X-ray static rapid multi-contrast signal analysis method based on single projection

Abstract

The invention provides an X-ray static rapid multi-contrast signal analysis method based on single projection, and belongs to the technical field of deep learning and X-ray grating contrast imaging. The method comprises the following steps: acquiring a sample grating projection; a convolutional neural network model is constructed based on Pix2pixHD, edge loss is added in a loss function, and the recovery capability of the network for image edge details is improved; training a convolutional neural network model; and obtaining high-quality absorption, phase and dark field projection by using a trained convolutional neural network and single grating projection. According to the method, the strong information extraction capability of the convolutional neural network is utilized, multi-contrast imaging of a single projection is realized, and compared with a traditional contrast signal analysis method, the radiation dose of the sample can be reduced, the time required by contrast imaging is reduced, and low-dose rapid imaging is realized.

Description

Technical Field

[0001] The present invention belongs to the technical fields of deep learning and X-ray grating contrast imaging, and particularly relates to a method for static and fast multi-contrast signal analysis of X-rays based on a single projection. Background Art

[0002] Compared with traditional absorption contrast imaging, grating interferometry can obtain absorption, phase, and dark-field contrast images of an object. Absorption images have good imaging effects for heavy elements such as metals, bones, and teeth. Differential phase contrast imaging can obtain higher imaging contrast for light element substances such as biological tissues. Dark-field contrast is optimal for imaging low-density tissues, such as the human lungs. The grating interferometry based on the Talbot-Lau effect can use a common X-ray source to generate coherent X-rays through a source grating G0, pass through a phase grating G1 and freely propagate a certain distance, pass through an absorption grating G2, and finally be received by a detector. However, since the grating will absorb most of the X-rays and the absorption grating G2 needs to be moved multiple times (usually 3 - 8 times) to complete the analysis of absorption, phase, and dark-field contrast signals, it greatly increases the radiation dose and time cost during the imaging process.

[0003] Using a single projection can effectively reduce the radiation dose and imaging time, but how to effectively analyze each contrast from it has become a major challenge. Summary of the Invention

[0004] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a method for static and fast multi-contrast signal analysis of X-rays based on a single projection, which can capture complex non-linear relationships in the data by using deep learning technology and achieve high-precision analysis of absorption, phase, and dark-field contrasts. This method not only reduces the requirement for radiation dose, but also reduces the dependence on mechanically moving gratings, avoids motion artifacts caused by multiple scans, improves the efficiency and reliability of imaging, and promotes the application of grating multi-contrast imaging technology in the clinical field.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for static and fast multi-contrast signal analysis of X-rays based on a single projection, comprising:

[0007] Step S101, using an X-ray grating multi-contrast imaging system to obtain at least three step projection sequences of a sample;

[0008] Step S102, constructing a convolutional neural network model based on Pix2pixHD and adding an edge loss to the loss function;

[0009] Step S103: Perform Fourier analysis on at least three step-projection sequences to obtain absorption, phase, and dark-field contrast projections respectively. Use the obtained absorption, phase, and dark-field contrast projections as the ground truth, and use the first step projection of at least three step-projection sequences as the source image. Input the source image and the ground truth at the same angle as paired data into a convolutional neural network for training;

[0010] Step S104: Use the trained convolutional neural network model to obtain absorption, phase, and dark-field contrast projections using a single step projection.

[0011] In a second aspect, the present invention provides an electronic device, including: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the foregoing X-ray static fast multi-contrast signal analysis method based on a single projection.

[0012] In a third aspect, the present invention provides a computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by a processor, the processor can be caused to implement the foregoing X-ray static fast multi-contrast signal analysis method based on a single projection.

[0013] The beneficial effects of the present invention are as follows:

[0014] Compared with the existing X-ray grating multi-contrast signal analysis method, the present invention can effectively reduce the radiation dose received by the sample while reducing the imaging time, and improves the application potential of the grating multi-contrast imaging technology; the deep neural network can capture the complex non-linear relationships in the data and achieve high-precision reconstruction of absorption, phase, and dark-field contrast. In addition, the deep learning-based method has good generalization ability and can adapt to different imaging conditions and objects, which is of great significance for practical applications.

