A method and system for dual energy decomposition of dual energy x-ray images

Abstract

The application discloses a method and system for dual-energy decomposition of dual-energy X-ray images. The method comprises the following steps: deducting dark fields in low-energy detector signals and high-energy detector signals, and performing normalization processing on bright field minus dark field image values; calculating a loss function; setting minimum initial values and maximum initial values of equivalent atomic numbers and mass thicknesses, using an iterative method, selecting two trial points according to a golden section search method in each iteration step, comparing relative sizes of first loss function values and second loss function values corresponding to the two trial points to solve; and obtaining final solutions of the equivalent atomic numbers and the mass thicknesses when a convergence condition is met. According to the application, the minimum value of the dual-energy decomposition loss function can be effectively found without increasing the number of iterations and without needing to derive the equivalent atomic number, so that the equation solution for the dual-energy decomposition of the dual-energy X-ray images is found.

Description

Technical Field

[0001] This invention relates to a method for dual-energy decomposition of dual-energy X-ray images, and also to a corresponding system, belonging to the field of medical imaging technology. Background Technology

[0002] Dual-energy X-ray image decomposition (DEXA) is an innovative medical imaging technique that acquires and analyzes patient images by combining two different energy levels of X-rays (typically high and low energy). Based on the differences in how different tissues absorb X-rays of varying energies, this technique effectively distinguishes and identifies bone, soft tissue, and other tissue types within the body. Compared to traditional single-energy X-ray imaging, DEXA not only provides images with higher contrast but also reveals more details about tissue characteristics and pathological conditions.

[0003] Dual-energy X-ray image decomposition technology has a wide range of applications, playing a vital role in medical diagnosis and treatment planning. In orthopedics, dual-energy X-ray image decomposition helps physicians assess the type and extent of fractures, as well as changes in bone density. In oncology, it can be used for tumor detection, grading, and monitoring of treatment effectiveness. Furthermore, in the diagnosis of vascular diseases, this technology helps identify calcified plaques within blood vessels, providing crucial information for cardiovascular disease risk assessment.

[0004] Despite the significant advantages of dual-energy X-ray image decomposition technology, its implementation faces several challenges. For example, Chinese invention patent application number 202210245557.6 discloses a dual-energy CT image decomposition method based on an iterative residual network. This method first selects models matching the detected body part from a human model library to construct a training dataset. Then, a specialized iterative residual network is designed and trained using this dataset, with the network parameters saved after training. After network training, dual-energy projection data of the detected object is collected, and a single-energy CT image reconstruction algorithm is applied to reconstruct high-energy and low-energy CT images. Finally, the trained network is used to decompose these two sets of CT images to obtain the matrix material density image of the detected object, thereby improving the accuracy and efficiency of medical image analysis. However, while this method can achieve effective image decomposition, its deployment requires specialized GPUs and software frameworks. Furthermore, the method's high dependence on the training dataset limits its transferability between different datasets. Summary of the Invention

[0005] The primary technical problem to be solved by this invention is to provide a method for dual-energy decomposition of dual-energy X-ray images.

[0006] Another technical problem to be solved by the present invention is to provide a system for dual-energy decomposition of dual-energy X-ray images.

[0007] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution:

[0008] According to a first aspect of the present invention, a method for dual-energy decomposition of dual-energy X-ray images is provided, comprising:

[0009] S1: Subtract the dark field from the low-energy detector signal and the high-energy detector signal, and normalize the image value of the bright field minus the dark field.

[0010] S2: Calculate the loss function using the following formula.

[0011] S3: Setting Z min Z max T min and T max The initial value is obtained by using a one-dimensional minimization method in [T]. min T max Solve Z within the range min The corresponding T min The optimal value of Z max The corresponding T max Optimal value, Loss min and Loss max ;

[0012] Among them, Z min Z is the smallest equivalent atomic number. max For the maximum equivalent atomic number, T min For minimum mass thickness, T max For maximum mass thickness, Loss min Loss is the minimum value of the loss function. max This represents the maximum value of the loss function.

