A CMUT ultrasound image reconstruction method based on wavelet domain compressed sensing

Abstract

The present invention discloses a CMUT ultrasonic image reconstruction method based on wavelet domain compressed sensing, which relates to the field of CMUT ultrasonic image compression reconstruction. First, a CMUT ultrasonic imaging detection system is built to obtain CMUT ultrasonic image echo data of the model under test; then, a discrete wavelet basis is selected as a sparse basis for compressing the CMUT ultrasonic raw echo data, and a random Gaussian matrix is ​​selected as a sensing matrix for wavelet domain compressed sensing, and an orthogonal matching pursuit algorithm is used to reconstruct the processed wavelet domain image data; secondly, the CMUT ultrasonic raw echo data is compressed and reconstructed under different compression rates and the reconstruction error is calculated; finally, the wavelet domain compressed reconstruction image result is compared with the frequency domain compressed reconstruction image result to evaluate the reconstruction effect. The present invention applies compressed sensing to the field of CMUT ultrasonic image compression reconstruction, solves the problem of large storage data volume caused by full-channel acquisition in the CMUT ultrasonic imaging detection system, and can effectively reduce the complexity of the CMUT ultrasonic imaging detection system.

Description

Technical Field

[0001] The present invention relates to the field of CMUT ultrasonic image compression reconstruction, and in particular to a CMUT ultrasonic image reconstruction method based on wavelet domain compressed sensing. Background Art

[0002] At present, most ultrasonic imaging systems at home and abroad use piezoelectric ultrasonic transducers. However, due to the problems of high-density piezoelectric transducers, such as high process difficulty, high cost and high power consumption, the further development of ultrasonic imaging systems is restricted. Capacitive micromachined ultrasonic transducers (CMUT) can accurately obtain the signal differences caused by small density changes inside tissues due to their low power consumption, small size, high integration and large bandwidth, and are suitable for high-density integrated ultrasonic imaging detection systems. In recent years, the field of ultrasonic imaging has increasingly higher requirements for imaging clarity, especially three-dimensional ultrasonic imaging. This has also increased the requirements for the array element spacing, array element size and number of array elements of the ultrasonic array. The number of array elements can reach hundreds or even thousands, resulting in a sharp increase in the amount of data received by the CMUT ultrasonic imaging detection system, and higher requirements for the storage, processing and analysis of CMUT ultrasonic signals.

[0003] The compressed sensing theory proposed by Donoho, Candès and others has been widely studied and effectively applied in the field of ultrasonic signal and image processing. The core idea of ​​the compressed sensing theory is to accurately reconstruct the original signal with a sampling number far lower than the Nyquist sampling theorem when the original signal is sparse or can be sparsely represented. Compared with traditional compression methods, compressed sensing combines signal compression with sampling, saving transmission and storage costs and reducing hardware system complexity.

[0004] At present, domestic research in the field of CMUT ultrasound imaging mainly focuses on the research of theoretical basis and imaging algorithms. There is still little research on the application of compressed sensing theory in the field of CMUT ultrasound image compression reconstruction. There is a large room for development of related research. The research results are of great significance and application value for high-resolution imaging of CMUT devices. Summary of the invention

[0005] In order to solve the problem of large amount of stored data in a high-density CMUT array ultrasonic imaging detection system, the present invention provides a CMUT ultrasonic image reconstruction method based on wavelet domain compressed sensing.

[0006] The present invention is implemented by the following technical solution: a CMUT ultrasound image reconstruction method based on wavelet domain compressed sensing, comprising the following steps:

[0007] a) Building a CMUT ultrasonic imaging detection system to obtain CMUT ultrasonic image echo data of the model under test for subsequent reconstruction and analysis; the CMUT ultrasonic imaging detection system comprises: a host controller, a data acquisition system and a CMUT linear array probe; the CMUT linear array probe transmits the real-time ultrasonic image echo data of the model under test to the host controller, and the specific method is: using the CMUT linear array probe to detect the target area of ​​the model under test, and sampling at a sampling frequency of 60MHz; after the target area is detected, extracting the CMUT ultrasonic B-scan image echo data containing the target area as raw data for subsequent analysis and processing.