[0015] The single-projection X-ray static fast multi-contrast signal analysis method in the present invention does not require modification of the existing imaging device, has strong scalability, and can effectively reduce the radiation dose. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of an X-ray static fast multi-contrast signal analysis method based on a single projection according to the present invention;

[0017] Figure 2 is a schematic imaging principle diagram of a grating interferometer;

[0018] Figure 3 is a structural diagram of a network generator;

[0019] Figure 4 is a structural diagram of a network discriminator;

[0020] Figure 5 This is the experimental flowchart of the embodiment of the present invention;

[0021] Figure 6 This is the comparison diagram of the implementation effects between the present invention and the prior art. Specific embodiments

[0022] Next, the technical solutions in this embodiment will be clearly and completely described in conjunction with the accompanying drawings in this embodiment. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0023] As Figure 1 shown, this embodiment discloses an X-ray static fast multi-contrast signal analysis method based on a single projection, including the following steps:

[0024] Step S101, obtaining at least three stepped projection sequences of the sample using an X-ray grating multi-contrast imaging system, as follows:

[0025] As Figure 2 shown, the X-ray grating differential phase contrast imaging device includes six parts: an X-ray source, a source grating G0, a sample, a phase grating G1, an absorption grating G2, and a detector;

[0026] Among them, the source grating G0 is used to generate coherent X-rays; the duty cycle of the phase grating G1 is 50%, and an X-ray is used to generate a phase shift of π; the duty cycle of the absorption grating G2 is 50%, and a part of it can completely absorb X-rays, and the other part can transmit X-rays; the detector is used to convert the X-ray energy into a digital image;

[0027] The corresponding relationship between the imaging experimental parameters of the X-ray grating multi-contrast imaging system is as follows:

[0028] (1)

[0029] (2)

[0030] (3)

[0031] (4)

[0032] Wherein, d is the distance between the phase grating G1 and the absorption grating G2, m is an integer representing m times the fractional Talbot distance, k = (L + d) / L is the magnification ratio, g1 is the period of the phase grating G1, λ is the wavelength of the X-ray, g2 is the period of the absorption grating G2, g0 is the period of the source grating G0, L is the distance between the source grating G0 and the phase grating G1, and s is the width of the source grating G0 that allows the X-ray to pass through in each period.

[0033] Step S102: Construct a convolutional neural network model based on Pix2pixHD, add an edge loss to the loss function, and improve the network's ability to restore image details.

[0034] The model consists of a generator and a discriminator, and can convert a single projection with a grating into a contrast projection. The generator consists of two parts, namely the global generator G g and the local enhancement generator G l , which gradually improves the resolution and detail quality of the image. Further, the discriminator adopts a multi-scale discriminator architecture to discriminate the generated contrast projection at different scales, ensuring the authenticity and richness of details of the generated image. The generator and the discriminator achieve Nash equilibrium through adversarial training, thus realizing a high-quality image conversion effect.

[0035] Figure 3 is the structural diagram of the generator in the network model of step S103, which includes the global generator G g and the local enhancement generator G l two modules. The global generator G g adopts a residual network structure, with the input being a single projection with a grating, and the output being a preliminarily generated low-resolution contrast projection, mainly focusing on the generation of the global structure; the local enhancement generator G l takes the output of G g as the input, enhances the local details, and adopts multi-level convolution and upsampling operations to gradually increase the resolution of the image, and finally generates a high-resolution contrast projection.

[0036] The global generator G g contains several residual blocks, and each residual block consists of two convolutional layers with a size of 3×3 and a stride of 1 and a skip connection. G g first performs convolution processing on the input single projection with a grating to extract low-level features, then captures high-level global features through stacking residual blocks, and finally outputs a preliminary low-resolution contrast projection through a convolutional layer.

[0037] The local enhancement generator G l receives the output of G g as the input, and adopts an alternating manner of upsampling layers and convolutional layers to gradually increase the resolution of the image. Gl It contains multiple upsampling modules, and each upsampling module consists of an upsampling layer and several convolutional layers. The upsampling layer uses nearest neighbor interpolation or transposed convolution operation to enlarge the size of the feature map; subsequently, the convolutional layer is used to extract detailed features and enhance the texture and edge information of the image. After multiple upsamplings and convolutions, G l outputs the final high-resolution contrast projection, achieving sufficient enhancement of image details.

[0038] Figure 4 The structure diagram of the discriminator in the network model for step S102 contains three multi-scale discriminators, which respectively process images with sizes of 512×512, 256×256, and 128×128. Each discriminator is sequentially connected by several convolutional layers with a size of 4×4 and a stride of 2, and the number of output channels increases layer by layer, such as 64, 128, 256, 512, etc. The role of the discriminator is to distinguish the difference between the generated contrast projection and the real contrast projection, and guide the generator to generate more realistic images. By making discriminations at different scales, the model can better capture the global structure and detailed features of the image, improving the quality of the generated image.

