High-b-value DWI image generation method

Abstract

The invention discloses a high-b-value DWI image generation method, which is derived from the urgent demand of high-b-value DWI in clinic and the defects of the prior art in the aspects of multi-modal fusion, noise suppression and model adaptability, and aims to break through the signal-to-noise ratio limitation and model simplification error of a traditional method by combining a double-index diffusion model and a complex signal averaging technology. And an efficient and reliable solution is provided for clinical popularization of the high-b-value DWI.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a method for generating a high b-value DWI image. Background Art

[0002] High b-value DWI is of great value in the diagnosis of prostate cancer, liver cancer and other diseases because it can significantly suppress background tissue signals and enhance the contrast between tumors and normal tissues. High b-value DWI can significantly suppress the signal intensity of normal prostate tissues such as the peripheral zone, reduce background tissue signals, reduce interference with lesions, and be more conducive to the detection of prostate cancer foci. Especially for inexperienced radiologists, the diagnostic accuracy is higher. In the peripheral zone, use b = 2000s / mm 2 When b = 1000s / mm, the accuracy of young doctors in diagnosing prostate cancer is significantly higher than that when b = 1000s / mm 2 Increased prostate cancer cell density and the disappearance of glandular duct structures limit the diffusion of water molecules, resulting in a high signal on high-b-value DWI. This contrasts sharply with the low-signal signal of normal tissue, facilitating the location and detection of cancer lesions, and is particularly effective for lesions that are small, hidden, or have similar signals to surrounding tissue. High-b-value DWI can effectively suppress the signal intensity of normal liver parenchyma, reducing the background signal of the liver, thereby more clearly displaying intrahepatic lesions and improving the detection rate of micro-liver cancers. This is especially true for small or early-stage liver cancers, as the contrast with surrounding normal liver tissue is more pronounced, facilitating early detection and diagnosis.

[0003] However, directly acquiring high-b-value DWI faces the following problems: First, directly acquiring high-b-value DWI requires a strong gradient field with rapid switching, which increases the cost of the scanner and is difficult to implement with low-field equipment, which places high demands on hardware performance. Second, directly acquiring high-b-value DWI results in severe signal attenuation at high b-values, a low signal-to-noise ratio, significant geometric distortion, and motion artifacts, especially in areas with inhomogeneous magnetic fields, such as the peripheral zone of the prostate, resulting in reduced image quality. Third, multiple excitations or repeated scans to improve the signal-to-noise ratio will prolong the examination time and affect patient tolerance.

[0004] Existing techniques use a single-exponential model to generate high-b-value images, failing to consider the complexity of tissue microstructure. Deep learning-based methods require extensive paired data training and lack physical model constraints, making them prone to overfitting or generating nonphysiological contrast. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides a method for generating a high b-value DWI image.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides a method for generating a high b-value DWI image, comprising the following steps:

[0008] Step S1, collecting at least 4 b-value DWI images, collecting each b-value DWI image for a certain length, and retaining the complex signal;

[0009] Step S2: aligning the phases of multiple scan data with the same b value using a cross-correlation algorithm;

[0010] Step S3, averaging the registered complex signals and taking the modulus of the averaged complex signals to obtain a low-noise DWI image;

[0011] Step S4: Generate a high b-value DWI image based on the bi-exponential model through a piecewise fitting method.

[0012] Preferably, in step S1, acquiring at least four b-value DWI images is achieved by using an MRI device greater than or equal to 3.0 T equipped with a phased array coil.

[0013] Preferably, in step S2, the phase alignment through the cross-correlation algorithm includes: selecting the complex image of the first scan and the complex image of the k-th scan, determining the maximum displacement by maximizing the cross-correlation function, and applying displacement compensation to the complex image of the k-th scan based on the maximum displacement to achieve phase alignment of the reference image and the image to be compensated.

[0014] Preferably, in step S3, the formula for averaging the registered complex signals is:

[0015] Where, N is the number of repetitions, S real,k (b) and S imag,k (b) are the real and imaginary signals of the kth scan.

[0016] Preferably, in step S4, the formula of the double exponential model is:

[0017] Where f is the perfusion fraction, D f Perfusion-related diffusion coefficient, D S True water diffusion coefficient.

[0018] Preferably, the segmented fitting method for generating high b-value DWI images includes: when the b-value is greater than or equal to 200 s / mm 2 When D S and S0(1-f); when b is less than 200s / mm 2 When D is obtained, S , solve f and D by nonlinear least squares method f .