[0013] S4: According to the golden section search method, take Z1 = Z min +0.382×(Z max -Z min Z2 = Z min +0.618×(Z max -Z min Using a one-dimensional minimization method in [T] min T max Find the optimal value of T1 corresponding to Z1, the optimal value of T2 corresponding to Z2, Loss1, and Loss2 within the range;

[0014] Where Z1 is the smallest trial value of Z in the current iteration step, Z2 is the largest trial value of Z in the current iteration step, T1 is the optimal T value corresponding to Z1 in the current iteration step, T2 is the optimal T value corresponding to Z2 in the current iteration step, Loss1 is the first loss function value, and Loss2 is the second loss function value.

[0015] S5: Compare the relative magnitudes of the first and second loss function values ​​to solve the problem;

[0016] S6: Repeat steps S4 to S5 until Z2 - Z1 is less than ΔZ; where ΔZ is a predetermined convergence threshold.

[0017] S7: The final solution for obtaining the equivalent atomic number is Z0 = (Z1 + Z2) / 2, and the final solution for the mass thickness is calculated.

[0018] Preferably, in step S5, if the first loss function value is greater than the second loss function value, then Z is taken. min =Z1, Z1=Z2, Z2=Z min +0.618×(Z max -Z min Loss min =Loss1, Loss1=Loss2, T min =T1, T1=T2, and in [T min Find the optimal value of T2 and the second loss function value corresponding to Z2 within the range of T1].

[0019] Preferably, in step S5, if the first loss function value is less than or equal to the second loss function value, then Z is taken. max =Z2, Z2=Z1, Z1=Z min +0.382×(Z max -Z min Loss max =Loss2, Loss2=Loss1, T max =T2, T2=T1, and in [T2, T max Find the optimal value of T1 and the first loss function value corresponding to Z1 within the range.

[0020] Preferably, in step S7, if T2 - T1 is less than ΔT, then the final solution of T is T0 = (T1 + T2) / 2; where ΔT is a predetermined convergence threshold. If T2 - T1 is greater than or equal to ΔT, then the corresponding optimal T0 is obtained through a one-dimensional minimization method, and the final solution of T is T0.

[0021] Preferably, the optimal T0 is obtained by the one-dimensional minimization method, which is achieved by the following formula:

[0022]

[0023] Where i = 1 or 2, 0 < α < 1.

[0024] According to a second aspect of the present invention, a system for dual-energy decomposition of dual-energy X-ray images is provided, comprising a central processing unit, a memory, and an image processor; wherein the memory is coupled to the central processing unit and the image processor respectively, and is used to store computer programs and images, wherein when the computer program is executed by the central processing unit, the central processing unit implements the above-described method.

[0025] Compared with existing technologies, this invention combines the characteristics of search and fitting methods, taking into account the properties of the loss function in dual-energy decomposition. This avoids the shortcomings of low accuracy and limited applicability of fitting methods, and also circumvents the large storage and computational requirements of search methods. Using this invention, the minimum value of the dual-energy decomposition loss function can be effectively found without significantly increasing the number of iterations or requiring differentiation of the equivalent atomic number. This leads to the solution to the equations for dual-energy decomposition of dual-energy X-ray images, namely, the equivalent atomic number and mass thickness (or electron density). Attached Figure Description

[0026] Figure 1A This is a schematic diagram of the two-dimensional optimization results of the error function obtained by the steepest descent method and the conjugate gradient method in the prior art.

[0027] Figure 1B This is a schematic diagram illustrating the two-dimensional optimization result of the error function obtained by the Hook-Jeeves method in the prior art.

