CN-122004941-A - Ultrasonic tomography method based on virtual data reconstruction

CN122004941ACN 122004941 ACN122004941 ACN 122004941ACN-122004941-A

Abstract

The disclosure provides an ultrasonic tomography method based on virtual data reconstruction, and relates to the technical field of medical ultrasound. The method comprises the specific implementation mode that wave field interpolation is carried out based on frequency domain scattered wave observation data to realize virtual data reconstruction, frequency domain virtual scattered wave observation data are obtained, multi-scale inversion is carried out according to the frequency domain virtual scattered wave observation data to optimize a sound velocity model, and then ultrasonic tomography is generated. According to the technical scheme, under the condition that the accuracy of the conventional FWI inversion is ensured to be close, the calculation cost of wave field solution in the operation process can be effectively reduced by reducing the wave field calculation domain, the calculation complexity is reduced, and the efficiency of ultrasonic tomography is improved.

Inventors

LI YUBING
LU XIANGWEI
SU CHANG
LIN WEIJUN

Assignees

中国科学院声学研究所

Dates

Publication Date: 20260512
Application Date: 20260228

Claims (10)

1. An ultrasound tomography method based on virtual data reconstruction, comprising: The real ultrasonic annular transducer performs full matrix data acquisition at discrete sampling time points under a load working condition and an idle working condition respectively to acquire real total field observation data and real background field observation data, wherein the load working condition is that a target body to be detected is placed in the real ultrasonic annular transducer, the full matrix data acquisition is that each array element of the real ultrasonic annular transducer sequentially transmits sound wave signals, and the other array elements receive the sound wave signals; performing field-wave separation by using the real total field observation data and the real background field observation data to obtain real scattered wave observation data, and obtaining frequency domain real scattered wave observation data through discrete Fourier transform; Selecting N f processing frequency points according to the central frequency of the emitted sound wave signals in the full matrix data acquisition, wherein N f is a positive integer greater than 1, and the N f processing frequency points are distributed on two sides of the central frequency; Constructing a virtual ultrasonic ring transducer, and performing wave field interpolation by utilizing frequency domain real scattered wave observation data of each processing frequency point to realize virtual data reconstruction to obtain frequency domain virtual scattered wave observation data of each processing frequency point, wherein the geometric centers of the virtual ultrasonic ring transducer and the real ultrasonic ring transducer are consistent; Setting an initial sound velocity model, performing multi-scale inversion by using frequency domain virtual scattered wave observation data and/or frequency domain real scattered wave observation data corresponding to each processing frequency point, and iteratively optimizing the sound velocity model; and generating ultrasonic tomography by using the iteratively optimized sound velocity model.
2. The method of claim 1, comprising performing a virtual data inversion for each of the N f processing frequency bins using its corresponding frequency domain virtual scattered wave observation data.
3. The method according to claim 1, characterized in that it comprises: Performing virtual data inversion on the frequency domain virtual scattered wave observation data corresponding to the lowest processing frequency point to the next highest processing frequency point; And carrying out conventional frequency domain full waveform inversion on the highest processing frequency point by adopting corresponding frequency domain true scattered wave observation data.
4. A method according to claim 1,2 or 3, wherein performing wave field interpolation using the frequency domain real scattered wave observation data of each processing frequency point to implement virtual data reconstruction, obtaining frequency domain virtual scattered wave observation data of each processing frequency point, comprises: performing angular conversion on the frequency domain real scattered wave observation data to obtain a real scattered wave angular spectrum; Performing single-step shifting by utilizing the real scattered wave angular spectrum to obtain a first virtual scattered wave angular spectrum representing the real ultrasonic annular transducer to emit an acoustic wave signal, wherein the virtual ultrasonic annular transducer receives the acoustic wave signal; according to reciprocity of a cylindrical wave angular spectrum analysis propagation theory, obtaining a second virtual scattered wave angular spectrum representing the sound wave signal emitted by the virtual ultrasonic annular transducer, wherein the second virtual scattered wave angular spectrum of the sound wave signal is received by the real ultrasonic annular transducer; performing single-step shifting by utilizing the second virtual scattered wave angular spectrum to obtain a third virtual scattered wave angular spectrum representing the sound wave signals transmitted and received by the virtual ultrasonic annular transducer; and carrying out scanning transformation on the third virtual scattered wave angular spectrum to obtain the frequency domain virtual scattered wave observation data.
5. The method as recited in claim 4, further comprising: Establishing the single-step moving reverse relation according to reversibility of a column wave angle spectrum analysis propagation theory; Constructing the single-step moving loss function by using an L2 norm according to the inverse relation; and optimizing the data obtained by the single step shifting by using a first local gradient optimization algorithm according to the loss function.
6. The method of claim 5, wherein the first local gradient optimization algorithm uses a line search method to obtain a step size and a steepest descent method to obtain a descent gradient.
7. The method of claim 2,3 or 6, wherein the virtual data inversion comprises: Constructing a Helmholtz equation of the frequency domain simulation total field observation data and the frequency domain simulation background field observation data by utilizing a second-order constant density acoustic wave equation, solving based on a convergence Berne series method to obtain the frequency domain simulation scattered field observation data, and realizing forward modeling; Constructing an objective function of the virtual data inversion by using the L2 norms of the frequency domain virtual scattered wave observation data and the frequency domain simulated scattered field observation data; and optimizing the sound velocity model by using a second local gradient optimization algorithm according to the target function of the virtual data inversion.
8. The method of claim 7, wherein the solving based on the converged berne series method to obtain the frequency domain simulated fringe field observation data comprises: Performing field wave separation and parameter separation by utilizing a Helmholtz equation of the frequency domain simulation total field observation data and the frequency domain simulation background field observation data to obtain a differential form Lipman-Shi Wenge equation about the frequency domain simulation scattered field observation data; and solving the differential form Liepman-Shi Wenge equation by adopting a convergence Beren series method to obtain the frequency domain simulation scattered field observation data.
9. The method of claim 7, wherein the solving based on the converged berne series method to obtain the frequency domain simulated fringe field observation data, further comprises: Solving a Helmholtz equation of the frequency domain simulation total field observation data and the frequency domain simulation background field observation data by adopting a convergence Berne series method to obtain the frequency domain simulation total field observation data and the frequency domain simulation background field observation data; And performing field wave separation by using the frequency domain simulation total field observation data and the frequency domain simulation background field observation data to obtain the frequency domain simulation scattered field data.
10. The method of claim 7, wherein the second local gradient optimization algorithm obtains a current gradient using a concomitant state method, optimizes the current gradient using a conjugate gradient method or an L-BFGS method, and obtains a step size using a line search method.