[0019] Preferably, the segmented fitting method for generating high b-value DWI images further comprises using a Levenberg-Marquardt algorithm

[0020] Minimize the residual:

[0021]

[0022] Preferably, generating a high b-value DWI image comprises: based on the fitting parameters (f, D f ,D S ) Calculate the target high b value (such as b = 2000s / mm 2 ) signal strength:

[0023]

[0024] Preferably, the generating of high b-value DWI images includes: retaining phase information (directly multiplexing the phase of low b-value data)

[0025] or interpolation) to generate a complex signal:

[0026] in is the phase diagram with b value zero.

[0027] Preferably, generating a high b-value DWI image includes performing complex averaging on different generated results to further reduce noise.

[0028] The high b-value DWI image generation method proposed in this patent is based on the urgent demand for high b-value DWI in clinical practice and the shortcomings of existing technologies in multimodal fusion, noise suppression and model adaptability. By combining the double exponential diffusion model with complex signal averaging technology, it aims to break through the signal-to-noise ratio limitations and model simplification errors of traditional methods, and provide an efficient and reliable solution for the clinical popularization of high b-value DWI. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A high b-value DWI image generation method provided by the present invention;

[0030] Figure 2 Another high b-value DWI image generation method provided by the present invention;

[0031] Figure 3 Another high b-value DWI image generation method provided by the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] Example 1

[0034] like Figure 1 As shown, this embodiment provides a method for generating a high b-value DWI image, which includes:

[0035] Step 101: DWI data acquisition and preprocessing:

[0036] In this step, at least 4 b-value data are collected, and the b-values ​​are 0, 50, 200, and 800 s / mm 2 The data collected covers the perfusion-sensitive area and the diffusion-dominated area. Specifically, a single-shot EPI sequence is used to collect b = 0 s / mm 2 The reference image of the reference image, when b = 50 / 200 = 0s / mm 2 In this case, the perfusion sensitivity signal is enhanced by a short diffusion gradient time Δ = 20-30ms and a high gradient field strength of 40mT / m to capture microcirculatory blood flow information. When b = 800s / mm 2 When the diffusion gradient time was extended to Δ = 50 ms, the gradient field strength was increased to 60 mT / m to enhance the dominant signal of the water molecule diffusion weight. To acquire data at least four b-values, a segmented k-space filling technique was used to complete a four-b-value coverage scan within a single breath-hold period, with a temporal resolution of 4 seconds per slice.

[0037] Preprocessing includes denoising and registration. Specifically, the non-local mean or deep learning-based denoising algorithm DnCNN is used to improve the signal-to-noise ratio, ensure strict alignment of images with different b values, and avoid motion artifacts.

[0038] Step 102: Decompose the DWI signal attenuation into two parts: fast diffusion and slow diffusion based on the bi-exponential model. The formula is: Where S0 is the signal intensity when b=0, f is the volume fraction of the fast-diffusing component, and D f is the rapid diffusion coefficient, which reflects blood perfusion and is expressed in mm 2 / s,D s The slow diffusion coefficient reflects the diffusion of water molecules, similar to ADC. Usually, D f Much larger than D s The dual-exponential module may be an IVIM model, wherein the rapid diffusion is perfusion-related diffusion and the slow diffusion is water molecule diffusion.

[0039] Step 103, Parameter fitting method:

[0040] Piecewise fitting strategy:

[0041] Step 1031, Separate perfusion and diffusion components

[0042] Estimate perfusion parameters f and D using a low b-value (e.g., b < 200 s / mm 2 ), assuming that the perfusion signal has completely decayed at high b-values: when b > 200 s / mm f , 2 , At this time, a single exponential model can be used to fit D s and S0(1 - f);

[0043] Step 1032, Joint optimization

[0044] Use a non-linear least squares algorithm such as Levenberg-Marquardt to globally optimize all parameters f, D f , D s to ensure model consistency.

[0045] Regularization constraint: Limit the parameter range, D f > 0.01 mm 2 / s, 0 < f < 0.3, to avoid overfitting.