[0028] Figure 2 This is a flowchart of a method for dual-energy decomposition of dual-energy X-ray images provided in an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of a system structure for dual-energy decomposition of dual-energy X-ray images, provided as an embodiment of the present invention. Detailed Implementation

[0030] The technical content of the present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0031] In existing technologies, dual-energy X-ray image decomposition methods mainly include the dual-effect decomposition method and the dual-material decomposition method. In the dual-effect decomposition method, the low-energy detector signal is:

[0032]

[0033] Where E is, E L For, SL Let T be the mass thickness (or electron density), and A be the electron density. OE f is the coefficient of the photoelectric effect. OE This is the formula for the photoelectric effect, where Z is the equivalent atomic number and A is... KN f is the coefficient of the Compton effect. KN The formula for the Compton effect; E L S represents the upper limit of X-ray energy received by the low-energy detector. L This is the X-ray energy spectrum received by low-energy detectors.

[0034] The high-energy detector signal is:

[0035]

[0036] Among them, E H S represents the upper limit of X-ray energy received by the high-energy detector. H This is the X-ray energy spectrum received by the high-energy detector.

[0037] In the dual-material decomposition method, the low-energy detector signal is:

[0038]

[0039] Wherein, B1 is the coefficient of the first material, μ1 is the X-ray attenuation coefficient of the first material, B2 is the coefficient of the second material, and μ2 is the X-ray attenuation coefficient of the second material.

[0040] The high-energy detector signal is:

[0041]

[0042] The expression for the error function is:

[0043]

[0044] Wherein, L0 is the measured low-energy detector signal and H0 is the measured high-energy detector signal, both of which have been normalized (normalization is a routine operation in X-ray image processing, and will not be elaborated here).

[0045] Since L and H are functions of the equivalent atomic number and mass thickness (the product of the X-ray attenuation coefficient of an object and the object's thickness, reflecting the overall attenuation effect of X-rays when passing through the object), and the error function is a function of L and H, the error function will change with the changes in the equivalent atomic number and mass thickness.

[0046] Typically, the objective function for optimization approximates a quadratic function in the neighborhood of the minimum point, with its contour lines forming an ellipse. This can be solved using methods such as steepest descent and conjugate gradient methods. However, see [link to other resources]. Figure 1A and Figure 1B As shown, the minimum of the loss function in the bienergy decomposition lies in a narrow valley with very steep sides and a very flat bottom, with minimal elevation change along the valley floor. The contour lines around the valley are not elliptical (O-shaped) but rather hyperbolic (X-shaped), and the valley width increases with distance from the minimum point and decreases closer. Common optimization methods, such as steepest descent, conjugate gradient, or Hook-Jeeves methods, struggle to reach the true minimum. This is because all optimization methods provide a driving force perpendicular to the valley, while the driving force along the valley towards the minimum point is weak, and the valley's funnel shape may even create a reverse driving force. Therefore, even after many iterations, the true minimum cannot be reached; finally, after reaching the maximum number of iterations, the process stops far from the minimum point.

[0047] After in-depth research, the inventors discovered that the two variables—equivalent atomic number and mass thickness—represent completely different physical meanings and have entirely different roles and contributions in the above equations. Conventional optimization methods treat all variables the same way, assuming they are essentially isotropic, failing to consider the highly anisotropic nature of the variable space. Furthermore, for the dual-material method, the X-ray attenuation coefficient μ is obtained from a commonly used database (such as the XCOM database from the National Institute of Standards and Technology), and is not an analytical expression. Therefore, it is not differentiable with respect to the equivalent atomic number, and consequently, the low-energy detector signal and the high-energy detector signal are also not differentiable with respect to the equivalent atomic number. This limits the application of some optimization methods; in practice, only a differentiable polynomial expression can be fitted before solving.

[0048] To address the above issues, this invention proposes a dual-energy X-ray image decomposition method that combines search and fitting methods. This method can effectively find the minimum value of the dual-energy decomposition loss function without significantly increasing the number of iterations or requiring differentiation of the equivalent atomic number. This allows for finding the solution to the equation for dual-energy X-ray image decomposition, namely the equivalent atomic number and mass thickness (or electron density).

