Search

CN-122020038-A - Ultrasonic shear wave viscoelasticity parameter inversion method, system, equipment and medium

CN122020038ACN 122020038 ACN122020038 ACN 122020038ACN-122020038-A

Abstract

The embodiment of the specification discloses an ultrasonic shear wave viscoelastic parameter inversion method, an ultrasonic shear wave viscoelastic parameter inversion system, ultrasonic shear wave viscoelastic parameter inversion equipment and an ultrasonic shear wave viscoelastic parameter inversion medium, wherein the parameter inversion method comprises the steps of obtaining original shear wave frequency dispersion data acquired by ultrasonic equipment, obtaining a viscoelastic parameter prediction value corresponding to the original shear wave frequency dispersion data based on a trained deep neural network, obtaining a sample set comprising a plurality of sample data, carrying out logarithmic conversion on training labels corresponding to each sample data in the sample set to obtain a logarithmic domain sample set comprising a plurality of logarithmic domain sample data, training the deep neural network to be trained based on the logarithmic domain sample set, and obtaining the trained deep neural network after convergence. The scheme realizes the ultrasonic viscoelasticity parameter inversion method with hardware universality, high precision, clinical safety and real-time processing capability.

Inventors

  • MA ZHICHUAN

Assignees

  • 聚融医疗科技(杭州)有限公司

Dates

Publication Date
20260512
Application Date
20260126

Claims (10)

  1. 1. The ultrasonic shear wave viscoelasticity parameter inversion method is characterized by comprising the following steps of: Acquiring original shear wave frequency dispersion data acquired by ultrasonic equipment; Obtaining a viscoelastic parameter predicted value corresponding to original shear wave frequency dispersion data based on the trained deep neural network, wherein the viscoelastic parameter predicted value is a logarithmic viscoelastic parameter set; the training method of the deep neural network comprises the following steps: Acquiring a sample set comprising a plurality of sample data, wherein the sample data comprises dispersion training data and training labels; Performing logarithmic conversion on training labels corresponding to each sample data in a sample set to obtain a logarithmic domain sample set comprising a plurality of logarithmic domain sample data, wherein the logarithmic domain sample data comprises dispersion training data and logarithmic domain training labels; Training the deep neural network to be trained based on the log domain sample set, and obtaining the trained deep neural network after convergence.
  2. 2. The ultrasonic shear wave viscoelastic parameter inversion method of claim 1, wherein the loss function of the deep neural network to be trained comprises a first error term and a second error term; The first error term characterizes the difference between the log domain training label of the input log domain sample data and the output viscoelastic parameter predicted value; The second error term characterizes the difference between the dispersion training data of the input logarithmic domain sample data and the theoretical dispersion curve obtained by the output viscoelastic parameter predicted value based on a Kelvin-Voigt dispersion equation.
  3. 3. The ultrasonic shear wave viscoelastic parameter inversion method of claim 1, further comprising: And carrying out frequency weighted noise processing and domain randomization processing on the frequency dispersion training data in the sample data to obtain the frequency dispersion training data in the logarithmic domain sample data.
  4. 4. The method of claim 1, wherein the raw shear wave frequency dispersion data comprises a plurality of discrete measurement frequency points and corresponding phase velocities thereof, and the acquiring the raw shear wave frequency dispersion data acquired by the ultrasonic device further comprises: and carrying out linear interpolation filling on the original shear wave frequency dispersion data in a preset frequency range, and taking the filled data as new original shear wave frequency dispersion data.
  5. 5. The ultrasonic shear wave viscoelastic parameter inversion method of claim 4, further comprising: and performing interpolation filling on an area exceeding an original data frequency range in a preset frequency range based on nearest neighbor measurement frequency points, wherein the original data frequency range is a value range of each measurement frequency point in original shear wave frequency dispersion data.
  6. 6. The ultrasonic shear wave viscoelastic parameter inversion method of claim 4, further comprising: based on the phase velocity in the dispersion training data corresponding to each sample data in the sample set, acquiring a sample mean value and a sample variance of the phase velocity in the sample set; The linear interpolation filling is carried out on the original shear wave frequency dispersion data in a preset frequency range, and the filled data is used as new original shear wave frequency dispersion data, and the method comprises the following steps: Performing linear interpolation filling on the original shear wave frequency dispersion data in a preset frequency range to obtain filled data; and carrying out Z-Score standardization processing on each phase velocity in the filled data based on the sample mean value and the sample variance, and taking the standardized data as new original shear wave frequency dispersion data.
  7. 7. The ultrasonic shear wave viscoelastic parameter inversion method according to claim 1, wherein the viscoelastic parameter set comprises elastic modulus and viscosity, the obtaining the viscoelastic parameter predicted value corresponding to the original shear wave frequency dispersion data based on the trained deep neural network comprises: The trained deep neural network adopts a Monte Carlo random inactivation method to obtain a plurality of groups of logarithmic-form viscoelastic parameter groups corresponding to the original shear wave frequency dispersion data; after each viscoelastic parameter set is subjected to index reduction, calculating parameter mean values and parameter standard deviations corresponding to the elastic modulus and the viscosity respectively, taking the parameter mean values corresponding to the elastic modulus and the viscosity respectively as inversion output of original shear wave frequency dispersion data, and taking the parameter standard deviations corresponding to the elastic modulus and the viscosity respectively as confidence of inversion output.
  8. 8. The ultrasonic shear wave viscoelastic parameter inversion system is characterized by comprising a data acquisition unit, a model prediction unit and a model training unit; The data acquisition unit acquires original shear wave frequency dispersion data acquired by ultrasonic equipment; The model prediction unit obtains a viscoelastic parameter prediction value corresponding to original shear wave frequency dispersion data based on a trained deep neural network, wherein the viscoelastic parameter prediction value is a logarithmic viscoelastic parameter set; The model training unit acquires a sample set comprising a plurality of sample data, wherein the sample data comprises frequency dispersion training data and training labels, performs logarithmic conversion on the training labels corresponding to the sample data in the sample set to obtain a logarithmic domain sample set comprising a plurality of logarithmic domain sample data, the logarithmic domain sample data comprises the frequency dispersion training data and the logarithmic domain training labels, and trains a deep neural network to be trained based on the logarithmic domain sample set, and converges to obtain the trained deep neural network.
  9. 9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1-7 when the computer program is executed.
  10. 10. A computer readable storage medium having stored thereon a computer program having instructions stored therein, which when run on a computer or processor, cause the computer or processor to perform the steps of the method according to any of claims 1-7.

