Calculation and inference method for physical characteristics of tissue, and related device

Abstract

Disclosed in the present invention are a calculation and inference method for the physical characteristics of tissue, and a related device. The method comprises: acquiring a data set, and training and testing a generative deep neural network model on the basis of the data set, so as to obtain a target network model; selecting target tissue, selecting a target area from the target tissue, scanning the target area to obtain tissue imaging information under the target area, and extracting an ultrasonic echo signal from the tissue imaging information; inputting the ultrasonic echo signal into the target network model, and simulating a wave transmission disturbance map of a mechanical wave in the target area; and obtaining trajectory information of the mechanical wave on the basis of the wave transmission disturbance map, and then obtaining the physical characteristics of the tissue in the target area on the basis of the trajectory information. In the present invention, a propagation disturbance process of the mechanical wave in the target area is simulated, and thus the wave transmission disturbance map can be obtained without actually generating such a mechanical wave disturbance, and the propagation speed of the mechanical wave can be obtained from the wave transmission disturbance map, and then physical properties related to the propagation speed are obtained.

Description

A computational reasoning method for tissue physical properties and related equipment Technical Field

[0001] The present invention relates to the technical field of tissue property analysis, and in particular to a computational reasoning method, system, terminal, and computer-readable storage medium for tissue physical properties. Background Art

[0002] Physical property measurement techniques can be applied to a wide range of tissues, including soft tissues that perform biomechanical functions and tissues with biomechanical properties, such as visceral tissue, muscle, and skin. Tissue physical property measurement has a wide range of applications. Generally speaking, it can be applied to basic research, such as analyzing how tissue physical properties change with age, thereby gaining insights into the nature of certain physiological phenomena.

[0003] Wave propagation disturbance images of tissues can be used to analyze the physical properties of tissues. Currently, there are various methods for measuring the physical properties of tissues. Non-invasive instruments and equipment for measuring the physical properties of tissues are expensive, usually large in size, and not easy to carry. In addition, the images collected by the instruments are subject to interference from mechanical wave signals, resulting in poor image quality. At the same time, such equipment often requires the operator to have a certain amount of experience in using the instrument, and the operator's subjectivity affects the imaging quality. These reasons limit the use scenarios of the instrument. The new method of using neural networks to generate tissue elasticity imaging has not yet been widely used, and this method is currently only applicable to the measurement of the physical properties of a single tissue. The extraction of local structural information of the tissue is still limited, and more tissue measurement usage scenarios are still waiting to be explored.

[0004] Therefore, the existing technology still needs to be improved and developed.

[0005] Summary of the Invention

[0006] The main purpose of the present invention is to provide a computational reasoning method, system, terminal and computer-readable storage medium for tissue physical properties, aiming to solve the problem in the prior art that the measurement of tissue physical properties has high equipment requirements and high equipment costs, and that ordinary ultrasound probes cannot be used to accurately measure tissue physical properties.

[0007] To achieve the above-mentioned object, the present invention provides a method for calculating and reasoning about the physical properties of tissues, the method comprising the following steps:

[0008] Obtaining a data set, and training and testing a generative deep neural network model based on the data set to obtain a target network model;

[0009] selecting a target tissue, selecting a target area from the target tissue, scanning the target area, obtaining tissue imaging information under the target area, and extracting an ultrasonic echo signal of the target area from the tissue imaging information;

[0010] Inputting the ultrasonic echo signal into the target network model to simulate a wave propagation disturbance diagram of the mechanical wave in the target area;

[0011] The trajectory information of the mechanical wave is obtained according to the wave transmission disturbance map, the transmission speed of the mechanical wave is calculated according to the trajectory information, and the tissue physical properties of the target area are calculated according to the transmission speed.

[0012] Optionally, in the computational reasoning method for tissue physical properties, the step of acquiring a data set specifically includes:

[0013] Using a dedicated transient elastic imaging device to measure a number of tissues used for training, ultrasonic echo signals and real wave transmission disturbance maps corresponding to the tissues used for training are obtained;

[0014] Using the ultrasonic echo signal as a feature and the real wave transfer disturbance map as a true label, generating a data set according to the feature and the true label;

[0015] The data set is divided into two parts according to a preset ratio, one part is used as a training set to train the generative deep neural network model, and the other part is used as a test set to detect the accuracy of the prediction results of the generative deep neural network model.

[0016] Optionally, in the computational reasoning method for tissue physical properties, wherein the generative deep neural network model includes a generator and a discriminator, the training and testing of the generative deep neural network model based on the data set to obtain a target network model specifically includes:

[0017] Inputting the training set into the generator for computational reasoning, and generating a plurality of virtual wave transfer disturbance maps through the generator;

[0018] Using the discriminator, the plurality of virtual wave transfer perturbation maps and the plurality of real labels are distinguished, and a tensor is output;

[0019] Calculating the target loss function of the discriminator and the target loss function of the generator according to the tensor, the ultrasonic echo signal, the true label, and the virtual wave transfer perturbation map until both the target loss function of the discriminator and the target loss function of the generator reach a preset convergence condition;

[0020] The test set is used to test and evaluate the generative deep neural network model that meets the convergence conditions until the preset conditions are met to obtain the target network model.

[0021] Optionally, the computational inference method for tissue physical properties, wherein the step of calculating the target loss function of the discriminator and the target loss function of the generator based on the tensor, the ultrasound echo signal, the true label, and the virtual wave transfer perturbation map, specifically includes:

[0022] Use the BCE Loss of the discriminator to calculate Real Loss, Fake Loss and the L1 Loss of the generator:

[0023] Where N is the number of samples, x i is the i-th value of x, y i represents the i-th value in y, x represents the output of the discriminator, and y represents a tensor with the same output dimension as x;

[0024] When using BCE Loss to calculate Real Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the true label. The y in BCE Loss represents a tensor with all values ​​1 that is consistent with the output dimension of the discriminator.