[0049] First Embodiment

[0050] like Figure 2 As shown, the first embodiment of the present invention provides a method for dual-energy decomposition of dual-energy X-ray images, comprising the following steps:

[0051] S1: Subtract the dark field from the low-energy detector signal L and the high-energy detector signal H, and normalize the image value of the bright field minus the dark field.

[0052] It should be noted that the embodiments of the present invention are based on a scene with known low-energy detector signal L and high-energy detector signal H (i.e., the image value of the dark field is subtracted and the bright field is normalized so that each pixel value is a real number between 0 and 1), and the corresponding equivalent atomic number Z and mass thickness or electron density T are solved.

[0053] In X-ray imaging, the dark field refers to the X-ray signal received by the detector when there is no object being measured. These signals include background noise and scattered rays, which can affect image quality and the accuracy of analysis. Dark field subtraction involves subtracting this background signal from the actual measured detector signal to eliminate or reduce these adverse factors. On the other hand, the bright field refers to the X-ray signal received by the detector when the object being measured is present. After dark field subtraction, to further improve image quality, bright field images are usually normalized. Normalization is a mathematical operation that scales the pixel values ​​of an image to a specific range, typically between 0 and 1. This improves image contrast, makes image details clearer, and provides standardized data for subsequent image analysis and processing steps.

[0054] The preprocessing steps described above are crucial for improving the accuracy of dual-energy X-ray image decomposition methods. Dark field subtraction reduces noise and artifacts in the image, while normalization ensures the consistency and comparability of image data in subsequent processing. These operations contribute to improving the performance of dual-energy X-ray image decomposition methods, thereby obtaining more accurate parameters such as equivalent atomic number and mass thickness.

[0055] In dual-energy X-ray image decomposition methods, the equivalent atomic number (Z) and mass thickness (T) are two key parameters, representing the equivalent value of the material's atomic number and the degree of X-ray absorption, respectively. Determining these two parameters is crucial for accurate image decomposition. The golden section search method and the one-dimensional minimization method are two different mathematical optimization methods, each with its own advantages in solving for these two parameters.

[0056] The Golden Section Search (GMS) is a derivative-free optimization method that uses the golden ratio (0.618) to find the extreme points of a function. Specifically, it gradually narrows the search interval and uses information from previous iterations to determine new search points, thus progressively approaching the optimal solution. This method does not require information about the derivative of the objective function, making it suitable for situations where derivatives are difficult to obtain or computationally expensive. When solving for the equivalent atomic number, since the atomic number is usually related to the absorption characteristics of X-rays but does not necessarily have a clear analytical expression or derivative, the GMS can effectively search for the optimal Z value within a certain range.

[0057] On the other hand, one-dimensional minimization methods refer to methods for finding the minimum value of a function in one-dimensional space, encompassing various approaches. Among these, the steepest descent method can be considered a special case of one-dimensional minimization. It finds the minimum value of the function by calculating the gradient (derivative) of the objective function with respect to the variables and updating the values ​​of the variables along the direction of the steepest gradient descent.

[0058] The core of the one-dimensional minimization method for solving mass thickness lies in precisely adjusting the mass thickness parameters to minimize the difference between the reconstructed image and the actual observed data after determining the equivalent atomic number. This method starts with an initial estimate of the mass thickness and then iteratively approaches the optimal solution. In each iteration, the first derivative (gradient) of the loss function with respect to the mass thickness is used to determine a search direction, and a linear search is performed in this direction to find the step size that reduces the loss function. By setting reasonable convergence conditions, such as the magnitude of the gradient or the magnitude of the change in mass thickness being below a preset threshold, the minimum point of the loss function can be effectively found while ensuring computational efficiency, thus determining the optimal mass thickness value.

[0059] S2: Calculate the loss function using the following formula.

