive the tissue measurements and information from the scan and output a UDFF. The UDFF is displayed.

[0005] In a second aspect, a system for ultrasound medical imaging is provided. A beamformer is configured to scan soft tissue of a patient with a transducer. The soft tissue is in a region of interest. The beamformer is configured to scan intervening tissue between the transducer and the soft tissue with the transducer. An image processor is configured to locate the region of interest and is configured to quantify a first property of the soft tissue from the scan of the soft tissue with a first machine learning model. The first property is an ultrasound-derived fat fraction, a shear wave property, and / or an elastography property. The first machine learning model receives a second property of the intervening tissue as an input for outputting the first property. A display is configured to display an ultrasound image representing the quantification of the first property of the soft tissue.

[0006] In a third aspect, a method is provided for ultrasonic quantification of soft tissue properties using an ultrasound scanner. An image processor detects anatomical structures to locate a region of interest in the soft tissue. The shear, elasticity, and / or fat fraction of the soft tissue within the region of interest is determined. The shear, elasticity, and / or fat fraction of the tissue in the region of interest is displayed.

[0007] Other aspects are summarized below in the exemplary embodiments. The present invention is defined by the claims, and nothing in this section should be considered as limiting the scope of the claims. Other aspects and advantages of the present invention are described below in conjunction with the preferred embodiments, and may be subsequently claimed individually or in combination. [Brief explanation of the drawings]

[0008] The components and drawings are not necessarily to scale, but rather have been exaggerated to illustrate the principles of the present invention. Moreover, in the drawings, like reference characters indicate corresponding parts throughout the various views.

[0009] [Figure 1] 1 is a flow chart of one embodiment of a method for ultrasound imaging using an ultrasound scanner with measurements of intervening tissue incorporated into quantification and consistent ROI location. [Figure 2] FIG. 10 is a diagram illustrating an ultrasound image for quantifying fat fraction. [Figure 3] 1 is a block diagram of one embodiment of a system for ultrasound medical quantification and / or consistent ROI positioning that accounts for intervening tissue. Detailed Description of the Drawings and Presently Preferred Embodiments

[0010] In one approach, tissue layer detection and modeling enhances quantitative ultrasound imaging. Intervening tissue layers are detected. Thickness and other measurements of the intervening layers are used as part of a model to estimate (quantify) tissue properties (quantification). In the case of UDFF, measurements of one or more tissue layers between the liver capsule and the transducer are fed into a model to estimate tissue properties (i.e., fat fraction). The model uses the input of measurements along with signals from the ROI and other information for quantification.

[0011] Modeling accuracy is improved when measurements of intervening layers are incorporated into a model for quantification (e.g., a UDFF model). For example, if an indicator is placed in the liver capsule, the thickness of the intervening tissue layer is determined. This thickness can be factored into models for assessing tissue properties, such as a UDFF model, to account for losses and / or wave distortion.

[0012] In another approach, the ROI is positioned automatically. The field of view is guided, and the ROI can be positioned appropriately within the desired field of view. To improve workflow and eliminate the need to measure the distance from the region of interest (ROI) to the capsule or the angle to the entrance, a liver capsule indicator (e.g., a "T" or "+") is incorporated into the ROI positioning. The liver capsule indicator is manually or automatically positioned on or along the liver capsule. Automatic positioning, such as by anatomical structure detection, reduces variability. This liver capsule indicator is used in conjunction with the integrated region of interest to define the region to be quantified. This region is scanned for quantification. The liver capsule indicator supports visual alignment and improves the reliability of the measurement ROI positioning. This indicator simplifies workflow and increases reliability in liver measurement or quantification (e.g., UDFF). Additionally, consistent positioning of the measurement ROI eliminates variability associated with ROI positioning.

[0013] FIG. 1 illustrates an example of a method for ultrasound imaging using an ultrasound scanner. An ultrasound quantification, such as fat fraction or elasticity, is calculated. This calculation incorporates one or more measurements of intervening layers of tissue into an artificial intelligence (AI) calculation of the region of interest. Modeling of the soft tissue property quantification is improved by considering the influence of intervening tissue in the quantification via input to the AI. Alternatively or additionally, an image processor consistently and appropriately positions the ROI through field of view guidance and / or anatomical structure detection. Anatomical structures distant from the ROI, such as the liver capsule, can be detected. The user can be guided to appropriately position the field of view and better position the ROI.

[0014] This method may be implemented by the system of Figure 3 or another system. A medical diagnostic ultrasound scanner performs measurements by acoustically generating waves with a beamformer and measuring the response. An image processor in the scanner, computer, server, or other device guides field of view positioning, detects anatomical structures, determines the location of indicators, lines, or ROIs, determines soft tissue properties (e.g., fat fraction, elasticity, or shear wave velocity), and generates images of quantification of the soft tissue properties. A display device, network, or memory is used to output the quantification images.

[0015] Additional or different steps, or fewer steps, may be provided. For example, if AI-based quantification using information from intervening tissue is used without visual field guidance and / or automated ROI positioning, steps 100, 110, and / or 120 are not provided. As another example, automated ROI positioning is provided when steps 100, 110, and / or 120 are used without AI-based quantification in step 140 and / or without measurements of intervening tissue in step 130. The steps are performed in the order described or illustrated (e.g., top to bottom or numerically), but may also be performed in other orders.

[0016] In process 100, an image processor guides the positioning of the ultrasound scanner and transducer field of view. The transducer is positioned relative to the patient's skin, and the sonographer may tilt, rotate, and / or translate the transducer to scan each part of the patient. The image processor guides the sonographer to position the field of view over appropriate soft tissue, such as the liver.