[0046] Step 104, Generate high b-value DWI images:

[0047] Based on the fitted parameters, calculate the signal intensity at the target high b-value (e.g., b = 2000 s / mm 2 ), and the calculation formula is as follows

[0048] The high b-value DWI image generation method provided in this Example 1 generates high b-value DWI images through a double exponential model, which not only retains microcirculation information but also significantly improves the lesion contrast under high diffusion weights, providing an efficient and economical solution for precision medicine. Compared with the single exponential model, the double exponential model can more realistically reflect the signal decay at high b-values, especially in regions with significant perfusion, reducing the underestimation of slow diffusion.

[0049] Example 2

[0050] As Figure 2 shown, this Example provides a high b-value DWI image generation method.

[0051] Step 201, Acquire at least 4 b-value DWI images, each b-value DWI image is acquired multiple times, and the complex signal is retained;

[0052] Step 202: aligning phases of multiple scan data with the same b value using a cross-correlation algorithm;

[0053] Step 203: averaging the registered complex signals, and obtaining a low-noise DWI image from the averaged complex modulus values;

[0054] Step 204 : Generate a high b-value DWI image based on the bi-exponential model by using a piecewise fitting method.

[0055] In step 201, the acquisition of at least four b-value DWI images is performed using a 3.0T or higher MRI device equipped with a phased array coil, which must cover the perfusion sensitive area (b<200s / mm 2 ) and diffusion-dominated region (b ≥ 200s / mm 2 ), for example: b = 0, 50, 100, 200, 400, 800 s / mm 2 .

[0056] In step S202, the phase alignment using the cross-correlation algorithm includes: selecting the complex image of the first scan and the complex image of the k-th scan, determining the maximum displacement by maximizing the cross-correlation function, and applying displacement compensation to the complex image of the k-th scan based on the maximum displacement to achieve phase alignment of the reference image and the image to be compensated.

[0057] In step S203, the formula for averaging the registered complex signals is:

[0058] Where, N is the number of repetitions, S real,k (b) and S imag,k (b) Real and imaginary signals of the kth scan.

[0059] In step S204, the formula of the double exponential model is: Where f is the perfusion fraction, D f Perfusion-related diffusion coefficient, D S True water diffusion coefficient.

[0060] The segmented fitting method for generating high b-value DWI images includes: when the b-value is greater than or equal to 200 s / mm 2 When D S and S0(1-f); when b is less than 200s / mm 2 When D is obtained, S , solve f and D by nonlinear least squares method f .

[0061] The piecewise fitting method shown to generate high b-value DWI images also includes minimizing the residual using the Levenberg-Marquardt algorithm:

[0062] The method of generating a high b-value DWI image comprises: based on the fitting parameters (f, D f ,D S ) Calculate the target high b value (such as b = 2000s / mm 2 ) signal strength:

[0063] Generating a high b-value DWI image includes: retaining phase information (directly multiplexing the phase of low b-value data or generating it by interpolation) to generate a complex signal: in This is the phase diagram when the b value is 0.

[0064] The generating of the high b-value DWI image further includes: performing complex averaging on different generated results to further reduce noise.

[0065] In summary, the high-b-value DWI image generation method proposed in this paper, by combining a biexponential diffusion model with complex signal averaging technology, aims to overcome the signal-to-noise ratio limitations and model simplification errors of traditional methods, and provide an efficient and reliable solution for the clinical popularization of high-b-value DWI.

[0066] Example 3

[0067] Perfusion Component in the Bi-Exponential IVIM Model and diffusion component When b<200s / mm 2 In order to overcome the signal aliasing problem in the low b-value interval, this embodiment introduces the SVD deconvolution algorithm based on the double exponential model, thereby optimizing parameter estimation and further improving the quality of the generated high b-value image. The implementation process is as follows: Figure 3 As shown, the details are as follows:

[0068] Step 301: Data preprocessing:

[0069] For multiple scans of the same b value, the phases are aligned by cross-correlation and then the complex average is taken to generate a low-noise DWI signal S(b); a multi-b data matrix is ​​constructed. Assuming that N voxels and M b values ​​are collected (e.g., b = 0, 50, 100, 200, 400, 800 s / mm 2 ), construct the signal matrix:

[0070]

[0071] Each row represents the signal attenuation curve of a voxel at different b values.

[0072] Step 302: SVD decomposition and perfusion component extraction:

[0073] S=U∑V T , where U represents the left singular vector matrix (spatial pattern), Σ represents the singular value diagonal matrix (energy ordering), and V represents the right singular vector matrix (b-value correlation pattern).