Description

Ultrasonic tomography method based on virtual data reconstruction Technical Field The disclosure relates to the technical field of medical ultrasound, in particular to an ultrasound tomography method based on virtual data reconstruction. Background Ultrasonic tomography (Ultrasound Computed Tomography, USCT) is a non-invasive medical imaging technique, and provides a reliable way for clinically acquiring structural information and acoustic parameter distribution of patient tissues by virtue of the remarkable advantages of no ionizing radiation and moderate cost. According to the technology, the ultrasonic transducer array is arranged, sound wave data are collected at multiple angles, and then the internal structure and parameter distribution are reconstructed by using a related signal processing algorithm, so that the technology has good diagnosis potential for early tissue pathological changes. At present, ultrasonic tomography methods can be mainly divided into two main categories, namely an imaging algorithm based on a ray propagation assumption, including a Delay and sum (DAS) algorithm and a time-of-flight tomography (Time of flight tomography, raft). The two methods have the advantages of high reconstruction speed and easy realization by means of approximate processing of the acoustic wave propagation path, and are suitable for rapid imaging. The other type is a full waveform inversion method (Full waveform inversion, FWI) based on wave equation and wave propagation theory, and the method can obtain imaging results with higher resolution and quantitative characteristics by fitting propagation waveforms, has remarkable potential in distinguishing microstructure changes, and can provide more accurate acoustic information for early lesion recognition. However, due to the complex iterative solution of partial differential equations and the large amount of computational resources involved, full waveform inversion currently suffers from deficiencies in imaging speed, has not reached a widely popular level in practical clinical applications, and is still in a continuous optimization and verification stage. To improve imaging efficiency, full waveform inversion may be reconstructed in the Frequency domain, a method commonly known as Frequency domain full waveform inversion (Frequency-Domain Full Waveform Inversion, FD-FWI). The multi-scale inversion strategy is adopted, namely, the inversion flow is started by the low-frequency data firstly, and then higher frequency components are gradually introduced, so that the calculation burden is reduced while the solving stability is ensured, and the image reconstruction on a personal computer is possible. Although the multi-scale strategy has greatly compressed the requirement of the algorithm on the computational resource, when facing the annular transducer array formed by large-base array elements, the number of independent acoustic wave propagation models required to be solved numerically is huge, so that the method cannot realize the rapid imaging performance meeting the clinical real-time requirement under the current hardware condition. Disclosure of Invention In a first aspect, the present disclosure provides an ultrasound tomography method based on virtual data reconstruction, comprising: The real ultrasonic annular transducer is used for acquiring full matrix data at discrete sampling time points under a loading working condition and an idle working condition respectively to acquire real total field observation data and real background field observation data, wherein the loading working condition refers to the condition that a target body to be detected is placed in the real ultrasonic annular transducer, the full matrix data acquisition refers to the condition that each array element of the real ultrasonic annular transducer sequentially transmits sound wave signals, and the rest array elements receive the sound wave signals. And performing field wave separation by using the real total field observation data and the real background field observation data to obtain real scattered wave observation data, and obtaining frequency domain real scattered wave observation data through discrete Fourier transform. According to the center frequency of the emitted sound wave signals in the full matrix data acquisition, N f processing frequency points are selected, wherein N f is a positive integer greater than 1, and N f processing frequency points are distributed on two sides of the center frequency. The method comprises the steps of constructing a virtual ultrasonic annular transducer, performing wave field interpolation by using frequency domain real scattered wave observation data of each processing frequency point to realize virtual data reconstruction, and obtaining frequency domain virtual scattered wave observation data of each processing frequency point, wherein the geometric centers of the virtual ultrasonic annular transducer and the real ultrasonic annular t