[0017] Various soft tissues may be of interest, such as the liver, kidneys, or other organs. The image processor may guide the sonographer by detecting the current field of view and / or by scoring whether the current field of view is the desired field of view for a given diagnostic test. For example, fat fraction is to be determined for the liver, so the field of view should include the liver imaged from an appropriate perspective. The orientation of the liver relative to the transducer may be important, such as positioning the liver capsule perpendicular to a line from the center of the transducer. The image processor guides the sonographer to move the transducer to provide the desired field of view.

[0018] In one approach, guidance relies on feedback on the current field of view: the field of view is examined to determine a quality score or other indicator of whether the field of view is adequate, and the user then refines the field of view by moving the transducer until the desired field of view is detected.

[0019] In another approach, guidance relies on feedback on how to adjust to have a better view. The image processor indicates how (e.g., rotate clockwise) and / or how much (e.g., 5 degrees) to move the transducer. For example, the user is guided to rock the transducer X degrees in a given direction.

[0020] The image processor is guided by an examination of the current field of view. One or more ultrasound images, such as B-mode images, are input. Template matching, pattern matching, location sensing, detection, segmentation, and / or other image processing may be used to identify anatomical structures represented in the images. This information may be used for scoring and / or to determine changes to be made.

[0021] In one approach, a machine learning model inspects the field and outputs feedback (e.g., quality / score for the current field and / or changes to be made to the next field). The machine learning model is trained using samples collected from sonographers locating and locating desired fields for diagnostic purposes (e.g., fat fraction determination). These samples include one or more images from the search, the expected final image, and vectors indicating transducer movements and modifications to reach the desired position. Machine learning uses optimization to learn values ​​for trainable parameters of a model, such as a neural network, that receives input (e.g., ultrasound images) and generates desired output (ground truth, such as a score or other quality of the current field, and / or modifications to reach the appropriate field). To indicate how and by how much to move the transducer, the machine learning model can be a policy learned through reinforcement learning. The policy can be implemented by a neural network with memory, such as long short-term memory. The training input is a sequence, and the policy learns what actions to take to progress to the desired field in the shortest time or fewest steps. With regard to scoring, the machine learning model can be a neural network, such as a convolutional neural network or a fully connected neural network, that outputs a score based on expert-curated ground truth scores used in training.

[0022] After training, the machine learning model outputs a quality (e.g., score) and / or change based on the input of the current image or a recent series of images. The current field of view is inspected by the machine learning model. The model itself outputs guidance, or the output from the model is used to generate output guidance for determining the position (pose) of the transducer to image the soft tissue of the target (e.g., the liver).

[0023] Guidance may include reducing the number or amount of shadowing (dark / black appearance) and / or blood vessels in the field of view. Shadowing and blood vessels near the region of interest are undesirable for soft tissue quantification. Shadowing caused by the patient's bone or other dense material can be reduced by moving the transducer (i.e., by changing the position of the field of view). The machine learning model may specifically detect and / or score shadowing and / or the amount of fluid (e.g., the number of blood vessels) and provide guidance for reducing them. Alternatively, the machine learning model may perform scoring in which the quality score is based in part on shadowing and / or blood vessels. For example, a sample may have a ground truth field of view (desired field of view) in which the transducer is moved to avoid fields with dark areas and / or blood vessels. The machine learning model is trained to score poorly and / or to prescribe changes when the field of view is inappropriate, such as when dark areas and / or blood vessels are present in the field of view.

[0024] In some approaches, the field of view is identified by a machine learning model, and guidance is provided to determine the transducer position to provide a quality image of the target anatomical structure (e.g., to limit dark areas and / or blood vessels in the field of view). In other approaches related to fat fractionation, the machine learning model identifies the liver capsule and guides the transducer position so that the liver capsule is appropriately located in the field of view (e.g., in the central third of the field of view with a boundary perpendicular to a line perpendicular to the center of the transducer).

[0025] Once the field of view is properly positioned, with or without guidance from an image processor-based inspection of the field of view, the ROI is located in step 120. The target anatomical structure, such as soft tissue (e.g., liver), is scanned to determine where quantification will be performed. The ROI should be located within the field of view. An ultrasound medical scanner scans the patient's tissue. A beamformer transmits acoustic beams and / or forms receive beams from the acoustic echoes. An array of transducer elements converts between acoustic and electrical energy. After B-mode detection, signals from the acoustic echoes are scan converted for imaging. The B-mode image is displayed to locate the ROI for quantification.

[0026] The ROI may be manually positioned by the sonographer and designates a region of the soft tissue of interest whose soft tissue properties are to be determined.

[0027] In another approach, ROI positioning is assisted. For example, an image processing device detects the location of the target anatomical structure in step 110. This location can be highlighted to assist ROI positioning. In the case of UDFF, the target anatomical structure is the liver capsule, and the liver capsule should be positioned near the lateral center of the ultrasound image by moving the transducer, for example, within the central one-third of the field of view. The liver capsule should also be substantially horizontal and perpendicular to a vertical line from the center of the transducer, with a ratio of substantially + / - 10 degrees. The liver capsule is the border of the liver and therefore should be substantially perpendicular to a line from the transducer through the center of the liver capsule.

[0028] The liver capsule and / or other anatomical structures are detected by pattern or template matching, filtering and thresholding, region reduction, skeletonization, and / or AI. For example, a machine learning model is trained with a large number of examples from previous patients curated by experts to detect the liver capsule in B-mode images of the liver. The location of the liver capsule is automatically detected in ultrasound images of the liver.