[0060] In dual-energy X-ray image decomposition methods, the loss function is defined based on the difference between the actual measured detector signal and the model's predicted signal. A smaller loss function value indicates that the model's prediction is closer to the actual observation, meaning the model performs better. In the Z-log(T) coordinate system, the loss function exhibits a monotonically decreasing trend in mass thickness T corresponding to its minimum value as the equivalent atomic number Z increases. This characteristic allows for effective prediction and constraint of the optimal solution range of T based on the value of Z during optimization, thereby improving the efficiency and accuracy of the algorithm in finding the optimal equivalent atomic number and mass thickness. In a preferred embodiment of the invention, because the bottom of the "valley" of the loss function is monotonically changing in the Z-log(T) coordinate system—that is, at larger equivalent atomic numbers Z, the mass thickness T at the minimum value of the loss function is smaller—the range of the mass thickness T during optimization can be constrained accordingly.

[0061] S3: Setting Z min Z max T min and T max The initial value is obtained by using a one-dimensional minimization method in [T]. min T max Solve Z within the range min The corresponding T min The optimal value of Zmax The corresponding T max The optimal value, the minimum value of the loss function. min and the maximum value of the loss function Loss max .

[0062] Among them, Z min Z is the smallest equivalent atomic number. max For the maximum equivalent atomic number, T min For minimum mass thickness, T max This represents the maximum mass thickness.

[0063] In step S3, it is required to first determine the minimum value (Z0) of the equivalent atomic number (Z) and mass thickness (T). min ,T min ) and maximum value (Z) max ,T max These initial values ​​constitute the search range. Within this range, step S3 uses a one-dimensional minimization method to find the minimum value of the loss function. min ) and maximum value (Loss) max This typically means that the optimal combination of Z and T parameters has been found, minimizing the difference between the model's predictions and the actual observed data. Simultaneously, the performance of the loss function under different Z and T values ​​is recorded to evaluate the stability and convergence of the optimization process. To ensure the efficiency of the optimization process, a convergence threshold ΔZ can be set. When the change in Z during consecutive iterations is less than this threshold, it is considered that the optimal solution has been approached, and further iterations can be stopped. The Z and T values ​​at this point are the optimal parameters sought.

[0064] S4: According to the golden ratio search method, take the trial point Z1 = Z min +0.382×(Z max -Z min Z2 = Z min +0.618×(Z max -Z min Using a one-dimensional minimization method in [T] min T max Within the range, find the optimal value of T1 corresponding to Z1 and the first loss function value Loss1 corresponding to T1, as well as the optimal value of T2 corresponding to Z2 and the second loss function value Loss2 corresponding to T2.

[0065] Where Z1 is the smallest trial value of Z in the current iteration step, Z2 is the largest trial value of Z in the current iteration step, T1 is the optimal T value corresponding to Z1 in the current iteration step, and T2 is the optimal T value corresponding to Z2 in the current iteration step.

[0066] In step S4, the golden section search method is used to further refine the search for the optimal value of the equivalent atomic number (Z). Specifically, within the initial search range of Z determined in step S3, two trial points Z1 and Z2 are selected using the golden ratio (approximately 0.618). These two points are located at Z1 and Z2 respectively. min and Z max Between these two points, a certain proportional relationship is maintained. Then, using the one-dimensional minimization method within the search range of mass thickness (T), the corresponding optimal T values ​​(T1 and T2) and the corresponding loss function values ​​(first loss function value Loss1 and second loss function value Loss2) are obtained for the two trial points respectively. The purpose of step S4 is to narrow the search range and approach the global minimum of the loss function by comparing the loss function values ​​of the two trial points, thereby gradually determining the optimal solutions for the equivalent atomic number Z and mass thickness T.

[0067] S5: Compare the relative magnitudes of the first loss function value and the second loss function value, and then solve the problem.