[0025] When using BCE Loss to calculate Fake Loss, the output of the discriminator is obtained by transferring the ultrasonic echo signal and the virtual wave perturbation map. The y in BCE Loss represents a tensor with all values ​​​​of 0 that is consistent with the output dimension of the discriminator.

[0026] The target loss function D Loss of the discriminator is calculated based on Real Loss and Fake Loss:

[0027] When using BCE Loss to calculate G Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the virtual wave transfer perturbation map. The y in BCE Loss represents a tensor with all values ​​1 that is consistent with the output dimension of the discriminator.

[0028] The target loss function G Loss of the generator is calculated as follows: G Loss = BCE Loss + L1 Loss·γ;

[0029] in, x' represents the virtual wave transfer perturbation map, y' represents the real label, and γ represents the weight of L1 Loss.

[0030] Optionally, the computational reasoning method for tissue physical properties, wherein the test set is used to test and evaluate the generative deep neural network model that meets the convergence condition until a preset condition is met to obtain the target network model, specifically includes:

[0031] According to the test set, the SSIM coefficient SSIM(x * ,y * ):

[0032] Testing and evaluating the virtual wave transfer perturbation map output from the generative deep neural network model that meets the convergence condition, and obtaining the target network model if the SSIM coefficient meets the preset condition;

[0033] Among them, x * is the true label, y * is the virtual wave propagation disturbance diagram, Represents the brightness feature mean of the true label, calculates the brightness feature mean of the true label

[0034] where x * i Represents the true label x * The i-th pixel value in ;

[0035] Represents the contrast of the true label, which is calculated by the grayscale standard deviation of the true label and the unbiased estimation of the standard deviation to calculate the contrast of the true label

[0036] in Represents the contrast difference between the virtual wave transmission perturbation map and the real label, and calculates the contrast difference between the virtual wave transmission perturbation map and the real label

[0037] Among them, y * i The virtual wave propagation disturbance graph y is represented by * The i-th pixel value in , represents the brightness characteristic mean of the virtual wave transfer disturbance map;

[0038] C1 and C2 are related constants representing structural characteristics: C1 = (K1L) 2 , C2=(K2L) 2 ;

[0039] Among them, K1, K2 and L are empirical values.

[0040] Optionally, the computational reasoning method for tissue physical properties, wherein the step of selecting a target area from the target tissue and obtaining tissue imaging information under the target area, specifically includes:

[0041] Positioning the target tissue to obtain the target area;

[0042] The target area is scanned by an ultrasonic probe to collect tissue imaging information of the target area.

[0043] Optionally, the computational reasoning method of tissue physical properties, wherein the tissue physical properties include: tissue physical properties, tissue elastic modulus, tissue viscosity and tissue viscoelasticity.

[0044] In addition, to achieve the above-mentioned purpose, the present invention further provides a computational reasoning system for tissue physical properties, wherein the computational reasoning system for tissue physical properties comprises:

[0045] A model training and testing module is used to obtain a data set, train and test the generative deep neural network model based on the data set, and obtain a target network model;

[0046] a target object acquisition module, configured to select a target tissue, choose a target area from the target tissue, scan the target area, obtain tissue imaging information under the target area, and extract an ultrasonic echo signal of the target area from the tissue imaging information;

[0047] a wave transfer disturbance map acquisition module, configured to input the ultrasonic echo signal into the target network model to simulate a wave transfer disturbance map of the mechanical wave in the target area;

[0048] The tissue physical property calculation module is used to obtain the trajectory information of the mechanical wave according to the wave transmission disturbance map, calculate the transmission speed of the mechanical wave according to the trajectory information, and calculate the tissue physical property of the target area according to the transmission speed.

[0049] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a computational reasoning program for the physical characteristics of tissues stored on the memory and runnable on the processor, and when the computational reasoning program for the physical characteristics of tissues is executed by the processor, the steps of the computational reasoning method for the physical characteristics of tissues as described above are implemented.

[0050] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computational reasoning program for the physical properties of tissues, and when the computational reasoning program for the physical properties of tissues is executed by a processor, the steps of the computational reasoning method for the physical properties of tissues as described above are implemented.

[0051] In the present invention, a data set is obtained, and a generative deep neural network model is trained and tested based on the data set to obtain a target network model; a target tissue is selected, a target area is selected from the target tissue, the target area is scanned, tissue imaging information under the target area is obtained, and an ultrasonic echo signal of the target area is extracted from the tissue imaging information; the ultrasonic echo signal is input into the target network model to simulate a wave transmission disturbance map of the mechanical wave in the target area; trajectory information of the mechanical wave is obtained based on the wave transmission disturbance map, and the tissue physical properties of the target area are obtained based on the trajectory information; the present invention simulates the propagation disturbance process of the mechanical wave, and thus, without actually generating such a mechanical wave disturbance, a wave transmission disturbance map can be obtained, from which the speed of mechanical wave propagation can be obtained, and further physical properties related to the propagation speed can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] FIG1 is a flow chart of a preferred embodiment of a method for calculating and reasoning about tissue physical properties according to the present invention;

[0053] FIG2 is a flowchart of training and testing a generative deep neural network model in the computational reasoning method for tissue physical properties of the present invention;

[0054] FIG3 is an architecture diagram of a generative deep neural network model in the computational reasoning method for tissue physical properties of the present invention;

[0055] FIG4 is a schematic diagram showing the principle of a preferred embodiment of a system for calculating and reasoning about tissue physical properties according to the present invention;

[0056] FIG5 is a schematic diagram of an operating environment of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0057] This application provides a method for calculating and reasoning tissue physical properties and related equipment. To clarify and clarify the purpose, technical solutions, and effects of this application, the application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are intended only to illustrate this application and are not intended to limit this application.