[0029] An example is shown in Figure 2. An ultrasound image 200 of the liver is shown. The liver capsule has been automatically detected. An indicator 230 has been placed on the detected liver capsule. The liver capsule is at the boundary between the liver and the transducer 220. One or more layers of intervening tissue 210 are between the liver capsule and the transducer 220. These layers of intervening tissue 210 are identified and differentiated, allowing the system to better understand the acoustic properties and boundaries of each layer.

[0030] In step 120, the image processing device automatically locates a line 240 from the transducer 220 through the liver capsule, an indicator 230 on the liver capsule, and an ROI 250 within the liver along the line 240. The size of the ROI 250 and / or its distance from the liver capsule can be set to a default or based on the detection of other tissues (e.g., 1 cm). The line 240 runs from the center of the transducer 220 through the liver capsule. The goal is to have a line that is perpendicular to the liver capsule and the transducer 220. The ROI 250 is located within the liver and is separated from the liver capsule by, for example, 1 cm or another distance depending on the setting. The user can change the position of the transducer 220, the line 240, the indicator 230, and / or the ROI.

[0031] The automatic detection of the liver capsule and the automatic positioning of the indicator 230, line 240, and / or ROI 250 based on the location of anatomical structures away from the ROI provide a real-time visualization tool to assist the clinician in determining what to quantify. This real-time visualization tool also assists the clinician in identifying and assessing intervening tissue layers 210 during the ultrasound examination, providing added-value context and improving the overall accuracy of the system.

[0032] Imaging is performed while positioning the ROI and measuring and / or quantitating intervening tissue. For example, B-mode imaging is performed to position and maintain the transducer's field of view and ROI at the desired tissue. Once the field of view and ROI are at the desired location within the patient, a scan is performed to estimate tissue properties. Imaging continues while the tissue properties are quantified. Alternatively, imaging is terminated while the tissue properties are quantified.

[0033] In step 130, the ultrasound scanner measures tissue between the ROI (e.g., the liver) and the ultrasound scanner's transducer 220. The intervening tissue is between the boundary of the soft tissue being quantified and the transducer, for example, the intervening tissue from the ROI 250 or liver capsule to the transducer 220. The intervening tissue measured or revealed may or may not include some of the intervening layers present, such as skin. The intervening tissue may be all or only some of the soft tissue and / or fluid through which echoes from the ROI path pass.

[0034] One or more tissue layers indicated by boundary markers (e.g., liver capsule with respect to the fat fraction) allow for identification and differentiation of various tissue layers during ultrasound examination, allowing the system to better understand the acoustic properties and boundaries of each layer of tissue measured from the organ of interest to the transducer.

[0035] The measurements are of any one or more characteristics of the intervening tissue. For the measurements, one or more tissue layers or the entire intervening tissue are detected and / or segmented. Filtering and thresholding, pattern or template matching, skeletonization, and / or AI perform the detection and / or segmentation. Image processing or machine learning-based models automatically segment and / or distinguish tissue layers, reducing the need for manual intervention and streamlining the estimation process. Alternatively, detection of the liver capsule or other anatomical structures relative to the transducer is used without additional detection and / or segmentation of the intervening tissue.

[0036] The measurement can be thickness. The thickness of each layer of tissue and / or the total thickness of intervening tissue is measured. Positioning of the ROI and / or detection of anatomical structures relative to the transducer indicates thickness. The thickness of intervening tissue layers can be incorporated into tissue property estimates, allowing for more accurate calculation of UDFF and other properties by accounting for the effect of layer thickness on acoustic properties.

[0037] In another approach, the measurements are backscatter and / or attenuation coefficients for the intervening tissue as a whole, for each layer, and / or for each location (e.g., pixel or voxel). The backscatter and / or attenuation coefficients can be used to determine losses (e.g., due to attenuation) and / or wave distortion (e.g., due to backscatter level) in the intervening tissue, so that quantification in the ROI is more accurate. For example, the level of backscatter through the intervening tissue can affect the level of attenuation in the ROI. Rather than using local attenuation and / or backscatter in the ROI, the overall attenuation and / or backscatter from the transducer to the tissue of the ROI can be considered.

[0038] In another approach, the image processor identifies tissue types within intervening tissue layers. Each type of tissue (e.g., skin, fat, muscle, other organs) has its own acoustic properties (e.g., different attenuation and / or backscatter). The measurement is the tissue type. A machine learning model, an in-image view combined with a knowledge base (e.g., template matching or lookup), and / or other detection or segmentation is used to distinguish between individual tissue layers and the corresponding tissue type for each layer. The tissue type can be used to look up the acoustic properties of that type of tissue from a database. The acoustic properties can be used as the measurement. The measurements can be the tissue type, the number of layers, the thickness of each layer, the acoustic properties of each layer, and / or other information for each layer.

[0039] Once the location of the ROI is determined in step 120, an ultrasound scanner scans the ROI, such as scanning a portion of the liver, in step 135, with or without measurements of intervening tissue in step 130. The scan is configured to determine tissue properties. For example, multiple scans are performed to determine UDFF. Scans are performed for backscatter calculations, attenuation calculations, elasticity, and / or shear wave velocity.

[0040] In step 140, the image processor determines (estimates) tissue properties of tissue within the patient's ROI. For example, the UDFF of the patient's liver is determined. Elasticity, shear wave propagation (e.g., velocity), or other tissue properties may be determined. Quantification is more than just B-mode (echo intensity) or Doppler mode (e.g., velocity, power, or variability of tissue motion), but rather represents properties of the tissue itself. Fat fraction, shear propagation, or elasticity may be determined. Because different tissues respond differently to shear waves (rather than ultrasound), shear wave velocity is indicative of tissue properties.