[0068] If the first loss function value Loss1 is greater than the second loss function value Loss2, then take Z. min =Z1, Z1=Z2, Z2=Z min +0.618×(Z max -Z min Loss min =Loss1, Loss1=Loss2, T min =T1, T1=T2, and in [T min Find the optimal value of T2 and the second loss function value Loss2 corresponding to Z2 within the range of T1].

[0069] If the first loss function value Loss1 is less than or equal to the second loss function value Loss2, then take Z. max =Z2, Z2=Z1, Z1=Z min +0.382×(Z max -Z min Loss max =Loss2, Loss2=Loss1, T max =T2, T2=T1, and in [T2, T max Find the optimal value of T1 and the first loss function value Loss1 corresponding to Z1 within the range.

[0070] In step S5, the two loss function values ​​(Loss1 and Loss2) obtained in step S4 are compared. These two values ​​correspond to two trial points Z1 and Z2 selected using the golden section search method, respectively. In this step, the relationship between Loss1 and Loss2 is first evaluated to determine which trial point is closer to the global minimum of the loss function. If Loss1 is less than Loss2, it indicates that the solution corresponding to the smaller trial value Z1 is better, and Z is updated. min Let Z1 be Z2, and then update Z2 to Z. min With Z max If a new, larger value is found, T2 and Loss2 are recalculated. Conversely, if Loss1 is greater than Loss2, Z is updated. max Let Z2 be Z1, and let Z1 be Z2. min With Z max The algorithm iterates through smaller trial values ​​and then recalculates new T1 and Loss1. This process helps to progressively narrow the search interval and, by iteratively approaching the minimum of the loss function, ultimately finds the optimal equivalent atomic number Z and mass thickness T. This method leverages the distribution characteristics of the loss function in the parameter space, effectively narrowing the search range and improving the efficiency and accuracy of the optimization algorithm by selectively updating trial points.

[0071] S6: Repeat steps S4 to S5 until Z2 - Z1 is less than ΔZ, where ΔZ is a predetermined convergence threshold.

[0072] Step S6 involves repeatedly applying steps S4 and S5 to progressively narrow the search range for the equivalent atomic number (Z) until a preset convergence condition is met. In this step, the optimization algorithm iterates continuously, updating the search range of Z each time based on the better attempt points determined in step S5, and recalculating the corresponding mass thickness (T) and loss function value. This process continues until the change in Z (ΔZ) during consecutive iterations is less than a pre-set convergence threshold. At this point, it can be considered sufficiently close to the global minimum of the loss function, and further iterations will not significantly improve the result.

[0073] S7: The final solution for obtaining the equivalent atomic number is Z0 = (Z1 + Z2) / 2, and the final solution for the mass thickness is calculated.

[0074] If T2 - T1 is less than ΔT, then the final solution of T is T0 = (T1 + T2) / 2, where ΔT is a predetermined convergence threshold.

[0075] If T2 - T1 is greater than or equal to ΔT, then the corresponding optimal T0 is obtained by one-dimensional minimization method, and the final solution of T is T0.

[0076] Step S7 is used to accurately determine the optimal values ​​of the equivalent atomic number (Z) and mass thickness (T) after the convergence condition is met. In this step, the average of the two trial points Z1 and Z2 obtained in the last iteration is taken as the final solution of the equivalent atomic number, i.e., Z0 = (Z1 + Z2) / 2. For the mass thickness T, if the difference (ΔT) between T1 and T2 calculated in the last iteration is less than the preset convergence threshold, then the final solution of T is the average of T1 and T2, T0 = (T1 + T2) / 2; if ΔT is greater than or equal to the convergence threshold, then the optimal value of T0 is further searched between T1 and T2 using a one-dimensional minimization method. Through the above steps, it is ensured that a physically reasonable and mathematically optimal parameter solution is obtained on the basis of meeting the convergence condition, thus completing the effective decomposition of the dual-energy X-ray image.

