Ultrasound 3D reconstruction preprocessing method based on dual-channel linear autoregressive model

Abstract

The present invention discloses an ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregression model, comprising the following steps: 1) collecting ultrasonic radio frequency signals based on the constructed ultrasonic signal acquisition device; 2) preprocessing the ultrasonic radio frequency signals to obtain high-fidelity two-dimensional images; 3) reconstructing the two-dimensional images obtained by the preprocessing based on the ray projection method. The ultrasonic three-dimensional reconstruction preprocessing method based on the dual-channel linear autoregression model proposed by the present invention, on the one hand, prevents the main signal from attenuating by a gain compensation method, and on the other hand, performs noise reduction processing on the signal by dual-channel linear autoregression. The high-fidelity two-dimensional images obtained by the final preprocessing method ensure the high quality of subsequent ultrasonic three-dimensional reconstruction imaging.

Description

Technical Field

[0001] The invention relates to the fields of ultrasonic detection and ultrasonic imaging, and in particular to an ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregressive model. Background Art

[0002] A scan means that the horizontal axis of the display is the propagation time or propagation distance of the ultrasonic wave in the material being tested, and the vertical axis is the amplitude of the ultrasonic reflection wave. B scan means that the image is displayed in two dimensions, and the screen displays a section parallel to the propagation direction of the sound speed and perpendicular to the measured surface of the workpiece.

[0003] Traditional ultrasonic three-dimensional reconstruction is a two-dimensional image obtained through B scanning. Although it is convenient and fast to obtain images, the images obtained are not clear and contain a lot of noise. Therefore, obtaining clear two-dimensional images is a prerequisite for obtaining high-quality three-dimensional reconstruction. The three-dimensional reconstruction method based on A scanning can be suitable for signal acquisition of small objects because the probe is small in size, which can make up for the shortcomings of B ultrasonic probe in this regard; the frequency of A ultrasonic probe is generally higher than that of other probes, so a higher-resolution two-dimensional image can be obtained. In addition, when performing three-dimensional reconstruction through ultrasonic A scanning, the acquisition of A scanning requires the use of a signal acquisition device to drive the probe or the scanned tissue to complete the acquisition of ultrasonic signals. Therefore, the pressure, scanning speed, and scanning direction of the probe on the scanned tissue can be strictly controlled, making it easier to generate high-quality three-dimensional reconstructed images.

[0004] In 2017, Han Zhihui et al. collected echo signals using a linear array transducer without elevation focusing, combined the collected signals, and constructed a three-dimensional image in frequency using Fourier transform in the article "3D ultrasound imaging in frequency domain with 1Darray transducer" (Ultrasonics, 2017, 76: 28-34). The direction proposed in the paper needs to improve the imaging depth in the axial direction and the influence of the edge waves of the shadow-free plane wave. The image quality needs to be improved through multiple transmissions, and the frame rate is also reduced.

[0005] In 2017, Du Juan et al. used a 3D electric scanning table to obtain ultrasonic echo signals on the tooth surface and adopted a gain compensation method for the radio frequency signal in the article "Three-Dimensional Reconstruction and Visualization of Human Enamel Ex Vivo Using High-Frequency Ultrasound" (Ournal of Medical and Biological Engineering, 2017, 37(1): 112-122). The three-dimensional reconstruction of human tooth enamel in vitro was performed by VTK, achieving a certain degree of accuracy. The electric scanning table alone cannot obtain all the image information on the tooth surface, so the image has certain defects. In 2017, Yan Yu et al. in the article "Ultrasonic Radio Frequency Signal Imaging Reconstruction Design Based on MATLAB" (China Medical Equipment, 2018, 33(08): 77-79+99) collected ultrasonic radio frequency signals and implemented ultrasonic radio frequency data imaging reconstruction through low-frequency filtering, dynamic orthogonal demodulation, bandpass filtering and downsampling, dynamic display range compression, scanning display imaging and other steps. Although the radio frequency signal has been processed many times, the imaging quality and imaging accuracy are low.