[0058] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0059] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features specified as "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but this must be based on the fact that ordinary technicians in this field can implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0060] The method for calculating and reasoning the physical properties of tissues according to a preferred embodiment of the present invention is shown in FIG1 . The method for calculating and reasoning the physical properties of tissues comprises the following steps:

[0061] Step S100: Acquire a data set, train and test a generative deep neural network model based on the data set, and obtain a target network model.

[0062] The acquiring of the data set specifically includes:

[0063] A dedicated transient elastic imaging device is used to measure a number of tissues used for training, and ultrasonic echo signals and real wave transmission disturbance maps corresponding to the tissues used for training are obtained.

[0064] The ultrasonic echo signal is used as a feature, the real wave transfer disturbance map is used as a real label, and a data set is generated according to the feature and the real label.

[0065] The data set is divided into two parts according to a preset ratio, one part is used as a training set to train the generative deep neural network model, and the other part is used as a test set to detect the accuracy of the prediction results of the generative deep neural network model.

[0066] Physical property measurement techniques can be applied to a wide range of tissues, including soft tissues that perform biomechanical functions and tissues with biomechanical properties, such as visceral tissue, muscle, and skin. Tissue physical property measurement has a wide range of applications. Generally speaking, it can be applied to basic research, such as analyzing changes in tissue physical properties with age to understand the nature of certain physiological phenomena.

[0067] It can be understood that when acquiring training data, the present application uses a dedicated elastic imaging instrument (such as an existing transient elastic imaging device such as FibroScan) to measure the ultrasonic echo signals corresponding to several tissues used for training (the tissues used for training can be one or more different tissues), as well as the real wave transmission disturbance diagram corresponding to the ultrasonic echo signal.

[0068] Optionally, the obtained ultrasonic echo signal is preprocessed, including envelope extraction, logarithmic stretching, and region selection. The envelope extraction algorithm is a signal analysis-based technique that decomposes the original signal into two components: an envelope signal and a modulation signal. The envelope signal is the amplitude-varying portion of the original signal, while the modulation signal is the frequency-varying portion of the original signal. By extracting the envelope signal, the main features of the signal can be obtained, thereby achieving the goals of signal processing and feature extraction.

[0069] A data set is generated based on the ultrasonic echo signals and the true labels, wherein several of the ultrasonic echo signals serve as features of the data set and several true labels serve as labels of the data set, wherein the true label refers to the correct output or category of an instance, which can also be called a target variable.

[0070] Furthermore, the data set is divided into two parts according to a preset ratio (8:2), one part is used as a training set for training the neural network model, and the other part is used as a test set for detecting the accuracy of the prediction results of the neural network model (the division ratio is selected according to actual conditions, and a training-test ratio of 8:2 is adopted in this embodiment).

[0071] As shown in FIG2 , further, the generative deep neural network model includes a generator and a discriminator. The generative deep neural network model is trained and tested according to the data set to obtain a target network model, specifically including:

[0072] Step S101: input the training set into the generator for computational reasoning, and generate a plurality of virtual wave transfer disturbance maps through the generator;

[0073] It can be understood that, as shown in Figure 3, taking the Generative Adversarial Network (GAN) of pix2pix as an example, the generative deep neural network model includes a generator and a discriminator. The input of the generative deep neural network model is the ultrasonic echo signal in the training set, and the output is a wave transfer disturbance map characterizing the physical properties of the tissue, and modules such as the Attention Mechanism are added.

[0074] Among them, the generator uses the basic U-net architecture as the feature extraction backbone network to extract the structural information of the image. According to the characteristics of the tissue image, the number of model input channels can be selected as single channel or multi-channel (for example, 3 channels are used for color images, 1 channel is used for grayscale images, and when a combination of multiple scan lines is input, the number of channels is selected according to the actual situation), so that the number of channels of the model is consistent with the number of channels of the input data, which is used to generate the wave transfer disturbance map.

[0075] The discriminator uses a series of fully connected layers to reduce the dimension and finally obtain tensor features, which are used to distinguish real data from virtual data generated by the generator (the specific network structure, such as the number of network layers, convolution kernel size, etc., varies slightly and can be set according to actual conditions, and is not limited here).

[0076] Specifically, several of the ultrasonic echo signals in the training set are input into the generator for calculation. The generator extracts the structural information of several of the ultrasonic echo signals and generates several virtual wave transfer perturbation maps. The task of the generator is to generate false data that is as realistic as possible to deceive the discriminator.

[0077] It should be noted that this embodiment uses the pix2pix generative network as an example. In addition to the pix2pix generative network, other neural network models (generative adversarial networks, diffusion models, physics-informed neural networks, or neural networks capable of image-to-image translation) can also be used to simulate wave propagation. Different methods may vary in accuracy and algorithm complexity, and their applicability to mobile devices may also vary.

[0078] Step S102: using the discriminator to distinguish a plurality of the virtual wave transfer perturbation maps and a plurality of the real labels, and outputting a tensor;

[0079] The discriminator's task is to distinguish between real data (from the training set) and fake data generated by the generator. It outputs a tensor, or predicted probability value. The positive class is the positive sample. Positive and negative samples are often associated with binary classification problems. Positive samples are considered positive for the target class corresponding to the true value. (Positive samples are the target class we are looking for in binary classification problems.) Negative samples are considered negative for all other target classes that do not correspond to the true value.