Description

Ultrasonic shear wave viscoelasticity parameter inversion method, system, equipment and medium Technical Field Embodiments of the present disclosure relate to the field of ultrasonic shear wave elastography, and in particular, to optimization of real-time and robustness of ultrasonic shear wave viscoelastic parameter inversion. Background Ultrasonic Shear Wave Elastography (SWE) has been widely used for hardness (elasticity) measurement of soft tissues such as liver, breast, etc. However, biological tissues have viscoelasticity, and the viscosity information is ignored by simple elasticity measurement, so that the viscosity has important pathological significance in fatty liver grading and tumor benign and malignant identification. The existing viscoelastic parameter inversion method mainly comprises a table look-up method (LUT) and a least squares fit method (Curve fit). The table look-up method calculates a plurality of possible corresponding relations between viscoelastic parameter combinations (namely elastic modulus mu and viscosity eta) and theoretical dispersion curves (phase velocity-frequency relations) through a physical model in advance, and constructs a huge table look-up to be stored in a memory. During inversion, the dispersion curve obtained by actual measurement is compared with all theoretical curves in the lookup table, and the best matched theoretical curve is found through certain similarity measurement, and the corresponding parameter combination is the inversion result. The least squares fit method is to construct the parametric inversion problem as a nonlinear optimization problem, and generally assume that the tissue mechanics behavior follows a certain viscoelastic model, and define a theoretical dispersion curve with this model. During inversion, the measured dispersion curve is taken as a target, and the viscoelastic parameters (mu, eta) are iteratively adjusted through an optimization algorithm, so that the difference between a theoretical curve and a measured curve calculated by the parameters through a physical model is minimized, and the parameter value at the moment is an inversion result. However, in order to ensure inversion accuracy, the lookup table needs to cover a sufficiently dense parameter space, so that the memory occupation is large, and the searching time is long when the full-image pixel-by-pixel inversion is performed, so that the real-time imaging requirement is difficult to meet. The fitting method has poor robustness under a high noise environment, particularly for deep tissues of a human body, the nonlinear fitting process is easy to fall into a local optimal solution, so that the inversion accuracy is reduced, and even a negative viscosity value which does not conform to a physical rule can be calculated. Disclosure of Invention The embodiment of the specification provides an ultrasonic shear wave viscoelastic parameter inversion method, an ultrasonic shear wave viscoelastic parameter inversion system, ultrasonic shear wave viscoelastic parameter inversion equipment and an ultrasonic shear wave viscoelastic parameter inversion medium, and solves the problems of low inversion efficiency and poor robustness of a fitting method in the existing viscoelastic parameter inversion method. The depth neural network is adopted to invert the viscoelastic parameters to meet the real-time imaging requirement, and the logarithmic domain physical constraint training is carried out on the depth neural network, so that the robustness of inversion prediction results is ensured. The technical scheme is as follows: in a first aspect, embodiments of the present disclosure provide an ultrasonic shear wave viscoelastic parameter inversion method, including the steps of: Acquiring original shear wave frequency dispersion data acquired by ultrasonic equipment; Obtaining a viscoelastic parameter predicted value corresponding to original shear wave frequency dispersion data based on the trained deep neural network, wherein the viscoelastic parameter predicted value is a logarithmic viscoelastic parameter set; the training method of the deep neural network comprises the following steps: Acquiring a sample set comprising a plurality of sample data, wherein the sample data comprises dispersion training data and training labels; Performing logarithmic conversion on training labels corresponding to each sample data in a sample set to obtain a logarithmic domain sample set comprising a plurality of logarithmic domain sample data, wherein the logarithmic domain sample data comprises dispersion training data and logarithmic domain training labels; Training the deep neural network to be trained based on the log domain sample set, and obtaining the trained deep neural network after convergence. As a preferred scheme, the loss function of the deep neural network to be trained comprises a first error term and a second error term; The first error term characterizes the difference between the log d