[0041] Fat fraction is used as an example herein. The fat fraction is patient-specific. One or more characteristics of the patient are used to derive the fat fraction for that patient. Some patients may have the same or similar fat fraction, while different patients may have different fat fractions. In one embodiment, the tissue is the patient's liver. The fat fraction of the patient's liver is obtained. The liver example is used here. In other embodiments, the tissue is the patient's kidney, bladder, breast, heart, muscle, or other soft tissue.

[0042] The determination is made by scanning the patient. The fat fraction may be obtained from a magnetic resonance scan of the patient, such as measured by a magnetic resonance proton density fat fraction (MR-PDFF) scan. Other modalities may also be used to measure or estimate the fat fraction. In another example, the fat fraction is obtained from an ultrasound scan. An ultrasound scan to estimate or measure the fat fraction may be more cost- and time-efficient and convenient.

[0043] For ultrasound estimation of tissue fat fraction (i.e., UDFF), an ultrasound medical scanner determines scattering and attenuation from scanning the tissue. Other combinations of quantitative ultrasound parameters may also be used. The complexity of human tissue may be measured using multiple quantitative ultrasound parameters for accurate characterization of that tissue. For example, liver fat fraction is estimated using a multiparametric approach that combines quantitative parameters extracted from the received signal of various wave phenomena, such as longitudinal wave scattering and attenuation from acoustic radiation force impulse (ARFI) excitation, shear wave propagation and attenuation, and / or on-axis wave propagation and attenuation. U.S. Patent Application Publication US2018 / 0289323A1 discloses estimation of fat fraction using ultrasound.

[0044] In one embodiment, liver fat fraction is estimated by transmitting and receiving a sequence of pulses to estimate scattering parameters and a sequence of pulses to obtain shear wave parameters. This estimation may include transmitting and receiving a sequence of pulses to estimate parameters from axial displacements caused by acoustic radiation force impulses (ARFIs). The parameters are estimated and combined to estimate fat fraction. Other information, such as non-ultrasound data (e.g., blood biomarkers), may also be included in the fat fraction estimation. A lookup table based on empirical studies may be used to relate values ​​of various parameters to fat fraction values. Alternatively, a machine learning model is trained using samples from various patients of scattering, attenuation, shear wave velocity, and / or other measurements for an ROI along with ground truth fat fractions based on magnetic resonance and / or biopsies. The machine learning model outputs a UDFF in response to input of measurements from the ROI.

[0045] Other inputs, such as input of measurements from intervening tissues, may also be used. The machine learning model is trained using training data of samples of inputs (ROI and intervening tissue measurements) and optimization from ground truth outputs. This training configures the machine learning model (e.g., by determining values ​​for trainable parameters) to generate outputs (e.g., UDFFs) in response to inputs (e.g., signals or measurements from the ROI and signals or measurements from the intervening tissues). Information from scanning the intervening tissues is input, such as inputting tissue type, thickness, backscatter coefficient, and / or attenuation coefficient, either as a whole or layer by layer. Information from scanning the ROI is input, such as inputting return signal, attenuation, shear wave velocity, and / or backscatter.

[0046] Information from the ROI is information from the liver that is used to calculate UDFF or other quantification (e.g., elasticity). Information from intervening tissue is information that accounts for losses and / or wave distortions caused by the tissue, which may affect quantification in the ROI. The machine learning model adapts to various tissue types. By including information from intervening tissues in the training, the machine learning model is trained to automatically adjust for different tissue types and their acoustic properties, improving the accuracy of the estimated ROI quantification, even when dealing with diverse and complex tissue structures. This compensation mechanism corrects for the effect of intervening layers on the ultrasound signal, allowing for more accurate estimation of tissue acoustic properties (e.g., backscatter coefficient and attenuation used in fat fraction estimation).

[0047] In one approach, the training samples are from a patient database. Other information can also be included. For example, a database of common tissue types and acoustic properties is used to create samples by ultrasound simulation. Commonly encountered tissue types and their respective acoustic properties are used to allow for more accurate estimation.

[0048] The quantification may be of fat fraction, elasticity, shear, or other tissue properties of the soft tissue in the region of interest. The quantification may be performed by a processor application of a machine learning model. In response to inputs such as properties of the intervening tissue and properties of the ROI, the machine learning model generates an output quantification (e.g., a UDFF value).

[0049] In UDFF, an ultrasound scanner generates measurements of scattering in tissue from a scan of a patient. To measure scattering, the ultrasound scanner scans the tissue with ultrasound. A series of transmit and receive events are performed to acquire signals for estimating quantitative ultrasound scattering parameters. The scattering measurements measure the tissue response to longitudinal waves transmitted from the ultrasound scanner. The scattering, or echoes, of the longitudinal waves striking the tissue are measured.

[0050] The scatter is greater than or equal to the intensity of the reflection (i.e., greater than or equal to the B-mode data). Any measure of scatter can be used, such as the spectral slope of the logarithm of the frequency-dependent backscatter coefficient. For example, the attenuation coefficient is measured. A reference phantom method is used, but other measures of the attenuation coefficient can also be used. The acoustic energy has an exponential decay with depth. Measurements of acoustic intensity as a function of depth are performed before or without depth gain correction. To remove system effects, the measurements are calibrated based on measurements of acoustic intensity as a function of depth in a phantom. Measurements can be made less susceptible to noise by averaging over one-, two-, or three-dimensional regions. The beamformed samples or acoustic intensity can be converted to the frequency domain, and calculations can be performed in the frequency domain.

[0051] Attenuation is measured as the slope of intensity as a function of depth. Other measures of attenuation, such as shear wave attenuation over distance or time, can also be used. Tissue displacement as a function of depth from the ARFI-induced longitudinal wave can be used to find tissue attenuation. The maximum displacement, the displacement as a function of depth, and / or the amount of displacement as a function of time are used to calculate attenuation. Other propagation measurements can be used instead of or as attenuation. For example, measurements of shear wave propagation or measurements of on-axis displacement (e.g., ARFI measurements) can be used.