[0080] Step S103: calculating the target loss function of the discriminator and the target loss function of the generator according to the tensor, the ultrasonic echo signal, the real label, and the virtual wave transfer perturbation map, until both the target loss function of the discriminator and the target loss function of the generator reach a preset convergence condition;

[0081] Specifically, the BCE (Binary Cross-Entropy) loss of the discriminator is calculated based on the probability value. This loss function is used to measure the real loss and fake loss of real data samples in the discriminator. The target loss function of the discriminator is calculated based on the real loss and fake loss. The loss function of the discriminator is intended to measure its performance in distinguishing real from fake data.

[0082] It can be understood that training a generative adversarial network (GAN) is a game-playing process, pitting the discriminator and generator against each other. The discriminator's goal is to maximize its ability to distinguish between real and fake data, while the generator's goal is to maximize the realism of the generated fake data. There is an adversarial relationship between the discriminator's loss function and the generator's loss function. Optimizing the discriminator's loss leads to better classification of real and fake data, making it more difficult to train the generator. Conversely, optimizing the generator results in more realistic fake data, increasing the difficulty for the discriminator. The ultimate goal of a GAN is to reach an equilibrium state where the data generated by the generator is realistic enough that the discriminator cannot reliably distinguish between real and fake data. This is the convergence condition for a GAN, where the performance of the generator is maximized and the performance of the discriminator is minimized. This equilibrium state corresponds to the distribution of data generated by the generator matching the real data distribution.

[0083] Furthermore, the calculating the target loss function of the discriminator and the target loss function of the generator according to the tensor, the ultrasonic echo signal, the real label and the virtual wave transfer perturbation map specifically includes:

[0084] Use the BCE Loss of the discriminator to calculate Real Loss, Fake Loss and the L1 Loss of the generator:

[0085] Where N is the number of samples, x i is the i-th value of x, y i represents the i-th value in y, x represents the output of the discriminator, and y represents a tensor with the same output dimension as x;

[0086] When using BCE Loss to calculate Real Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the true label. The y in BCE Loss represents a tensor with all values ​​1 that is consistent with the output dimension of the discriminator.

[0087] When using BCE Loss to calculate Fake Loss, the output of the discriminator is obtained by transferring the ultrasonic echo signal and the virtual wave perturbation map. The y in BCE Loss represents a tensor with all values ​​​​of 0 that is consistent with the output dimension of the discriminator.

[0088] The target loss function D Loss of the discriminator is calculated based on Real Loss and Fake Loss:

[0089] It can be understood that BCE Loss is used to measure the difference between the output of real data samples in the discriminator and the label "1" (indicating real) (Real Loss), allowing the discriminator to correctly identify real data, and the discriminator's output should be as close to 1 as possible. At the same time, this loss function is used to measure the difference between the discriminator's output of data samples generated by the generator and the label "0" (indicating fake or generated) (Fake Loss). To enable the discriminator to correctly identify generated data as fake, for fake data, the discriminator's output should be as close to 0 as possible. The discriminator's final target loss function, D Loss, is half of the sum of the two.

[0090] When BCE Loss is used to calculate G Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the virtual wave transfer perturbation map, and y in BCE Loss represents a tensor whose values ​​are all 1, consistent with the output dimension of the discriminator.

[0091] The target loss function G Loss of the generator is calculated as follows: G Loss = BCE Loss + L1 Loss·γ;

[0092] in, x' represents the virtual wave transfer perturbation map, y' represents the real label, and γ represents the weight of L1 Loss (a hyperparameter, and a specific value can be selected according to actual conditions).

[0093] It is understandable that during the generator loss calculation process, the fake samples (y_fake) generated by the generator are input into the discriminator with the real samples (x) to obtain a discriminator's authenticity score (D_fake) for these fake samples. This score is then combined with a tensor of the same size whose elements are all 1 to calculate the BCE loss function, so that the fake samples generated by the generator can deceive the discriminator, making the discriminator think that these samples are real. The L1 loss is also calculated (selected according to the actual situation) to measure the difference between the generated fake samples (y_fake) and the real labels (y). The L1 loss calculates the absolute difference between the values ​​at each corresponding position between the two, encouraging the fake samples generated by the generator to be as close to the real samples as possible.

[0094] It should be noted that when the generator calculates L1 Loss (also known as MAE (mean abs error)), it can also use loss functions such as L2 Loss (also known as MSE (mean square error)) and Hubor Loss as alternatives. The main purpose is to minimize the difference between the generated results and the real results. BCE Loss in the generator and discriminator can be replaced with Focal Loss or Cross Entropy Loss, making the generator's results indistinguishable from the real ones and the discriminator better able to distinguish the authenticity of the results; the evaluation indicators of the generated results can be replaced with other indicators, such as Peak Signal to Noise Ratio (PSNR).

[0095] Step S104: Use the test set to test and evaluate the generative deep neural network model that meets the convergence condition until the preset condition is met to obtain the target network model.

[0096] Specifically, the test set is used to verify the difference between the output result of the generative deep neural network model that meets the convergence condition and the true label, and the performance value of the generative deep neural network model that meets the convergence condition is calculated based on the difference; the performance value is compared with a preset threshold, and if the performance value reaches the preset threshold, the target model that meets the requirements is obtained; if the performance value does not reach the preset threshold, the generative deep neural network model is repeatedly trained until the performance reaches the preset threshold, and the target model that meets the requirements is obtained.