[0039] The loss function of the model consists of three parts. The loss function of the convolutional neural network model includes adversarial loss, feature matching loss, and edge loss. The adversarial loss is used to train the adversarial relationship between the generator and the discriminator. The feature matching loss is used to match the feature distributions of the generated high-resolution contrast projection and the real contrast projection in the intermediate layer of the discriminator. The edge loss is used to calculate the difference in edge features between the generated high-resolution contrast projection and the real contrast projection:

[0040] (5)

[0041] (6)

[0042] E(I) represents the edge of image I, where I is the input image, and S x is the Sobel horizontal direction operator S y is the Sobel vertical direction operator, represents the convolution operation. L edge is the edge loss, I g is the generated image, I t is the target image, E(I g ) is the edge of the generated image, E(I t ) is the edge of the target image, is the L1 norm, representing the sum of the absolute differences of the corresponding elements of the two edge maps.

[0043] Step S103: Use the first image of at least three step-projection sequences obtained as the source image, and use the three contrast projections obtained by Fourier analysis of the step-projection sequences as the ground truth. Input the paired data of the source image and the ground truth at the same angle into a convolutional neural network for training. During the imaging process, at least three projection sequences are obtained through a step absorption grating, and then the curve of the light intensity varying with the position is obtained by fast Fourier transform:

[0044] (7)

[0045] where x represents the position of the absorption grating, f(x) represents the X-ray intensity corresponding to the absorption grating at position x, a0 is the constant term, a1 is the amplitude, φ1 is the initial phase, and g2 is the period of the absorption grating. Subsequently, the absorption, phase, and dark-field contrast maps of the sample are obtained using the formula:

[0046] (8)

[0047] (9)

[0048] (10)

[0049] where A represents the absorption contrast map, and respectively represent the constant terms of the sample and the background in formula (7), φ represents the phase contrast map, and respectively represent the initial phases of the sample and the background in formula (7), V represents the dark-field contrast map, v s and v r are the dark-field contrast maps of the sample and the background respectively, and respectively represent the amplitudes of the sample and the background in formula (7). Use the first image of the step-projection sequence containing the sample as the source image, and use the three contrast projections obtained by Fourier analysis of the step-projection sequence as the ground truth. Input the paired source image and the ground truth at the same angle into the neural network for training.

[0050] Step S104: Use the trained convolutional neural network model to obtain three contrast projections using a single step projection: Input the first projection of the step-projection sequence into the trained network to obtain absorption, phase, and dark-field contrast projection images.

[0051] In a second aspect, the present invention provides an electronic device, including: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the foregoing method for X-ray static fast multi-contrast signal analysis based on a single projection.

[0052] In a third aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, and when the instructions are executed by a processor, the processor can implement the foregoing X-ray static fast multi-contrast signal analysis method based on a single projection.

[0053] Figure 5 The specific experimental process is shown. First, the PS projection of the sample is obtained using the experimental device. The first image in the projection sequence is used as the source image, and the three contrast projections obtained by Fourier analysis are used as the ground truth for paired training. Finally, the untrained projection is input into the trained neural network to obtain the three contrast projections.

[0054] As Figure 6 shown, experiments were conducted using mice. The left half of the picture shows the absorption, phase, and dark field image results of Fourier analysis, and the right half shows the results of the three contrast images obtained by the present invention. It can be seen that while greatly reducing the dose and time required for imaging, the present invention does not significantly reduce the imaging quality.

[0055] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A single-projection X-ray static fast multi-contrast signal analysis method, characterized in that: The steps include: Step S101, using an X-ray grating multi-contrast imaging system to acquire at least three step projection sequences of a sample; Step S102: construct a convolutional neural network model based on Pix2pixHD, and add edge loss to the loss function; Step S103, performing Fourier analysis on at least three stepping projection sequences to obtain absorption, phase, and dark field contrast projections respectively, taking the absorption, phase, and dark field contrast projections obtained by analysis as true values, taking the first stepping projection of at least three stepping projection sequences as a source image, taking the source image and the true value at the same angle as paired data, and inputting them into a convolutional neural network for training; Step S104: using the trained convolutional neural network model to use a single step projection to obtain three contrast projections: absorption, phase, and dark field.