[0052] Fat fraction can be determined from attenuation and scattering or other combinations of quantitative ultrasound information. Attenuation or propagation and backscatter measurement values ​​are input into a machine learning classifier or lookup table that outputs a value for fat fraction. A machine-trained classifier provides a nonlinear model. A lookup table may provide a linear model. A predetermined or preprogrammed function relates input values ​​to output values. The function and / or weights used for the function can be empirically determined. For example, the weights are obtained by least-squares minimization using magnetic resonance-proton density fat fraction (MR-PDFF) values ​​or biopsy values ​​of fat fraction. Other approaches for measuring fat fraction with ultrasound medical scanners can also be used.

[0053] A tissue property (e.g., UDFF) is determined for the ROI. One value may be determined for the entire ROI. In other approaches, separate determinations are made for different subregions. A field of values ​​(e.g., UDFF values) distributed within the ROI is determined. Figure 2 shows 5x3 subdivisions of the ROI. A value is determined for each subdivision in the ROI of an organ (e.g., the liver). Even higher resolution may be used.

[0054] In step 150, the ultrasound medical scanner or processor (e.g., an image processing unit of a medical scanner) generates an image of the determined quantification. The generated image is transmitted to a display, memory, or computer network. When displayed, the generated image is displayed on a display device to provide diagnostically useful quantification characterizing the tissue within the ROI.

[0055] The image is a graph, alphanumeric text, and / or tissue representation. In one approach, quantification or tissue characterization values ​​are displayed as annotations or numbers overlaid on the B-mode image. The overlay includes a graphic for the ROI. In another approach, the ROI is coded (e.g., colored or highlighted) with one or more quantities (e.g., UDFF values). For example, each of the subregions is color-coded based on magnitude. The color coding highlights each subregion according to its value. In yet another approach, the user selects a location within the ROI. The quantity determined for the selected location is displayed as text or as a marker along a scale. The image is a representation of the shear, elasticity, and / or fat fraction determined for the ROI.

[0056] Figure 3 illustrates one embodiment of a system 300 for ultrasound medical imaging. System 300 performs the method of Figure 1 or other methods. System 300 uses (1) measurements of intervening tissue as input to AI for quantification in the ROI, (2) automatic location of the ROI, (3) automatic location of a liver capsule indicator, and / or (4) field of view guidance (e.g., by image scoring). Any of these can be used to provide consistent and / or accurate quantification of soft tissue (e.g., UDFF or elasticity).

[0057] The system 300 includes a transmit beamformer 310, a transducer (XDCR) 320, a receive beamformer 330, an image processor 340, a display 360, and a memory 350. Additional or different components may be provided, or fewer components may be provided. For example, a user input device may be provided for user interaction with the system.

[0058] System 300 is a medical diagnostic ultrasound imaging system. In alternative embodiments, system 300 may include a computer, workstation, picture archiving and communication system (PACS) station, or other device configuration for real-time or post-acquisition imaging, either co-located or distributed over a network.

[0059] The transmit and receive beamformers 310, 330 form beamformers that scan (e.g., transmit and receive activation) using the transducer 320. A series of pulses are transmitted and responses are received based on the activation or configuration of the beamformers. The beamformers scan to determine quantification and imaging tissue.

[0060] The transmit beamformer 310 may be an ultrasound transmitter, memory, pulse generator, analog circuitry, digital circuitry, or a combination thereof. The transmit beamformer 310 is operable to generate waveforms for multiple channels with individual or relative amplitudes, delays, and / or phases. Acoustic waves are transmitted from the transducer 320 in response to the generated electrical waveforms, forming one or more beams.

[0061] The transducer 320 is a 1D, 1.25D, 1.5D, 1.75D, or 2D array of piezoelectric or capacitive membrane elements. The transducer 320 contains multiple elements for transducing between acoustic and electrical energy. Receive signals are generated in response to ultrasound energy (echoes) striking each element of the transducer 320. These elements connect to channels of the transmit beamformer 310 and the receive beamformer 330. Alternatively, a single element with mechanical focus may be used.

[0062] The receive beamformer 330 includes multiple channels with amplifiers, delays, and / or phase rotators, and one or more summers. Each channel connects to one or more transducer elements. The receive beamformer 330 is configured in hardware, firmware, or software and applies relative delays, phases, and / or apodization to form one or more receive beams in response to each imaging transmission. The receive beamformer 330 uses the received signals to output data representing spatial locations. The relative delays and / or phasing and summation of the signals from the individual elements provide the beamforming.

[0063] In cooperation with the transmit beamformer 310, the receive beamformer 330 generates data representing a field of view. Data (e.g., beamformed samples) are generated by scanning the field of view using ultrasound. By repeating the scan, ultrasound data representing the field of view at different times is acquired. The transmit beamformer 310 and the receive beamformer 330 are configured to scan the patient's soft tissue together with the transducer 320. The field of view for scanning the soft tissue is established by the position (orientation) of the transducer 320 relative to the patient. The transducer 320 can be moved to change the field of view until the target soft tissue is captured. An ROI can then be positioned within the soft tissue to scan the soft tissue. Another scan, such as another transmit and receive sequence, can be used for the ROI other than for other locations. Similarly, another scan can be used for segmented or identified intervening tissue regions. A scan sequence is performed to obtain measurements. A scan sequence is performed to perform quantification in the ROI. A B-mode scan is performed for other locations within the field of view.