[0097] Furthermore, the test set is used to test and evaluate the generative deep neural network model that meets the convergence condition until the preset condition is met to obtain the target network model, which specifically includes:

[0098] According to the test set, the SSIM coefficient SSIM(x * ,y * ):

[0099] Testing and evaluating the virtual wave transfer perturbation map output from the generative deep neural network model that meets the convergence condition, and obtaining the target network model if the SSIM coefficient meets the preset condition;

[0100] Among them, x * is the true label, y * is the virtual wave propagation disturbance diagram, Represents the brightness feature mean of the true label, and calculates the brightness feature mean of the true label

[0101] where x * i Represents the true label x * The i-th pixel value in ;

[0102] Represents the contrast of the true label, which is calculated by the grayscale standard deviation of the true label and the unbiased estimation of the standard deviation to calculate the contrast of the true label

[0103] in Represents the contrast difference between the virtual wave transmission perturbation map and the real label, and calculates the contrast difference between the virtual wave transmission perturbation map and the real label

[0104] Among them, y * i The virtual wave propagation disturbance graph y is represented by * The i-th pixel value in , represents the brightness characteristic mean of the virtual wave transfer disturbance map;

[0105] C1 and C2 are related constants representing structural characteristics: C1 = (K1L) 2 , C2=(K2L) 2 ;

[0106] Among them, K1, K2 and L are empirical values.

[0107] Step S200 , selecting a target tissue, selecting a target area from the target tissue, scanning the target area, obtaining tissue imaging information under the target area, and extracting an ultrasonic echo signal of the target area from the tissue imaging information.

[0108] The step of selecting a target area from the target tissue, scanning the target area, and obtaining tissue imaging information of the target area specifically includes:

[0109] Positioning the target tissue to obtain the target area;

[0110] The target area is scanned by an ultrasonic probe to collect tissue imaging information of the target area.

[0111] It can be understood that ordinary medical imaging equipment (such as traditional ultrasonic probe equipment, which does not require external instrument vibration to generate mechanical waves and does not require tissue disturbance) is used to scan the target tissue, locate the target tissue, obtain the target area, collect the tissue imaging information from the target area, and extract the ultrasonic echo signal of the target area from the tissue imaging information to prepare for subsequent calculation of tissue physical properties; the use of professional tissue physical property measurement equipment is avoided, and the measurement cost is greatly reduced.

[0112] It should be noted that during the signal acquisition process, in addition to directly using single or multiple scan line signals from the B-ultrasound image generated by the ultrasound device, other relevant ultrasonic echo information (such as ultrasonic radio frequency signals) can also be used. After certain preprocessing (such as envelope extraction and logarithmic compression), a neural network can be used to generate wave propagation disturbance maps representing tissue physical properties for tissue physical property assessment. Any signal that can directly or after processing and represents tissue structural information can serve as input data for the present invention.

[0113] Furthermore, the results of two or more scan lines may be used for combined input. One possible input is that the results of multiple scan lines are input as different channels.

[0114] Step S300: input the ultrasonic echo signal into the target network model to simulate a wave propagation disturbance diagram of the mechanical wave in the target area.

[0115] The ultrasonic echo signal is input into the target network model for inference calculation, which can simulate the wave transmission disturbance map of the mechanical wave in the target area. The present application combines a generative neural network to infer (simulate) the propagation process of mechanical waves in the tissue structure, and then obtains a wave transmission disturbance map related to the physical properties of the tissue (such as the liver), which can be used to analyze the physical information representation related to the tissue. In addition, the user's operation requirements are low, the subjective influence is small, and the tolerance for tissue imaging quality is relatively high.

[0116] As an example, the wave propagation disturbance map includes: shear wave propagation image (shear wave propagation image), shear wave propagation maps (shear wave propagation map), elastogram images (elastic map image) or strain image (strain image), etc. of Transient Elastography; it should be noted that the above content is only for example and is not used to limit this solution.

[0117] Step S400: Obtaining trajectory information of the mechanical wave according to the wave transmission disturbance map, calculating the transmission speed of the mechanical wave according to the trajectory information, and calculating the tissue physical properties of the target area according to the transmission speed.

[0118] It is understood that the tissue physical properties include but are not limited to: tissue physical properties, tissue elastic modulus, tissue viscosity and tissue viscoelasticity.

[0119] Specifically, a wave transfer disturbance diagram containing mechanical wave trajectory information is inferred from the target network model. Based on the wave transfer disturbance, the transfer time and transfer depth of the mechanical wave in the target area (tissue) can be obtained. The transfer speed of the ultrasonic echo signal can be calculated by dividing the transfer depth by the transfer time. From the propagation speed, many physical properties of the tissue can be derived, such as hardness, elasticity, viscosity, viscoelasticity, etc.

[0120] For example, the process of calculating the tissue hardness of the target area according to the transmission speed is:

[0121] The tissue hardness E of the target area is calculated according to the transmission speed: E=3ρV 2 ;

[0122] Where ρ represents tissue density and V represents the shear wave transmission velocity.

[0123] It should be noted that in addition to manual methods, image segmentation and other processing techniques can be used to obtain physical property values ​​from wave propagation disturbance maps. Furthermore, neural networks can be used to obtain information about wave propagation depth and time, thereby determining wave propagation velocity and tissue physical property values. Other physical properties (such as viscosity) can also be measured and calculated based on feature extraction from the inferred wave propagation disturbance map.

[0124] This invention uses tissue hardness measurement as an example. The velocity of wave propagation reflects tissue hardness: faster wave propagation in tissue indicates greater hardness, and vice versa. By applying image enhancement methods such as multi-scale image enhancement, contrast enhancement, and color balancing to the wave propagation disturbance map, the image is de-enhanced. The wave propagation trajectory information from the wave propagation map (which records the wave propagation trajectory, similar to the elastogram in transient elastography) is extracted. This information, namely the propagation time and depth of the wave in tissue propagation, is used to infer the wave propagation velocity.