[0077] In a preferred embodiment of the present invention, the corresponding optimal T0 is obtained by a one-dimensional minimization method, which is achieved by the following formula:

[0078]

[0079] Where i = 1 or 2 indicates that the search starts from T1 or T2; 0 < α < 1 is a manually set reduction factor used to control the search step size, for example, α = 0.1 can be set.

[0080] because and Loss(L, H) has an analytical expression, therefore and All of these have analytical expressions and can be calculated directly. The above formula calculates the gradient (i.e., derivative) of the loss function at T1 or T2, and searches in the opposite direction of the gradient with a step size α, thereby updating the value of T. This process is repeated until the minimum point of the loss function is found, at which point the value of T is the optimal mass thickness T0.

[0081] Second Embodiment

[0082] Based on the first embodiment described above, the second embodiment of the present invention provides a system for dual-energy decomposition of dual-energy X-ray images. For example... Figure 3As shown, the system includes a central processing unit 21, a memory 22, and an image processor 23. The memory 22 is coupled to both the central processing unit 21 and the image processor 23, and is used to store one or more programs, as well as dual-energy X-ray images obtained from external sources. These images can be high- and low-energy images obtained from a dual-energy X-ray scanning imaging system or a DR flat panel imaging system, high- and low-energy projection images obtained from a dual-energy CT system, or high- and low-energy slice images reconstructed from a CT system. When the one or more programs are executed by the central processing unit 21, the central processing unit 21 implements the dual-energy decomposition method for dual-energy X-ray images as described in the above embodiment.

[0083] The central processing unit 21 controls the overall operation of the method for dual-energy X-ray image decomposition, completing all or part of the steps of the method. The image processor 23 executes the dual-energy X-ray image decomposition method described in the above embodiment on all pixels of the image in parallel. It should be noted that, in some cases, the image processor 23 can be replaced by the central processing unit 21.

[0084] Memory 22 is used to store various types of data and images to support operation on the system. This data may include, for example, instructions for any application or method used to operate on the system, as well as application-related data. These images may include, for example, dual-energy X-ray images obtained externally; these may be high- and low-energy images obtained by a dual-energy X-ray scanning imaging system or a DR flat panel imaging system; high- and low-energy projection images obtained by a dual-energy CT system; or high- and low-energy slice images obtained after reconstruction by a CT system. Memory 22 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, etc.

[0085] In one exemplary embodiment, the system may be implemented by a computer chip or physical entity, or by a product with certain functions, for performing the aforementioned method of dual-energy X-ray image decomposition and achieving the same technical effect as described above. A typical embodiment is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interface device, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0086] In another exemplary embodiment, the present invention also provides a computer-readable storage medium including program instructions that, when executed by a central processing unit, implement the steps of the method for dual-energy decomposition of dual-energy X-ray images in any of the above embodiments. For example, the computer-readable storage medium may be the memory including the program instructions described above, which can be executed by a central processing unit to complete the method for dual-energy decomposition of dual-energy X-ray images and achieve the same technical effects as the method described above.

[0087] It should be noted that the above embodiments are merely examples, and the technical solutions of each embodiment can be combined, all of which are within the protection scope of this invention.

[0088] Compared with existing technologies, this invention combines the characteristics of search and fitting methods, taking into account the properties of the loss function in dual-energy decomposition. This avoids the drawbacks of low accuracy and limited applicability of fitting methods, while also mitigating the high storage and computational demands of search methods. Using this invention, the minimum value of the dual-energy decomposition loss function can be effectively found without significantly increasing the number of iterations or requiring differentiation of the equivalent atomic number. This leads to the solution to the equations for dual-energy decomposition of dual-energy X-ray images, namely, the equivalent atomic number and mass thickness (or electron density).

[0089] The method and system for dual-energy X-ray image decomposition provided by this invention have been described in detail above. Any obvious modifications made by those skilled in the art without departing from the essence of this invention will constitute an infringement of the patent rights of this invention and will incur corresponding legal liability.