[0006] In 2020, Huang Qinghua and others from Northwestern Polytechnical University, in the patent "Ultrasound 3D Reconstruction Method Based on Self-Supervised 3D Fully Convolutional Neural Network" (application publication number: CN 111815764 A), obtained a 3D ultrasound image of human tissue by scanning with an ultrasound probe in order to improve the accuracy of 3D reconstruction. They trained 3D convolution kernels of different scales separately and connected the convolution kernels in series to obtain a cascaded 3D fully convolutional neural network, and used the cascaded network to reconstruct the 3D ultrasound image. The extraction of images from 3D ultrasound images by 3D fully convolutional neural networks has the disadvantages of large amount of calculation and long calculation time. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the prior art and provide an ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregressive model, thereby effectively improving the quality of A-scan imaging.

[0008] The object of the present invention is achieved by: a preprocessing method for ultrasonic three-dimensional reconstruction based on a dual-channel linear autoregressive model, comprising the following steps:

[0009] Step 1) collecting ultrasonic radio frequency signals based on the constructed ultrasonic signal collection device;

[0010] Step 2) processing the radio frequency signal based on a dual-channel linear regression model;

[0011] Step 3) Reconstruct the preprocessed two-dimensional image into three dimensions based on the ray projection method.

[0012] As a further limitation of the present invention, the ultrasonic signal acquisition device in step 1) includes a base, the base is connected to the Y-axis movable guide screw by bolts, the Y-axis movable guide screw is connected to the X-axis movable guide screw by bolts, and a loading platform is fixed above the X-axis movable guide screw; the base is connected to the vertical support plate by bolts, the vertical support plate is fixedly connected to the Z-axis movable guide screw by bolts, and the ultrasonic probe clamping device is fixed to the Z-axis movable guide screw by screws.

[0013] As a further limitation of the present invention, the step of collecting the ultrasonic radio frequency signal in step 1) specifically includes: after scanning the first point from the origin, the probe moves to the next scanning position along the X direction, repeats the scanning until the set maximum value in the X direction is reached, and then moves one scanning position in the Y direction, and moves along the -X axis direction after the scanning is completed until the scanning covers all the set areas.

[0014] As a further limitation of the present invention, the step 2) specifically comprises:

[0015] Step 2.1) Accurately remove noise based on a dual-channel linear regression model;

[0016] Step 2.2) Maintain the main signal information based on the gain compensation method to improve the imaging clarity of the two-dimensional image.

[0017] As a further limitation of the present invention, the step 2.1) specifically includes: performing wavelet denoising processing on the radio frequency signal collected in step 1) as the original signal; dividing the signal into low-frequency and high-frequency signals, blocking all high-frequency signals from passing through the low-frequency signal, and blocking all low-frequency signals from passing through the high-frequency signal, thereby achieving dual-channel high and low frequency denoising, and obtaining dual-channel high and low frequency denoised signals;

[0018] When removing noise, the high-pass filter removes low-frequency interference mixed in the RF signal, and demodulates the RF signal after removing the high-frequency signal. The standard demodulation is:

[0019] P(n)=r(n)cos(2πf c t n ) (1)

[0020] Q(n)=r(n)sin(2πf c t n ) (2)

[0021] Among them, r(n) is the RF signal after high-pass filtering, f c is the center frequency of the transmitting probe, t nIt is the time corresponding to the current sampling point in the direction of sound speed.

[0022] The adjustment method is used to extract two orthogonal signals P and Q from the high-frequency echo signal. The attenuation of the transmitted signal and the frequency deviation of the received signal will increase with the detection depth of the ultrasonic beam. After the standard demodulation, the correction term is added to obtain the dynamic orthogonal demodulation formula:

[0023]

[0024]

[0025] Among them, W r is the frequency bandwidth ratio, a is the object attenuation coefficient, and d is the transmission frequency depth at the current sampling point;

[0026] After adjustment, the signal contains high-frequency harmonic components. The high-frequency harmonic components of the signal are removed by low-pass filtering. Based on linear regression signal compensation, the signal reception model is first constructed as follows:

[0027]

[0028] Among them, f c is the center frequency of the transmitting probe, δ o is the variance of the spectrum, and x is the propagation distance.