[0064] The image processor 340 may include a B-mode detector, a Doppler detector, and / or a pulsed wave Doppler detector for detecting and processing information for display from beamformed ultrasound samples. The image processor 340 may include a processor that may implement such detectors or may be separate from the detectors. The processor may be a control processor, a general-purpose processor, a digital signal processor, a tensor processor, a graphics processing unit, an application-specific integrated circuit, a field-programmable gate array, a network, a server, a cluster of processors, combinations thereof, or other now-known or later-developed device for quantifying tissue properties from ultrasound data (e.g., beamformed data or detected data). The image processor 340 may be configured with hardware, firmware, and / or software to perform any combination of one or more of the processes shown in FIG. 1 .

[0065] In one approach, the image processor 340 is configured to guide the field of view of the transducer 320 to position the region of interest. The image processor 340 may be configured with a machine learning model 355. The model 355 is trained to receive images related to the current field of view and score the quality of the image. The quality may indicate viewing an organ from a desired perspective, the number or amount of shadowing, and / or the number of blood vessels in the field of view. By communicating the quality to the user, the user may move the transducer to locate a field of view with sufficient or best quality. In another approach, the machine learning model 355 is trained (configured) to indicate the movement and / or magnitude of movement to reposition the field of view given an input of the current field of view or a series of views.

[0066] In another approach, the image processor 340 is configured to position the region of interest. Given an image of the current field of view, the image processor 340 is configured to position an ROI within the field of view. Segmentation or detection of an anatomical structure (e.g., an organ of interest) may be used. Alternatively, a machine learning model 355 is trained (configured) to position an ROI given an input image. In the case of UDFF, the image processor 340 is configured to automatically detect the liver capsule. The soft tissue of interest is the liver. The ROI should be positioned within the liver relative to the liver capsule. The liver capsule is detected. A liver capsule indicator and / or a line through the liver capsule can be positioned by the image processor based on the position of the detected liver capsule. The ROI is positioned relative to the liver capsule, such as laterally centered on a line from the transducer to the liver capsule and at a default distance from the liver capsule. The image processor 340 automatically positions the ROI based on the detected liver capsule. The user can adjust the positioning. In an alternative approach, the ROI and / or liver capsule indicator are positioned manually by the user.

[0067] In another approach, the image processor 340 is configured to quantify soft tissue properties from scans of the soft tissue. An ROI is scanned. Quantification is performed using scan data and / or measurements derived from the scan data (e.g., backscatter coefficient, attenuation coefficient, and / or shear wave velocity). The properties quantified are UDFF, shear wave properties, and / or elastography properties. In one embodiment, the image processor 340 is configured with a machine learning model 355 trained to output values ​​of the properties in response to inputs such as scan data and / or measurements from the ROI.

[0068] In another embodiment, the image processor 340 is configured with a machine learning model 355 that outputs quantification in response to input of information from the ROI and information from the intervening tissue. The image processor 340 identifies the intervening tissue and receives measurements of the intervening tissue. The measurements characterize the intervening tissue, such as tissue type, tissue acoustic properties, tissue layer, tissue thickness, tissue attenuation, and / or tissue backscatter. Measurements of one or more of these properties are input to the machine learning model 355. Through previous training of the model 355, the model 355 has been trained to account for loss and / or wave distortion of the intervening tissue. This accounting does not directly calculate loss and / or wave distortion. In other embodiments, loss and / or wave distortion are calculated or estimated, and measurements of the loss or wave distortion are input. By accounting for the acoustic effects of the intervening tissue (e.g., between the transducer and the liver capsule or between the transducer and the ROI), more accurate quantification may be provided.

[0069] The image processor 340 is configured to generate one or more images. For example, B-mode, contrast, M-mode, flow, color mode, and / or other types of images may be generated. Quantification (e.g., UDFF) may be presented as an overlay in the image. The quantification may change color at a location within the ROI or be an annotation on the image. The quantification may be included in the image or displayed sequentially or substantially simultaneously. For example, a tissue property estimation image may be displayed simultaneously with other images. One or more values ​​of fat fraction or other quantification may be mapped to display information. If fat fraction or other quantification is measured at different locations, the values ​​may be generated as a color overlay on the region of interest in the B-mode image. Shear wave velocity and fat fraction or other quantification data may be combined as a single overlay on a B-mode image. Alternatively, the fat fraction or quantification value may be displayed as text or numerical values ​​adjacent to or overlaid on the B-mode or shear wave imaging image.

[0070] The image processor 340 may be configured to generate other displays, such as a shear wave velocity image displayed next to a graph, text, or graphical indicator of fat fraction or other quantification. Rather than being in a separate two-dimensional or three-dimensional representation, the quantification information is presented at one or more locations in the region of interest, such as when the user selects a location, and the ultrasound scanner presents the fat fraction at that location.

[0071] The image processor 340 operates according to instructions stored in memory 350 or other memory. Memory 350 is a non-transitory computer-readable storage medium. Instructions for executing the processes, methods, and / or techniques described herein may be provided in a computer-readable storage medium or memory, such as a cache, buffer, RAM, removable media, a hard drive, or other computer-readable storage medium. Computer-readable storage media include various types of volatile and non-volatile storage media. The functions, processes, or tasks illustrated in the figures or described herein are performed in response to one or more instruction sets stored in a computer-readable storage medium. These functions, processes, or tasks are independent of the particular type of instruction set, storage medium, processor, or processing strategy and may be performed by software, hardware, integrated circuits, firmware, microcode, etc., operating alone or in combination. Similarly, processing strategies may include multiprocessing, multitasking, parallel processing, etc.