Claims

1. A method for dual-energy decomposition of dual-energy X-ray images, characterized in that... Includes the following steps: S1: Subtract the dark field from the low-energy detector signal and the high-energy detector signal, and normalize the image value of the bright field minus the dark field. S2: Calculate the loss function using the following formula. S3: The mass thickness is solved using a one-dimensional minimization method; S4: The equivalent atomic number is solved by the golden section search method, and the first and second loss function values ​​are obtained by the one-dimensional minimization method. S5: Compare the relative magnitudes of the first and second loss function values ​​to solve the problem; S6: Repeat steps S4 to S5 until Z2 - Z1 is less than ΔZ; where ΔZ is a predetermined convergence threshold. S7: The final solution for obtaining the equivalent atomic number is Z0 = (Z1 + Z2) / 2, and the final solution for the mass thickness is calculated.

2. The method as described in claim 1, characterized in that... In step S3, Z is set min Z max T min and T max The initial value is obtained by using a one-dimensional minimization method in [T]. min T max Solve Z within the range min The corresponding T min The optimal value of Z max The corresponding T max Optimal value, Loss min and Loss max ; Among them, Z min Z is the smallest equivalent atomic number. max For the maximum equivalent atomic number, T min For minimum mass thickness, T max For maximum mass thickness, Loss min Loss is the minimum value of the loss function. max This represents the maximum value of the loss function.

3. The method as described in claim 2, characterized in that... In step S4, according to the golden section search method, the trial point Z1 = Z is selected. min +0.382×(Z max -Z min Z2 = Z min +0.618×(Z max -Z min Using a one-dimensional minimization method in [T] min T max Within the range, find the optimal value of T1 corresponding to Z1 and the Loss1 corresponding to T1, as well as the optimal value of T2 corresponding to Z2 and the Loss2 corresponding to T2; Where Z1 is the smallest trial value of Z in the current iteration step, Z2 is the largest trial value of Z in the current iteration step, T1 is the optimal T value corresponding to Z1 in the current iteration step, T2 is the optimal T value corresponding to Z2 in the current iteration step, Loss1 is the first loss function value, and Loss2 is the second loss function value.

4. The method as described in claim 3, characterized in that... In step S5, if the first loss function value is greater than the second loss function value, then Z is taken. min =Z1, Z1=Z2, Z2=Z min +0.618×(Z max -Z min Loss min =Loss1, Loss1=Loss2, T min =T1, T1=T2, and in [T min Find the optimal value of T2 and the second loss function value corresponding to Z2 within the range of T1].

5. The method as described in claim 4, characterized in that... In step S5, if the first loss function value is less than or equal to the second loss function value, then Z is taken. max =Z2, Z2=Z1, Z1=Z min +0.382×(Z max -Z min Loss max =Loss2, Loss2=Loss1, T max =T2, T2=T1, and in [T2, T max Find the optimal value of T1 and the first loss function value corresponding to Z1 within the range.

6. The method as described in claim 5, characterized in that... In step S7, if T2 - T1 is less than ΔT, then the final solution of T is T0 = (T1 + T2) / 2; where ΔT is a predetermined convergence threshold.

7. The method as described in claim 6, characterized in that... In step S7, if T2 - T1 is greater than or equal to ΔT, then the corresponding optimal T0 is obtained by one-dimensional minimization method, and the final solution of T is T0.

8. The method as described in claim 7, characterized in that The optimal T0 is found using a one-dimensional minimization method, achieved through the following formula: Where i = 1 or 2, 0 < α < 1.

9. A system for dual-energy decomposition of dual-energy X-ray images, characterized in that... It includes a central processing unit, a memory, and an image processor; wherein the memory is coupled to the central processing unit and the image processor respectively, and is used to store computer programs and images. When the computer program is executed by the central processing unit, the central processing unit implements the method described in any one of claims 1 to 7.

10. The system as described in claim 9, characterized in that: The image processor is replaced by the central processing unit.

Images