[0008] b) using discrete wavelet transform to sparse the discrete non-stationary CMUT ultrasound original image sequence signal, selecting a random Gaussian matrix as the sensing matrix of wavelet domain compressed sensing, and using the orthogonal matching pursuit algorithm to reconstruct the CMUT ultrasound image from the processed wavelet domain image data;

[0009] c) Combined with the method in step b), the CMUT ultrasound image echo data of the tested model is compressed and reconstructed using discrete wavelet transform at different compression rates, and the reconstruction effect is evaluated as follows: the obtained CMUT ultrasound original image echo data is sparsely processed using discrete wavelet transform; the compression rate is defined as the ratio of the compressed signal length to the original signal length; part of the wavelet domain data is compressed, and the compression rate range is set to 30%-80%, with each 10% as a level, and the processed wavelet domain data is reconstructed using compressed sensing for CMUT ultrasound image; structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE are selected as indicators for evaluating the accuracy of the image reconstruction algorithm, and the CMUT ultrasound image reconstruction effect at different compression rates is analyzed.

[0010] d) In combination with the method of step b), using the sensor matrix and reconstruction algorithm selected in step b), the obtained CMUT ultrasound original image echo data is subjected to discrete Fourier transform, part of the frequency domain data is compressed, the compression rate range is set to 30%-80%, and every 10% is a level, and the processed frequency domain data is reconstructed using compressed sensing; the structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE of the reconstructed image under frequency domain conditions are calculated, and compared with the wavelet domain reconstruction image results, and the CMUT ultrasound image reconstruction effect is analyzed;

[0011] e) Within the same compression rate range, the CMUT ultrasound image of the tested model after discrete wavelet transform and discrete Fourier transform is reconstructed. Finally, the image results of wavelet domain compression reconstruction are compared with those of frequency domain compression reconstruction to evaluate the CMUT ultrasound image reconstruction effect under different sparse basis conditions.

[0012] Compared with the prior art, the present invention has the following beneficial effects: a CMUT ultrasonic image reconstruction method based on wavelet domain compressed sensing provided by the present invention: ① applying the compressed sensing theory to the field of CMUT ultrasonic image compression reconstruction, using discrete wavelet transform to perform wavelet domain compression reconstruction on the original image data, and achieving good results; ② by building a CMUT ultrasonic imaging detection experimental platform, the tested model is reconstructed at different compression rates, and the structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE of the reconstructed images under different compressions are compared; the results show that, using the CMUT ultrasonic image reconstruction method based on wavelet domain compressed sensing proposed by the present invention, good image reconstruction results can be obtained with sampling point data far lower than the requirements of the Nyquist sampling theorem, which confirms the application potential of the method in the field of CMUT ultrasonic image compression reconstruction; ③ by comparing the wavelet domain compressed sensing image reconstruction results with the frequency domain compressed sensing image reconstruction results, the CMUT ultrasonic image reconstruction effect of the method is further proved. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of a CMUT ultrasound image reconstruction method based on wavelet domain compressed sensing.

[0014] Figure 2 This is a schematic diagram of the structure of the CMUT ultrasonic breast imaging detection system.

[0015] Figure 3 Structural similarity SSIM results of images reconstructed in wavelet domain at different compression rates.

[0016] Figure 4 The peak signal-to-noise ratio (PSNR) results of the image reconstructed in the wavelet domain at different compression rates.

[0017] Figure 5 It is the root mean square error (NRMSE) results of the image reconstructed in the wavelet domain under different compression rates.

[0018] Figure 6 This is the structural similarity SSIM comparison result of the reconstructed image in wavelet domain and frequency domain.

[0019] Figure 7 Comparison results of peak signal-to-noise ratio (PSNR) of reconstructed images in wavelet domain and frequency domain.

[0020] Figure 8 Comparison results of the root mean square error (NRMSE) of the reconstructed image under wavelet domain and frequency domain conditions. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in detail below with reference to the accompanying drawings and specific examples.