[0125] Furthermore, the present invention can hide or skip the stages from wave propagation disturbance map generation to tissue physical property measurement and calculation, using neural networks to directly predict tissue physical properties. Alternatively, multiple neural networks can be combined to handle different stages separately. For example, a generative neural network could be used to generate ultrasound echo information into a wave propagation disturbance map, while another neural network could be used to calculate tissue physical property values ​​from the wave propagation disturbance map. Multiple neural networks can be combined in explicit or implicit ways.

[0126] As can be seen, the present invention, based on ultrasonic imaging and signal acquisition technology (medical imaging equipment, such as ultrasonic imaging equipment) and generative neural networks, proposes a method for inferring the transmission disturbance process of mechanical waves during tissue deformation (wave transmission disturbance map, similar to transient elastography) from tissue structure (ultrasonic echo information). This method can directly utilize ordinary imaging equipment to obtain ultrasonic echo information of the tissue, and then, through the mapping established by the neural network, directly obtain the wave transmission process within the tissue. There is no need for external instrument vibration to generate mechanical waves or induce tissue disturbances, thereby obtaining the physical properties of the tissue. It does not require special equipment to measure tissue physical properties (such as shear wave elastography equipment or transient elastography equipment), nor does it rely on dedicated ultrasonic elasticity measurement equipment. Instead, it can achieve the acquisition of tissue physical properties in ordinary ultrasound probes, such as ultrasound probes, greatly reducing the cost of detection. It can be widely used in various existing imaging equipment and can perform low-cost tissue physical property detection.

[0127] Furthermore, as shown in FIG4 , based on the above-mentioned method for calculating and reasoning about tissue physical properties, the present invention also provides a system for calculating and reasoning about tissue physical properties, wherein the system for calculating and reasoning about tissue physical properties includes:

[0128] A model training and testing module 51 is used to obtain a data set, train and test a generative deep neural network model based on the data set, and obtain a target network model;

[0129] A target object acquisition module 52 is configured to select a target tissue, choose a target area from the target tissue, scan the target area, obtain tissue imaging information of the target area, and extract an ultrasonic echo signal of the target area from the tissue imaging information;

[0130] a wave transfer disturbance map acquisition module 53 for inputting the ultrasonic echo signal into the target network model to simulate a wave transfer disturbance map of the mechanical wave in the target area;

[0131] The tissue physical property calculation module 54 is used to obtain the trajectory information of the mechanical wave according to the wave transmission disturbance map, calculate the transmission speed of the mechanical wave according to the trajectory information, and calculate the tissue physical property of the target area according to the transmission speed.

[0132] Furthermore, as shown in FIG5 , based on the above-described method and system for calculating and reasoning about tissue physical properties, the present invention also provides a terminal, which includes a processor 10, a memory 20, and a display 30. FIG5 shows only some of the components of the terminal, but it should be understood that implementation of all shown components is not required, and more or fewer components may be implemented instead.

[0133] In some embodiments, the memory 20 can be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 can also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Furthermore, the memory 20 can also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as program code of the installation terminal, etc. The memory 20 can also be used to temporarily store data that has been output or is about to be output. In one embodiment, a calculation reasoning program 40 of tissue physical properties is stored on the memory 20, and the calculation reasoning program 40 of tissue physical properties can be executed by the processor 10, thereby realizing the calculation reasoning method of tissue physical properties in this application.

[0134] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, configured to execute program codes or process data stored in the memory 20, such as executing computational reasoning methods for the physical properties of the tissue.

[0135] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0136] In one embodiment, when the processor 10 executes the computational reasoning program 40 of the tissue physical properties in the memory 20, the following steps are implemented:

[0137] Obtaining a data set, and training and testing a generative deep neural network model based on the data set to obtain a target network model;

[0138] selecting a target tissue, selecting a target area from the target tissue, scanning the target area, obtaining tissue imaging information under the target area, and extracting an ultrasonic echo signal of the target area from the tissue imaging information;

[0139] Inputting the ultrasonic echo signal into the target network model to simulate a wave propagation disturbance diagram of the mechanical wave in the target area;

[0140] The trajectory information of the mechanical wave is obtained according to the wave transmission disturbance map, the transmission speed of the mechanical wave is calculated according to the trajectory information, and the tissue physical properties of the target area are calculated according to the transmission speed.

[0141] The step of obtaining a data set specifically includes:

[0142] Using a dedicated transient elastic imaging device to measure a number of tissues used for training, ultrasonic echo signals and real wave transmission disturbance maps corresponding to the tissues used for training are obtained;

[0143] Using the ultrasonic echo signal as a feature and the real wave transfer disturbance map as a true label, generating a data set according to the feature and the true label;

[0144] The data set is divided into two parts according to a preset ratio, one part is used as a training set to train the generative deep neural network model, and the other part is used as a test set to detect the accuracy of the prediction results of the generative deep neural network model.

[0145] The generative deep neural network model includes a generator and a discriminator, and the generative deep neural network model is trained and tested according to the data set to obtain a target network model, specifically including:

[0146] Inputting the training set into the generator for computational reasoning, and generating a plurality of virtual wave transfer disturbance maps through the generator;

[0147] Using the discriminator, the plurality of virtual wave transfer perturbation maps and the plurality of real labels are distinguished, and a tensor is output;

[0148] Calculating the target loss function of the discriminator and the target loss function of the generator according to the tensor, the ultrasonic echo signal, the true label, and the virtual wave transfer perturbation map until both the target loss function of the discriminator and the target loss function of the generator reach a preset convergence condition;

[0149] The test set is used to test and evaluate the generative deep neural network model that meets the convergence conditions until the preset conditions are met to obtain the target network model.