[0029] As a further limitation of the present invention, the step 2.2) specifically includes: performing signal envelope detection on the signal after high and low frequency denoising of the dual channels in step 2.1), performing template matching on the signal after envelope detection with the original signal, comparing whether the tip signal and abrupt signal of the original signal are removed, and then digitally compressing the signal. If the main signal after compression is attenuated, performing gain compensation by simulating an ideal signal model, and then performing dynamic orthogonal adjustment; if the main signal is not attenuated, directly performing dynamic orthogonal adjustment, and finally performing scan conversion to generate a two-dimensional image;

[0030] Estimate the attenuation coefficient β:

[0031]

[0032] Among them, θ(x k ,y k ,z k ) is in (x k ,y k ,z k ) point after filtering, is the data after compensation at this point;

[0033] By taking the derivative of f in equation (6), we can get the maximum power frequency:

[0034] f max =f c -4βxδ0 2 (7)

[0035] Among them, the β attenuation coefficient, f c is the center frequency of the transmitting probe, δ o is the variance of the spectrum, x is the propagation distance;

[0036] An autoregressive model is established to output the linear sum of previous samples of the process. The autoregressive model is:

[0037]

[0038] Among them, a c is the coefficient, r(n) is the collected signal;

[0039] Establish a moving average model:

[0040]

[0041] Among them, b e is the coefficient, u(n) is white noise;

[0042] The autoregressive moving average model of the sampling signal r(n) is established as:

[0043]

[0044] The power spectral density function r(f) of the sampled signal r(n):

[0045]

[0046] in, The variance with mean zero, a c is the coefficient, b e is the coefficient;

[0047] Define the relative error of attenuation:

[0048]

[0049] Where η represents the theoretical value of attenuation, is an estimated value.

[0050] As a further limitation of the present invention, the signal envelope detection is to divide the original ultrasonic radio frequency signal into two signals and merge them to obtain an envelope signal:

[0051]

[0052] Among them, P(n) and Q(n) are decomposed signals of the radio frequency signal.

[0053] The present invention adopts the above technical scheme, and compared with the prior art, the beneficial effects are as follows: (1) To address the attenuation problem of the main signal, a method of comparing the simulated ideal signal with the set one and adding gain compensation is used to prevent the main signal from attenuating, thereby improving the imaging quality of the two-dimensional image; (2) To address the noise problem of the original signal, a dual-channel linear autoregression is proposed to perform noise reduction processing on the signal, and the processed signal is compared with the original signal to determine whether the noise signal has been eliminated. Finally, the high-fidelity two-dimensional image obtained by the preprocessing method ensures the high quality of the subsequent ultrasonic three-dimensional reconstruction imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the ultrasonic signal acquisition device of the present invention.

[0055] Figure 2 Schematic diagram of the scanning movement trajectory of the probe of the present invention.

[0056] Figure 3 Flowchart of the present invention.

[0057] Among them, 1 is a base, 2 is a right side moving guide screw of the Y axis, 3 is a vertical support plate, 4 is a moving guide screw of the Z axis, 5 is a ultrasonic probe clamping device, 6 is a stage, 7 is a moving guide screw of the X axis, and 8 is a left side moving guide screw of the Y axis. DETAILED DESCRIPTION

[0058] like Figure 3 The ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregressive model includes the following steps:

[0059] Step 1) collecting ultrasonic radio frequency signals based on the constructed ultrasonic signal collection device;

[0060] First build Figure 1 The ultrasonic signal acquisition device shown includes a base 1, which is connected to the Y-axis movable guide screw by bolts, and the Y-axis movable guide screw includes a Y-axis right movable guide screw 2 and a Y-axis left movable guide screw 8; the Y-axis movable guide screw is connected to the X-axis movable guide screw 7 by bolts, and a loading platform 6 is fixed above the X-axis movable guide screw 7; the base 1 is connected to the vertical support plate 3 by bolts, and the vertical support plate 3 is fixedly connected to the Z-axis movable guide screw 4 by bolts, and the ultrasonic probe clamping device 5 is fixed to the Z-axis movable guide screw 4 by screws.