[0072] In one embodiment, the instructions are stored on a removable media device for reading by a local or remote system, in another embodiment, the instructions are stored remotely for transfer over a computer network or over telephone lines, and in yet another embodiment, the instructions are stored within a given computer, CPU, GPU, or system.

[0073] Display 360 is a device for displaying one-dimensional or two-dimensional images or three-dimensional representations, such as a CRT, LCD, projector, plasma, or other display. Two-dimensional images represent spatial distribution within an area. Three-dimensional representations are rendered from data representing spatial distribution within a volume. Display 360 is configured by image processor 340 or other device upon input of data to be displayed as an image. Display 360 displays images representing tissue, such as the liver. Display 360 is configured to display ultrasound images showing quantification of soft tissue properties in a ROI.

[0074] Various exemplary embodiments are summarized below. The combinations shown or other combinations may be provided. Forms, functions, or approaches in one type of exemplary embodiment (e.g., a method or a system) may also be used in another type (e.g., a system or a method). Any of the forms, functions, or approaches may be used in a computer program (product) or instructions stored on a non-transitory computer-readable storage medium.

[0075] Exemplary embodiment 1: 1. A method of ultrasound imaging using an ultrasound scanner, comprising: measuring with the ultrasound scanner the tissue between the liver and a transducer of the ultrasound scanner; scanning a region of interest within the liver with the ultrasound scanner; determining an ultrasound-derived fat fraction (UDFF) of the liver using a first machine learning model configured to receive the tissue measurements and information from the scan and output the UDFF; displaying the UDFF.

[0076] Exemplary embodiment 2: 2. The method of Exemplary Embodiment 1, wherein the measuring step includes measuring a thickness as the measurement value.

[0077] Exemplary embodiment 3: the measuring step includes measuring a backscattering coefficient and / or attenuation of the tissue as the measurement value; The method of any of Exemplary Embodiments 1-2, wherein the determining includes determining a region where the information includes the backscatter coefficient and / or attenuation of the liver.

[0078] Exemplary embodiment 4: the measuring step includes measuring the tissue between a liver capsule of the liver and the transducer; The method of any of Exemplary Embodiments 1-3, wherein the liver capsule is indicated by an indicator on a display.

[0079] Exemplary embodiment 5: The method of any of Exemplary Embodiments 1-4, wherein the measuring includes measuring acoustic properties based on the type of the tissue as the measurement.

[0080] Exemplary embodiment 6: the measuring includes identifying each tissue layer of the tissue; The method of any of exemplary embodiments 1-5, wherein the measurements are from each of the tissue layers.

[0081] Exemplary embodiment 7: 7. The method of exemplary embodiment 6, wherein each property of each tissue layer is input as the measurement into the first machine learning model.

[0082] Exemplary embodiment 8: The method of any of Exemplary Embodiments 1-7, wherein the first machine learning model is configured to account for losses and / or wave distortions caused by the tissue through training.

[0083] Exemplary embodiment 9: automatically detecting the location of the liver capsule within the ultrasound image of the liver; automatically locating a line from the transducer through the liver capsule, an indicator on the liver capsule, and the region of interest within the liver along the line; The method of any one of exemplary embodiments 1 to 8, wherein determining the UDFF includes determining a UDFF within the region of interest.

[0084] Exemplary embodiment 10: testing the visual field with a second machine learning model; 10. The method of any of exemplary embodiments 1-9, further comprising: outputting guidance for positioning the transducer to image the liver based on an output of the second machine learning model.

[0085] Exemplary embodiment 11: said inspecting includes inspecting for dark areas and / or blood vessels; The method of exemplary embodiment 10, wherein the induction reduces the dark areas and / or blood vessels in the field of view.

[0086] Exemplary embodiment 12: 11. The method of Exemplary Embodiment 10, wherein the inspecting includes scoring the field of view for automatic location of the region of interest.

[0087] Exemplary embodiment 13: determining the UDFF as a field of UDFF values ​​distributed in the region of interest within the liver; 13. The method of any of Example Embodiments 1-12, wherein the displaying includes displaying an ultrasound image with the region of interest coded by the UDFF values.

[0088] Exemplary embodiment 14: 1. A system for ultrasound medical imaging, comprising: A transducer; a beamformer configured to scan soft tissue of a patient using the transducer, the soft tissue being in a region of interest, and configured to scan intervening tissue between the transducer and the soft tissue with the transducer; an image processing device configured to locate the region of interest and configured to quantify a first property of the soft tissue from the scan of the soft tissue with a first machine learning model, the first property comprising ultrasound-guided fat fraction, shear wave properties, and / or elastography properties, the first machine learning model receiving a second property of the intervening tissue as an input for outputting the first property; a display configured to display an ultrasound image indicative of the quantification of the first property of the soft tissue.

[0089] Exemplary embodiment 15: The system of exemplary embodiment 14, wherein the first machine learning model is configured to account for loss of intervening tissue and / or wave distortion through training.

[0090] Exemplary embodiment 16: 16. The system of any of exemplary embodiments 14-15, wherein the image processing device is configured to guide the field of view of the transducer to locate the region of interest by a second machine learning model.

[0091] Exemplary embodiment 17: the soft tissue comprises a liver; the intervening tissue includes tissue between the liver capsule and the transducer; 17. The system of any one of exemplary embodiments 14-16, wherein the image processing device is configured to automatically detect the liver capsule and automatically position the region of interest based on the detected liver capsule.

[0092] Exemplary embodiment 18: 1. A method for ultrasonic quantification of soft tissue properties using an ultrasound scanner, comprising: Locating a region of interest within the soft tissue by imaging device detection of anatomical structures; quantitating the shear, elasticity, and / or fat fraction of the soft tissue within the region of interest; Displaying the shear, elasticity, and / or fat fraction of regional tissue within the region of interest.