[0022] A CMUT ultrasound image reconstruction method based on wavelet domain compressed sensing, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0023] a) Building a CMUT ultrasonic imaging detection system to obtain CMUT ultrasonic image echo data of the model under test for subsequent reconstruction and analysis; the CMUT ultrasonic imaging detection system comprises: a host controller, a data acquisition system and a CMUT linear array probe; the CMUT linear array probe transmits the real-time ultrasonic image echo data of the model under test to the host controller, and the specific method is: using the CMUT linear array probe to detect the target area of ​​the model under test, and sampling at a sampling frequency of 60MHz; after the target area is detected, extracting the CMUT ultrasonic B-scan image echo data containing the target area as raw data for subsequent analysis and processing.

[0024] b) using discrete wavelet transform to sparse the discrete non-stationary CMUT ultrasound original image sequence signal, selecting a random Gaussian matrix as the sensing matrix of wavelet domain compressed sensing, and using the orthogonal matching pursuit algorithm to reconstruct the CMUT ultrasound image from the processed wavelet domain image data;

[0025] c) Combined with the method in step b), the CMUT ultrasound image echo data of the tested model is compressed and reconstructed using discrete wavelet transform at different compression rates, and the reconstruction effect is evaluated as follows: the obtained CMUT ultrasound original image echo data is sparsely processed using discrete wavelet transform; the compression rate is defined as the ratio of the compressed signal length to the original signal length; part of the wavelet domain data is compressed, and the compression rate range is set to 30%-80%, with each 10% as a level, and the processed wavelet domain data is reconstructed using compressed sensing for CMUT ultrasound image; structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE are selected as indicators for evaluating the accuracy of the image reconstruction algorithm, and the CMUT ultrasound image reconstruction effect at different compression rates is analyzed.

[0026] d) In combination with the method of step b), using the sensor matrix and reconstruction algorithm selected in step b), the obtained CMUT ultrasound original image echo data is subjected to discrete Fourier transform, part of the frequency domain data is compressed, the compression rate range is set to 30%-80%, and every 10% is a level, and the processed frequency domain data is reconstructed using compressed sensing; the structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE of the reconstructed image under frequency domain conditions are calculated, and compared with the wavelet domain reconstruction image results, and the CMUT ultrasound image reconstruction effect is analyzed;

[0027] e) Within the same compression rate range, the CMUT ultrasound image of the tested model after discrete wavelet transform and discrete Fourier transform is reconstructed. Finally, the image results of wavelet domain compression reconstruction are compared with those of frequency domain compression reconstruction to evaluate the CMUT ultrasound image reconstruction effect under different sparse basis conditions.

[0028] like Figure 2 As shown, this embodiment builds a CMUT ultrasonic breast imaging detection system to obtain CMUT ultrasonic image echo data of the tested breast model for subsequent analysis, and the specific steps are as follows;

[0029] a1) The CMUT ultrasound breast imaging detection system includes: a host controller, a data acquisition system and a CMUT linear array probe; wherein a plurality of simulated tumors are distributed inside the breast model to be tested.

[0030] a2) The CMUT linear array probe is used to detect the target tumor area of ​​the breast model under test, and sampling is performed using a 60 MHz sampling frequency; after the tumor area is detected, the CMUT ultrasound B-scan image echo data containing the tumor location is extracted as raw data for subsequent processing.

[0031] a3) This embodiment does not limit the CMUT ultrasonic probe and data acquisition hardware system, and any device that can complete the above functions can be used.

[0032] b) Selecting the sparse basis, sensor matrix and recovery and reconstruction algorithm in wavelet domain compressed sensing, and selecting the index for evaluating the accuracy of the reconstructed image, the specific operation is as follows;

[0033] b1) using discrete wavelet transform to perform sparse processing on the discrete non-stationary CMUT ultrasound original image sequence signal;

[0034] b2) Selecting a random Gaussian matrix as the sensing matrix for wavelet domain compressed sensing;

[0035] Construct an M×N matrix Φ so that each element in Φ independently obeys a Gaussian distribution with a mean of 0 and a variance of 1 / , that is,

[0036]

[0037] b3) Reconstruct the processed CMUT ultrasound image wavelet domain sparse data using the orthogonal matching pursuit algorithm. The specific algorithm flow is as follows:

[0038] Find a matrix D from the overcomplete dictionary that matches the initial residual e 0 The atom φ with the largest absolute value of inner product 1 , construct a sparse approximation by 0 Subtract its 1 The orthogonal projection on the spanned space gives the residual e1 , the residual is always orthogonal to the atoms that have been selected, and then the atom with the largest absolute value of the inner product with the residual is selected. The above steps are iterated on the residual until a specified stopping criterion is reached and the algorithm is stopped. Finally, the original signal can be represented by the linear sum of the last residual value and the atom;

[0039] b4) The compression ratio CR is defined as the ratio of the compressed signal length to the original signal length;

[0040]

[0041] Where, L is the measured signal length, and N is the original signal length;

[0042] b5) Select structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE as indicators for evaluating the accuracy of the reconstruction algorithm;

[0043] The structural similarity SSIM is defined as:

[0044]

[0045] where μ x is the mean value of x, μ y is the mean value of y, is the variance of x, is the variance of y, σ xy is the covariance of x and y, C 1 , C 2 It is a constant used to maintain stability.

[0046] The peak signal-to-noise ratio (PSNR) is defined as:

[0047]

[0048] Where: MAX I is the maximum value representing the color of an image point, and MSE is the mean square error, which is defined as:

[0049]

[0050] The normalized root mean square error NRMSE is defined as:

[0051]

[0052] c) The CMUT ultrasound image echo data of the tested breast model was compressed and reconstructed using discrete wavelet transform at different compression rates to evaluate the CMUT ultrasound image reconstruction effect at different compression rates:

[0053] c1) using discrete wavelet transform to perform sparse processing on the obtained CMUT ultrasound breast raw image data;

[0054] c2) using the sensor matrix and reconstruction algorithm selected in step b), compressing part of the wavelet domain data, setting the compression rate range to 30%-80%, with each 10% as a level, and reconstructing the CMUT ultrasound image of the processed wavelet domain data using compressed sensing;

[0055] c3) Calculate the structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE of the reconstructed image, and compare and analyze the calculation results;

[0056] c4) Analyze the CMUT ultrasound image reconstruction results:

[0057] Depend on Figure 3 It can be seen that when the compression rate is within 70%, the structural similarity of the reconstructed image remains at a high level. When the compression rate is 80%, the structural similarity decreases slightly but is still above 0.9. Figure 4 It can be seen that when the compression rate is within 60%, the peak signal-to-noise ratio of the reconstructed image remains at a high level. When the compression rate is greater than 60%, the peak signal-to-noise ratio value decreases slightly. Figure 5 It can be seen that the root mean square error increases with the increase of compression rate. When the compression rate is within 70%, the root mean square error is kept within 0.22. The experimental results show that when the CMUT ultrasound image is reconstructed based on wavelet domain compressed sensing, the error of the reconstructed image is small and the similarity is high when the compression rate is within 70%, and the original CMUT ultrasound image can be accurately reconstructed.

[0058] d) The CMUT ultrasound image echo data of the tested breast model was compressed and reconstructed using discrete Fourier transform at different compression rates to evaluate the CMUT ultrasound image reconstruction effect at different compression rates:

[0059] d1) using the sensor matrix and reconstruction algorithm selected in step b), performing discrete Fourier transform on the obtained CMUT ultrasonic raw echo data to obtain frequency domain echo data;

[0060] d2 compresses part of the frequency domain data, sets the compression rate range to 30%-80%, and uses compressed sensing to reconstruct the image of the frequency domain CMUT ultrasound echo data;

[0061] d3) Calculate the structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE of the frequency domain compressed reconstructed image, compare them with the wavelet domain reconstruction results and analyze the results.

[0062] d4) Analysis of CMUT ultrasound image reconstruction results:

[0063] e) Within the same compression rate range, the CMUT ultrasound image of the tested model after discrete wavelet transform and discrete Fourier transform is reconstructed. Finally, the image results of wavelet domain compression reconstruction are compared with those of frequency domain compression reconstruction to evaluate the CMUT ultrasound image reconstruction effect under different sparse basis conditions.