[0061] Build the ultrasonic signal acquisition device and follow the Figure 2The acquisition sequence shown is used to acquire signals from the object. After scanning the first point from the origin, the probe moves to the next scanning position along the X direction, repeats the scanning until it reaches the set maximum value in the X direction, and then moves one scanning position in the Y direction. After the scanning is completed, it moves along the -X axis direction until the scanning covers all the set areas.

[0062] Step 2) processing the radio frequency signal based on a dual-channel linear regression model;

[0063] Step 2.1) accurately removes the noise based on the dual-channel linear regression model; uses the RF signal collected in step 1) as the original signal for wavelet denoising; divides the signal into low-frequency and high-frequency signals, blocks all high-frequency signals from passing through the low-frequency signal, and blocks all low-frequency signals from passing through the high-frequency signal, thereby achieving dual-channel high- and low-frequency denoising and obtaining dual-channel high- and low-frequency denoised signals;

[0064] When collecting ultrasonic radio frequency signals, the ultrasonic equipment will introduce low-frequency interference generated by the hardware. In order to improve the quality of reconstructed A-scan images, wavelet denoising removes noise, and the high-pass filter removes the low-frequency interference mixed in the radio frequency signal. After removing the high-frequency signal, the radio frequency signal is demodulated. The standard demodulation is:

[0065] P(n)=r(n)cos(2πf c t n ) (1)

[0066] Q(n)=r(n)sin(2πf c t n ) (2)

[0067] Among them, r(n) is the RF signal after high-pass filtering, f c is the center frequency of the transmitting probe, t n is the time corresponding to the current sampling point in the direction of sound speed;

[0068] In order to improve the results of the modulation, the modulation method is used to extract two orthogonal signals P and Q from the high-frequency echo signal. The attenuation of the transmitted signal and the frequency deviation of the received signal will increase with the detection depth of the ultrasonic beam. After the standard demodulation, the correction term is added to obtain the dynamic orthogonal demodulation formula:

[0069]

[0070]

[0071] Among them, W r is the frequency bandwidth ratio, a is the object attenuation coefficient, and d is the transmission frequency depth at the current sampling point;

[0072] After adjustment, the signal contains high-frequency harmonic components. The high-frequency harmonic components of the signal are removed by low-pass filtering. Based on linear regression signal compensation, the signal reception model is first constructed as follows:

[0073]

[0074] Among them, f c is the center frequency of the transmitting probe, δ o is the variance of the spectrum, x is the propagation distance;

[0075] Step 2.2) Maintain the information volume of the main signal based on the gain compensation method to improve the imaging clarity of the two-dimensional image; perform signal envelope detection on the signal after high and low frequency denoising of the dual channels in step 2.1), perform template matching on the signal after envelope detection, compare whether the tip signal and abrupt signal of the original signal are removed, and then perform digital compression on the signal. If the compressed main signal is attenuated, perform gain compensation by simulating an ideal signal model, and then perform dynamic orthogonal adjustment; if the main signal is not attenuated, perform dynamic orthogonal adjustment directly, and finally perform scan conversion to generate a two-dimensional image;

[0076] Estimate the attenuation coefficient β:

[0077]

[0078] Among them, θ(x k ,y k ,z k ) is in (x k ,y k ,z k ) point after filtering, is the data after compensation at this point;

[0079] By taking the derivative of f in equation (6), we can get the maximum power frequency:

[0080] f max =f c -4βxδ0 2 (7)