[0074] Perfusion component identification: Perfusion characteristics: The perfusion-related signal decays rapidly at low b values, and the corresponding right singular vector (column of V) is in the low b value region (e.g. b<200s / mm 2 ) have significant changes.

[0075] Select principal components: retain the components corresponding to the first K largest singular values ​​(e.g. K = 2, corresponding to perfusion and diffusion respectively), and reconstruct the signal: Among them S perf A perfusion-dominated signal matrix.

[0076] Step 303: fitting the parameters of the double exponential model:

[0077] Deperfusion processing: Subtract the perfusion component from the original signal to obtain the diffusion-dominated signal, the formula is S diff =SS perf ,

[0078] Parameter fitting: S diff Apply a single exponential model to fit D s : Reusing the original signal and fitting D s , infer f and D through low b value data f .

[0079] Step 304: Generate a high b-value image:

[0080] Signal prediction: based on fitting parameters f,D f ,D s , generate target high b value (such as b = 2000s / mm 2 ) signal, the formula is

[0081] Complex signal reconstruction: Multiplex the phase information with a b value of 0 to generate a complex high b value image. The formula is in This is the phase diagram when the b value is 0.

[0082] This embodiment introduces the SVD deconvolution algorithm based on the double exponential model, applies a singular matrix decomposition to the original signal to extract the perfusion component, subtracts the perfusion component from the original signal to obtain the diffusion-dominant signal, and applies a single exponential model to fit the diffusion-dominant signal. s , then use the original signal and fitting D s , infer f and D through low b value data f, thereby optimizing the parameters f and D f Estimate and further improve the quality of the generated high b-value image to overcome the signal aliasing problem in the low b-value range.

[0083] It should be noted that the technical features in the above embodiments can be combined in any way, and the technical solutions formed by the combination all fall within the scope of protection of this application. In this article, terms such as "including", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0084] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating a high b-value DWI image, comprising the following steps: Step S1, acquiring at least four b-value DWI images, acquiring each b-value DWI image multiple times, and retaining complex signals; Step S2: aligning the phases of multiple scan data with the same b value using a cross-correlation algorithm; Step S3, averaging the registered complex signals, and obtaining a low-noise DWI image from the averaged complex modulus values; Step S4: Generate a high b-value DWI image based on the bi-exponential model through a piecewise fitting method.

2. The method for generating a high b-value DWI image according to claim 1, wherein: In step S1, acquiring at least four b-value DWI images is achieved by using an MRI device greater than or equal to 3.0 T equipped with a phased array coil.

3. The method for generating a high b-value DWI image according to claim 1, wherein: In step S2, the phase alignment using the cross-correlation algorithm includes: selecting the complex image of the first scan and the complex image of the k-th scan, determining the maximum displacement by maximizing the cross-correlation function, and applying displacement compensation to the complex image of the k-th scan based on the maximum displacement to achieve phase alignment of the reference image and the image to be compensated.

4. The method for generating a high b-value DWI image according to claim 1, wherein: In step S3, the formula for averaging the registered complex signals is: Where, N is the number of repetitions, S real,k (b) and S imag,k (b) Real and imaginary signals of the kth scan.

5. The method for generating a high b-value DWI image according to claim 1, wherein: In step S4, the formula of the double exponential model is: Where S0 is the signal intensity when b=0, f is the perfusion fraction, and D f Perfusion-related diffusion coefficient, D s True water diffusion coefficient.

6. The method for generating a high b-value DWI image according to claim 5, wherein: The segmented fitting method for generating high b-value DWI images includes: when the b-value is greater than or equal to 200 s / mm 2 When D s and S0(1-f); when b is less than 200s / mm 2 When D is obtained, s , solve f and D by nonlinear least squares method f .

7. The method for generating a high b-value DWI image according to claim 6, wherein: The piecewise fitting method shown to generate high b-value DWI images also includes minimizing the residual using the Levenberg-Marquardt algorithm:

8. The method for generating a high b-value DWI image according to claim 7, wherein: The method of generating a high b-value DWI image comprises: based on the fitting parameters (f, D f ,D s ) Calculate the signal intensity of the target high b value:

9. The method for generating a high b-value DWI image according to claim 8, wherein: The generating of a high b-value DWI image comprises: retaining phase information to generate a complex signal, in is the phase diagram with b value zero.

10. The method for generating a high b-value DWI image according to claim 9, wherein: The generating of the high b-value DWI image includes: performing complex averaging on different generated results to further reduce noise.

Images