[0064] Figure 6 , Figure 7 , Figure 8 The structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE indicators of the reconstructed images under wavelet domain and frequency domain conditions are compared. The reconstruction accuracy under frequency domain compressed sensing conditions decreases with the increase of compression rate. Under the same compression rate, the accuracy of frequency domain reconstructed images is quite different from that of wavelet domain reconstruction. The image reconstructed by wavelet domain compressed sensing is significantly better than the image restored by frequency domain compressed sensing under various indicators. Therefore, compared with the frequency domain compressed sensing algorithm, the wavelet domain compressed sensing algorithm has a significant effect on CMUT breast ultrasound image reconstruction.

[0065] In summary, this embodiment uses the wavelet sparse basis through the above steps to apply the compressed sensing algorithm to the field of CMUT ultrasonic image compression reconstruction, performs sparse reconstruction on the CMUT ultrasonic image echo data in the wavelet domain, and achieves good results.

[0066] The scope of protection claimed by the present invention is not limited to the above specific embodiments. For those skilled in the art, the present invention may have various deformations and changes. Any modifications, improvements and equivalent substitutions made within the concept and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A CMUT ultrasound image reconstruction method based on wavelet domain compressed sensing, characterized in that: The steps include: a) Build a CMUT ultrasonic imaging detection system to obtain CMUT ultrasonic image echo data of the tested model for subsequent reconstruction and analysis; b) using discrete wavelet transform to sparse the discrete non-stationary CMUT ultrasound original image sequence signal, selecting a random Gaussian matrix as the sensing matrix of wavelet domain compressed sensing, and using the orthogonal matching pursuit algorithm to reconstruct the CMUT ultrasound image from the processed wavelet domain image data; c) combining the method of step b), compressing and reconstructing the CMUT ultrasound image echo data of the tested model using discrete wavelet transform at different compression rates, selecting structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE as indicators for evaluating the accuracy of the image reconstruction algorithm, and evaluating the CMUT ultrasound image reconstruction effect at different compression rates; The evaluation of reconstruction effect is as follows: the obtained CMUT ultrasound original image echo data is sparsely processed using discrete wavelet transform; the compression rate is defined as the ratio of the compressed signal length to the original signal length; some wavelet domain data is compressed, and the compression rate range is set to 30%-80%, with a 10% interval as a level, and the processed wavelet domain data is reconstructed into CMUT ultrasound images using compressed sensing; structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE are selected as indicators for evaluating the accuracy of image reconstruction algorithms, and the CMUT ultrasound image reconstruction effects under different compression rates are analyzed; d) performing discrete Fourier transform on the CMUT ultrasound original image sequence signal, and using the sensor matrix and reconstruction algorithm selected in step b), compressing and reconstructing the frequency domain image data of the model under test at different compression rates; selecting structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE as indicators for evaluating the accuracy of the image reconstruction algorithm, and evaluating the CMUT ultrasound image reconstruction effect at different compression rates; Specifically, the sensor matrix and reconstruction algorithm selected in step b) are used to perform discrete Fourier transform on the obtained CMUT ultrasound original image echo data, and part of the frequency domain data is compressed, and the compression rate range is set to 30%-80%, with each 10% as a level, and compressed sensing is used to reconstruct the processed frequency domain data; the structural similarity SSIM, peak signal-to-noise ratio PSNR and root mean square error NRMSE of the reconstructed image under frequency domain conditions are calculated, and compared with the wavelet domain reconstruction image results, and the CMUT ultrasound image reconstruction effect is analyzed; e) Within the same compression rate range, the CMUT ultrasound image of the tested model after discrete wavelet transform and discrete Fourier transform is reconstructed. Finally, the image results of wavelet domain compression reconstruction are compared with those of frequency domain compression reconstruction to evaluate the CMUT ultrasound image reconstruction effect under different sparse basis conditions.

2. The CMUT ultrasonic image reconstruction method based on wavelet domain compressed sensing according to claim 1, characterized in that: In step a), the CMUT ultrasonic imaging detection system includes: a host controller, a data acquisition system and a CMUT linear array probe; the CMUT linear array probe transmits the real-time ultrasonic image echo data of the model under test to the host controller, and the specific method is: using the CMUT linear array probe to detect the target area of ​​the model under test, and sampling at a sampling frequency of 60MHz; after the target area is detected, the CMUT ultrasonic B-scan image echo data containing the target area is extracted as raw data for subsequent analysis and processing.

Images