[0150] The calculating of the target loss function of the discriminator and the target loss function of the generator according to the tensor, the ultrasonic echo signal, the real label and the virtual wave transfer perturbation map specifically includes:

[0151] Use the BCE Loss of the discriminator to calculate Real Loss, Fake Loss and the L1 Loss of the generator:

[0152] Where N is the number of samples, x i is the i-th value of x, y i represents the i-th value in y, x represents the output of the discriminator, and y represents a tensor with the same output dimension as x;

[0153] When using BCE Loss to calculate Real Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the true label. The y in BCE Loss represents a tensor with all values ​​1 that is consistent with the output dimension of the discriminator.

[0154] When using BCE Loss to calculate Fake Loss, the output of the discriminator is obtained by transferring the ultrasonic echo signal and the virtual wave perturbation map. The y in BCE Loss represents a tensor with all values ​​​​of 0 that is consistent with the output dimension of the discriminator.

[0155] The target loss function D Loss of the discriminator is calculated based on Real Loss and Fake Loss:

[0156] When using BCE Loss to calculate G Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the virtual wave transfer perturbation map. The y in BCE Loss represents a tensor with all values ​​1 that is consistent with the output dimension of the discriminator.

[0157] The target loss function G Loss of the generator is calculated as follows: G Loss = BCE Loss + L1 Loss·γ;

[0158] in, x' represents the virtual wave transfer perturbation map, y' represents the real label, and γ represents the weight of L1 Loss.

[0159] The step of using the test set to test and evaluate the generative deep neural network model that meets the convergence conditions until the preset conditions are met to obtain the target network model specifically includes:

[0160] According to the test set, the SSIM coefficient SSIM(x * ,y * ):

[0161] Testing and evaluating the virtual wave transfer perturbation map output from the generative deep neural network model that meets the convergence condition, and obtaining the target network model if the SSIM coefficient meets the preset condition;

[0162] Among them, x * is the true label, y * is the virtual wave propagation disturbance diagram, Represents the brightness feature mean of the true label, and calculates the brightness feature mean of the true label

[0163] where x * i Represents the true label x * The i-th pixel value in ;

[0164] Represents the contrast of the true label, which is calculated by the grayscale standard deviation of the true label and the unbiased estimation of the standard deviation to calculate the contrast of the true label

[0165] in Represents the contrast difference between the virtual wave transmission perturbation map and the real label, and calculates the contrast difference between the virtual wave transmission perturbation map and the real label

[0166] Among them, y * i The virtual wave propagation disturbance graph y is represented by * The i-th pixel value in , represents the brightness characteristic mean of the virtual wave transfer disturbance map;

[0167] C1 and C2 are related constants representing structural characteristics: C1 = (K1L) 2 , C2=(K2L) 2 ;

[0168] Among them, K1, K2 and L are empirical values.

[0169] The step of selecting a target area from the target tissue and obtaining tissue imaging information of the target area specifically includes:

[0170] Positioning the target tissue to obtain the target area;

[0171] The target area is scanned by an ultrasonic probe to collect tissue imaging information of the target area.

[0172] The tissue physical properties include: tissue physical properties, tissue elastic modulus, tissue viscosity and tissue viscoelasticity.

[0173] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computational reasoning program for tissue physical properties, and when the computational reasoning program for tissue physical properties is executed by a processor, the steps of the computational reasoning method for tissue physical properties as described above are implemented.

[0174] In summary, the present invention provides a computational reasoning method for tissue physical properties and related equipment, the method comprising: acquiring a data set, training and testing a generative deep neural network model based on the data set to obtain a target network model; selecting target tissue information, selecting a target area from the target tissue, scanning the target area, obtaining tissue imaging information under the target area, and extracting an ultrasonic echo signal of the target area from the tissue imaging information; inputting the ultrasonic echo signal into the target network model to simulate a wave transmission disturbance map of a mechanical wave in the target area; obtaining trajectory information of the mechanical wave based on the wave transmission disturbance map, and obtaining the tissue physical properties of the target area based on the trajectory information; the present invention simulates the propagation disturbance process of a mechanical wave, and thus can obtain a wave transmission disturbance map without actually generating such a mechanical wave disturbance, from which the speed of mechanical wave propagation can be obtained, and further physical properties related to the propagation speed can be obtained.

[0175] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal comprising the element.

[0176] Of course, those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0177] It should be understood that the application of the present invention is not limited to the above examples. For those skilled in the art, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for computational reasoning of tissue physical properties, characterized in that: The computational reasoning method of the tissue physical properties includes: Acquire a data set, and train and test a generative deep neural network model according to the data set to obtain a target network model; Selecting a target tissue, selecting a target area from the target tissue, scanning the target area, obtaining tissue imaging information under the target area, and extracting an ultrasonic echo signal of the target area from the tissue imaging information; Inputting the ultrasonic echo signal into the target network model to simulate a wave transmission disturbance diagram of the mechanical wave in the target area; The trajectory information of the mechanical wave is obtained according to the wave transfer disturbance map, the transmission speed of the mechanical wave is calculated according to the trajectory information, and the tissue physical characteristics of the target area are calculated according to the transmission speed.

2. The method for calculating and reasoning the physical properties of tissue according to claim 1, characterized in that: The obtaining of the data set specifically includes: Measuring a number of tissues for training by using a dedicated transient elastic imaging device to obtain ultrasonic echo signals and real wave transmission disturbance diagrams corresponding to the tissues for training; Taking the ultrasonic echo signal as a feature and the real wave transfer disturbance map as a real label, generating a data set according to the feature and the real label; The data set is divided into two parts according to a preset ratio, one part is used as a training set to train the generative deep neural network model, and the other part is used as a test set to detect the accuracy of the prediction results of the generative deep neural network model.