[0081] Among them, the β attenuation coefficient, f c is the center frequency of the transmitting probe, δ o is the variance of the spectrum, x is the propagation distance;

[0082] Establish an autoregressive model and output the linear sum of previous samples of the process. The autoregressive model is:

[0083]

[0084] Among them, a cis the coefficient, r(n) is the collected signal;

[0085] Establish a moving average model:

[0086]

[0087] Among them, b e is the coefficient, u(n) is white noise;

[0088] The autoregressive moving average model of the sampling signal r(n) is established as:

[0089]

[0090] The power spectral density function r(f) of the sampled signal r(n):

[0091]

[0092] in, The variance with mean zero, a c is the coefficient, b e is the coefficient;

[0093] Define the relative error of attenuation:

[0094]

[0095] Where η represents the theoretical value of attenuation, is an estimated value.

[0096] Signal envelope detection is to divide the original ultrasonic radio frequency signal into two signals and merge them to obtain the envelope signal:

[0097]

[0098] Among them, P(n) and Q(n) are decomposed signals of the radio frequency signal.

[0099] Step 3) reconstructing the preprocessed two-dimensional image into three dimensions based on the ray projection method;

[0100] First, the generated two-dimensional image is grayed out, and the grayed image can be used as a two-dimensional image for three-dimensional reconstruction. Then, the generated two-dimensional image is subjected to Gaussian filtering to remove the noise of the image. The denoised image is subjected to coordinate transformation and ray projection method to reconstruct the image in three dimensions.

[0101] The present invention aims at the attenuation problem of the main signal, and compares the set simulated ideal signal and adds a gain compensation method to prevent the main signal from attenuating, thereby improving the imaging quality of the two-dimensional image. In view of the noise problem of the original signal, a dual-channel linear autoregression is proposed to perform noise reduction processing on the signal, and the processed signal is compared with the original signal to determine whether the noise signal is eliminated. The ultrasonic three-dimensional reconstruction preprocessing method based on the dual-channel linear autoregression model proposed by the present invention, on the one hand, prevents the attenuation of the main signal by the gain compensation method, and on the other hand, performs noise reduction processing on the signal by the dual-channel linear autoregression. The high-fidelity two-dimensional image obtained by the final preprocessing method ensures the high quality of the subsequent ultrasonic three-dimensional reconstruction imaging.

[0102] The present invention is not limited to the above-mentioned embodiments. On the basis of the technical solution disclosed in the present invention, technicians in this field can make some substitutions and deformations to some technical features therein according to the disclosed technical content without creative labor, and these substitutions and deformations are all within the protection scope of the present invention.