[0093] Exemplary embodiment 19: The determining step includes: identifying the field of view with a first machine learning model; 20. The method of Exemplary Embodiment 18, comprising guiding transducer positioning to limit dark areas and / or blood vessels within the field of view.

[0094] Exemplary embodiment 20: The determining step includes: identifying the liver capsule with a first machine learning model; 20. The method of any of exemplary embodiments 18-19, comprising guiding the positioning of the transducer so that the liver capsule is within the central lateral third of the ultrasound image of the liver and so that the liver capsule is substantially perpendicular to a line from the center of the transducer to the liver capsule.

[0095] Exemplary embodiment 21: measuring a first property of an intervening tissue between a transducer of the ultrasound scanner and the region of interest; 21. The method of any of exemplary embodiments 18-20, wherein the quantifying includes quantifying by a machine learning model responsive to an input of the first characteristic.

[0096] While the present invention has been described above with reference to various embodiments, it will be understood that many changes and modifications can be made without departing from the scope of the invention. Accordingly, the foregoing detailed description is to be interpreted as illustrative rather than limiting, and it is the following claims, including all equivalents, that are intended to define the spirit and scope of the invention.

Claims

1. 1. A method of ultrasound imaging using an ultrasound scanner, comprising: measuring with the ultrasound scanner the tissue between the liver and a transducer of the ultrasound scanner; scanning a region of interest within the liver with the ultrasound scanner; determining an ultrasound-derived fat fraction (UDFF) of the liver using a first machine learning model configured to receive the tissue measurements and information from the scan and output the UDFF; displaying the UDFF.

2. The method of claim 1 , wherein the measuring step includes measuring a thickness as the measurement value.

3. the measuring step includes measuring a backscattering coefficient and / or attenuation of the tissue as the measurement value; The method of claim 1 , wherein said determining comprises determining a region where said information includes said backscatter coefficient and / or attenuation of said liver.

4. the measuring step includes measuring the tissue between a liver capsule of the liver and the transducer; The method of claim 1 , wherein the liver capsule is indicated by an indicator on a display.

5. The method of claim 1 , wherein the measuring comprises measuring acoustic properties based on the tissue type as the measurement.

6. the measuring includes identifying each tissue layer of the tissue; The method of claim 1 , wherein the measurements are from each of the tissue layers.

7. The method of claim 6 , wherein each property of each tissue layer is input as the measurement into the first machine learning model.

8. The method of claim 1 , wherein the first machine learning model is configured through training to account for losses and / or wave distortions caused by the tissue.

9. automatically detecting the location of the liver capsule within the ultrasound image of the liver; automatically locating a line from the transducer through the liver capsule, an indicator on the liver capsule, and the region of interest within the liver along the line; The method of claim 1 , wherein determining the UDFF comprises determining a UDFF within the region of interest.

10. examining the visual field with a second machine learning model; 10. The method of claim 1, further comprising: outputting guidance for positioning the transducer to image the liver based on an output of the second machine learning model.

11. said inspecting includes inspecting for dark areas and / or blood vessels; The method of claim 10 , wherein the induction reduces the dark areas and / or blood vessels in the field of view.

12. The method of claim 10 , wherein the inspecting comprises scoring the field of view for automatic location of the region of interest.

13. determining the UDFF as a field of UDFF values ​​distributed in the region of interest within the liver; The method of claim 1 , wherein the displaying comprises displaying an ultrasound image with the region of interest coded with the UDFF values.

14. 1. A system for ultrasound medical imaging, comprising: A transducer; a beamformer configured to scan soft tissue of a patient using the transducer, the soft tissue being in a region of interest, and configured to scan intervening tissue between the transducer and the soft tissue with the transducer; an image processing device configured to locate the region of interest and configured to quantify a first property of the soft tissue from the scan of the soft tissue with a first machine learning model, the first property comprising ultrasound-guided fat fraction, shear wave properties, and / or elastography properties, the first machine learning model receiving a second property of the intervening tissue as an input for outputting the first property; a display configured to display an ultrasound image indicative of the quantification of the first property of the soft tissue.

15. 15. The system of claim 14, wherein the first machine learning model is configured to account for the loss of intervening tissue and / or wave distortion upon training.

16. The system of claim 14 , wherein the image processor is configured to guide the field of view of the transducer to locate the region of interest with a second machine learning model.

17. the soft tissue comprises a liver; the intervening tissue includes tissue between the liver capsule and the transducer; The system of claim 14 , wherein the image processor is configured to automatically detect the liver capsule and automatically position the region of interest based on the detected liver capsule.

18. 1. A method for ultrasonic quantification of soft tissue properties using an ultrasound scanner, comprising: Locating a region of interest within the soft tissue by imaging device detection of anatomical structures; quantification of the shear, elasticity, and / or fat fraction of the soft tissue within the region of interest; Displaying the shear, elasticity, and / or fat fraction of the soft tissue within the region of interest.

19. The determining the position includes: identifying the field of view with a first machine learning model; 20. The method of claim 18, comprising guiding transducer positioning to limit dark areas and / or blood vessels in the field of view.

20. The determining the position includes: identifying the liver capsule with a first machine learning model; 20. The method of claim 18, comprising guiding the positioning of the transducer so that the liver capsule is within the central lateral third of an ultrasound image of the liver and the liver capsule is substantially perpendicular to a line from the center of the transducer to the liver capsule.

21. measuring a first property of an intervening tissue between a transducer of the ultrasound scanner and the region of interest; The method of claim 18 , wherein the quantifying comprises quantifying with a machine learning model responsive to an input of the first characteristic.

Images