3. The method for calculating and reasoning the physical properties of tissue according to claim 2, characterized in that: The generative deep neural network model includes a generator and a discriminator, and the generative deep neural network model is trained and tested according to the data set to obtain a target network model, specifically including: Inputting the training set into the generator for computational reasoning, and generating a plurality of virtual wave transfer disturbance graphs through the generator; Using the discriminator, a plurality of the virtual wave transfer perturbation images and a plurality of the real labels are distinguished, and a tensor is output; The judgment is calculated according to the tensor, the ultrasonic echo signal, the real label and the virtual wave transfer disturbance map. The target loss function of the discriminator and the target loss function of the generator are converged until the target loss function of the discriminator and the target loss function of the generator reach a preset convergence condition; The test set is used to test and evaluate the generative deep neural network model that meets the convergence condition until the preset condition is met to obtain the target network model.

4. The method for calculating and reasoning the physical properties of tissue according to claim 3, characterized in that: The calculating the target loss function of the discriminator and the target loss function of the generator according to the tensor, the ultrasonic echo signal, the real label and the virtual wave transfer perturbation map specifically includes: Use the BCE Loss of the discriminator to calculate Real Loss, Fake Loss and L1 Loss of the generator: Where N is the number of samples, x i is the ith value of x, y i represents the i-th value in y, x represents the output of the discriminator, and y represents a tensor with the same output dimension as x; When using BCE Loss to calculate Real Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the real label, and y in BCE Loss represents a tensor whose values ​​are all 1 and consistent with the output dimension of the discriminator; When using BCE Loss to calculate Fake Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the virtual wave transmission perturbation map, and y in BCE Loss represents a tensor whose values ​​are all 0 and consistent with the output dimension of the discriminator; The target loss function D Loss of the discriminator is calculated according to Real Loss and Fake Loss: When using BCE Loss to calculate G Loss, the output of the discriminator is obtained by the ultrasonic echo signal and the virtual wave transfer perturbation map, and y in BCE Loss represents a tensor whose values ​​are all 1 and are consistent with the output dimension of the discriminator; The target loss function G Loss of the generator is calculated: G Loss=BCE Loss+L1 Loss·γ; in, x' represents the virtual wave transfer perturbation map, y' represents the real label, and γ represents the weight of L1 Loss.

5. The method for calculating and reasoning the physical properties of tissue according to claim 4, characterized in that: The test set is used to test and evaluate the generative deep neural network model that meets the convergence condition until the preset condition is met to obtain the target network model, specifically including: According to the test set, the SSIM coefficient SSIM(x * ,y * ): Testing and evaluating the virtual wave transfer disturbance map outputted from the generative deep neural network model that meets the convergence condition, and obtaining the target network model if the SSIM coefficient meets the preset condition; Among them, x * is the true label, y * is the virtual wave transfer disturbance diagram, Represents the brightness feature mean of the true label, and calculates the brightness feature mean of the true label where x * i Represents the true label x * The i-th pixel value in ; Represents the contrast of the true label, which is calculated by the grayscale standard deviation of the true label and the standard deviation is unbiasedly estimated to calculate the contrast of the true label in Represents the contrast difference between the virtual wave transmission perturbation map and the real label, and calculates the contrast difference between the virtual wave transmission perturbation map and the real label Among them, y * i The virtual wave propagation disturbance graph y is represented by * The i-th pixel value in , represents the brightness characteristic mean of the virtual wave transfer disturbance map; C1 and C2 are related constants representing structural characteristics: C1 = (K1L) 2 , C2=(K2L) 2 ; Among them, K1, K2 and L are empirical values.

6. The method for calculating and reasoning the physical properties of tissue according to claim 1, characterized in that: The step of selecting a target area from the target tissue, scanning the target area, and obtaining tissue imaging information of the target area specifically includes: Positioning the target tissue to obtain the target area; The target area is scanned by an ultrasonic probe to acquire tissue imaging information of the target area.

7. The method for calculating and reasoning the physical properties of tissue according to claim 1, characterized in that: The tissue physical properties include: tissue physical properties, tissue elastic modulus, tissue viscosity and tissue viscoelasticity.

8. A computational reasoning system for tissue physical properties, characterized in that: The computational inference system for tissue physical properties includes: A model training and testing module is used to obtain a data set, train and test the generative deep neural network model according to the data set, and obtain a target network model; A target object acquisition module, used to select a target tissue, select a target area from the target tissue, scan the target area, obtain tissue imaging information under the target area, and extract an ultrasonic echo signal of the target area from the tissue imaging information; A wave transfer disturbance map acquisition module, used for inputting the ultrasonic echo signal into the target network model to simulate a wave transfer disturbance map of the mechanical wave in the target area; The tissue physical property calculation module is used to obtain the trajectory information of the mechanical wave according to the wave transfer disturbance map, calculate the transmission speed of the mechanical wave according to the trajectory information, and calculate the tissue physical property of the target area according to the transmission speed.

9. A terminal, characterized in that: The terminal includes: a memory, a processor, and a computational reasoning program for physical properties of tissues stored in the memory and executable on the processor. When the computational reasoning program for physical properties of tissues is executed by the processor, the steps of the computational reasoning method for physical properties of tissues as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computational reasoning program for physical properties of tissues, and when the computational reasoning program for physical properties of tissues is executed by a processor, the steps of the computational reasoning method for physical properties of tissues as described in any one of claims 1 to 7 are implemented.

Images