Claims

1. A preprocessing method for ultrasonic three-dimensional reconstruction based on a dual-channel linear autoregressive model, characterized in that: The following steps are involved: Step 1) collecting ultrasonic radio frequency signals based on the constructed ultrasonic signal collection device; Step 2) processing the radio frequency signal based on a dual-channel linear autoregressive model; The step 2) specifically includes: Step 2.1) Accurately remove the noise based on a dual-channel linear autoregressive model; Step 2.2) Maintaining the information volume of the main signal based on the gain compensation method and improving the imaging clarity of the two-dimensional image; The step 2.1) specifically includes: using the radio frequency signal collected in step 1) as the original signal for wavelet denoising; dividing the signal into low-frequency and high-frequency signals, blocking all high-frequency signals from passing through the low-frequency signal, and blocking all low-frequency signals from passing through the high-frequency signal, to achieve dual-channel high and low frequency denoising, and obtain dual-channel high and low frequency denoised signals; When removing noise, a high-pass filter can be used to remove low-frequency interference mixed in the RF signal, and then the RF signal is demodulated, where the standard demodulation is: P(n)=r(n)cos(2πf c t n ) (1) Q(n)=r(n)sin(2πf c t n ) (2) Among them, P(n) and Q(n) are the decomposed signals of the RF signal; r(n) is the RF signal after high-pass filtering, and f c is the center frequency of the transmitting probe, t n is the time corresponding to the current sampling point in the direction of sound speed; The adjustment method is used to extract two orthogonal signals P and Q from the high-frequency echo signal. The attenuation of the transmitted signal and the frequency deviation of the received signal will increase with the detection depth of the ultrasonic beam. After adding the correction term after the standard demodulation, the dynamic orthogonal demodulation formula is obtained: Among them, W r is the frequency bandwidth ratio, a is the object attenuation coefficient, and d is the transmission frequency depth at the current sampling point; After adjustment, the signal contains high-frequency harmonic components. The high-frequency harmonic components of the signal are removed by low-pass filtering. Based on linear regression signal compensation, the signal reception model is first constructed as follows: Among them, f c is the center frequency of the transmitting probe, δ o is the variance of the spectrum, x is the propagation distance; f is the real-time power of the transmitting probe; The step 2.2) specifically includes: performing signal envelope detection on the signal after high and low frequency denoising of the dual channels in step 2.1), performing template matching on the signal after envelope detection with the original signal, comparing whether the tip signal and abrupt signal of the original signal are removed, and then digitally compressing the signal. If the main signal after compression is attenuated, performing gain compensation by simulating an ideal signal model, and then performing dynamic orthogonal adjustment; if the main signal is not attenuated, directly performing dynamic orthogonal adjustment, and finally performing scan conversion to generate a two-dimensional image; Estimate the attenuation coefficient β: Among them, θ(x k ,y k ,z k ) is in (x k ,y k ,z k ) point after filtering, is the data after compensation at this point; fz k It is the transmitting probe Z k Real-time power; x k ,y k , z k is the coordinate of point k in the X, Y, and Z directions; By taking the derivative of f in equation (6), we can get the maximum power frequency: f max =f c -4βxδ0 2 (7) Where β is the attenuation coefficient, f c is the center frequency of the transmitting probe, δ o is the variance of the spectrum, x is the propagation distance; An autoregressive model is established to output the linear sum of previous samples of the process. The autoregressive model is: Among them, a c is the coefficient, r(n) is the acquired signal; c is the sequence number of the previous sample sequence; p is the maximum sequence number of the previous sample sequence; k = c; Establish a moving average model: Among them, b e is the coefficient, u(n) is white noise; e is the sequence number of the white noise sequence; q is the maximum sequence number of the white noise sequence; l = e; The autoregressive moving average model of the sampling signal r(n) is established as: The power spectral density function r(f) of the sampled signal r(n): in, The variance with mean zero, a c is the coefficient, b e is the coefficient; Define the relative error of attenuation: Where η represents the theoretical value of attenuation, is an estimated value; Step 3) Reconstruct the preprocessed two-dimensional image into three dimensions based on the ray projection method.

2. The ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregressive model according to claim 1, characterized in that: The ultrasonic signal acquisition device in step 1) includes a base, the base is connected to the Y-axis movable guide screw by bolts, the Y-axis movable guide screw is connected to the X-axis movable guide screw by bolts, and a loading platform is fixed above the X-axis movable guide screw; the base is connected to the vertical support plate by bolts, the vertical support plate is fixedly connected to the Z-axis movable guide screw by bolts, and the ultrasonic probe clamping device is fixed to the Z-axis movable guide screw by screws.

3. The ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregressive model according to claim 1, characterized in that: The step of collecting the ultrasonic radio frequency signal in step 1) specifically includes: after scanning the first point from the origin, the probe moves to the next scanning position along the X direction, repeats the scanning until the set maximum value in the X direction is reached, and then moves one scanning position in the Y direction. After the scanning is completed, it moves along the -X axis direction until the scanning covers all the set areas.

4. The ultrasonic three-dimensional reconstruction preprocessing method based on a dual-channel linear autoregressive model according to claim 1, characterized in that: The signal envelope detection is to divide the original ultrasonic radio frequency signal into two signals and merge them to obtain an envelope signal: Among them, P(n) and Q(n) are decomposed signals of the radio frequency signal.

Images