2. The X-ray static fast multi-contrast signal analysis method based on a single projection according to claim 1, characterized in that: In step S101, the X-ray grating multi-contrast imaging system includes an X-ray source, a source grating, a sample, a phase grating, an absorption grating and a detector arranged in sequence along an optical path; The corresponding relationship between the imaging experimental parameters of the X-ray grating multi-contrast imaging system is as follows: (1) (2) (3) (4) Wherein, d is the distance between the phase grating and the absorption grating, m is an integer, representing m times the fractional Talbot distance, k = (L+d) / L is the magnification ratio, g1 is the period of the phase grating, λ is the wavelength of the X-ray, g2 is the period of the absorption grating, g0 is the period of the source grating, L is the distance between the source grating and the phase grating, and s is the width of the source grating that allows X-rays to pass through in each period.

3. The X-ray static fast multi-contrast signal analysis method based on a single projection according to claim 1, characterized in that: The convolutional neural network model in step S102 includes a generator and a discriminator, the generator includes a global generator and a local enhancement generator, the discriminator adopts a multi-scale discriminator architecture to discriminate the generated contrast projections at different scales, and the generator and the discriminator reach Nash equilibrium through adversarial training.

4. The X-ray static fast multi-contrast signal analysis method based on a single projection according to claim 3, characterized in that: The global generator of the generator includes a first convolutional layer, several residual blocks, and a second convolutional layer connected in series in sequence. The input single grating projection is convolved through the first convolutional layer to extract low-level features, and then stacked through several residual blocks to extract high-level global features. Finally, a low-resolution contrast projection is output through the second convolutional layer.

5. The single-projection-based X-ray static fast multi-contrast signal analysis method according to claim 3, characterized in that: The local enhancement generator of the generator includes multiple serially connected upsampling modules, each of which is composed of an upsampling layer and several convolutional layers. The upsampling layer uses nearest neighbor interpolation or deconvolution operations to enlarge the size of the low-resolution contrast projection output by the global generator, and then extracts detail features through several convolutional layers. After passing through multiple serially connected upsampling modules, the local enhancement generator outputs the final high-resolution contrast projection.

6. The X-ray static fast multi-contrast signal analysis method based on a single projection according to claim 3, characterized in that: The discriminator includes three multi-scale discriminators, which process images of different sizes respectively. Each multi-scale discriminator is sequentially connected by a number of convolutional layers with a size of 4×4 and a step size of 2, and the number of output channels increases layer by layer. The discriminator is used to discriminate the difference between the generated high-resolution contrast projection and the real contrast projection.

7. The X-ray static fast multi-contrast signal analysis method based on a single projection according to claim 1, characterized in that: The loss function of the convolutional neural network model includes adversarial loss, feature matching loss and edge loss. The adversarial loss is used to train the adversarial relationship between the generator and the discriminator. The feature matching loss is used to match the feature distribution of the generated high-resolution contrast projection and the real contrast projection in the middle layer of the discriminator. The edge loss is used to calculate the difference in edge features between the generated high-resolution contrast projection and the real contrast projection: （5） （6） Where E(I) represents the edge of the input image I, S x represents the Sobel horizontal operator, S y represents the Sobel vertical operator, represents the convolution operation, L edge is the edge loss, I g To generate a high-resolution contrast projection, I t is the real contrast projection, E(I g ) is the edge of the generated high-resolution contrast projection, E(I t ) is the edge of the real contrast projection, is the L1 norm.

8. According to the single-projection-based X-ray static fast multi-contrast signal analysis method of claim 1, in step S103, a step curve of light intensity variation with position is obtained by fast Fourier transform for the at least three step projection sequences obtained in step S101: (7) in, x represents the position of the absorption grating, f(x) represents the X-ray intensity corresponding to the absorption grating at position x, a0 is a constant term, a1 is the amplitude, φ1 is the initial phase, g2 is the period of the absorption grating, and the absorption, phase and dark field contrast images of the sample are obtained using the formula: (8) (9) (10) Where A represents the absorption contrast map, and They represent the constant terms of the sample and the background in formula (7), φ represents the phase contrast image, and They represent the initial phases of the sample and the background in formula (7), V represents the dark field contrast image, and v s and v r They are the dark field contrast images of the sample and the background, and They represent the amplitude of the sample and the background in formula (7) respectively. The first image of the step projection sequence containing the sample is taken as the source image. The three contrast projections obtained by Fourier analysis of the step projection sequence are taken as the true values. The source image and the true value at the same angle are input into the neural network in pairs for training.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; Wherein, when one or more programs are executed by the one or more processors, the one or more processors implement the X-ray static fast multi-contrast signal analysis method based on a single projection as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, and when the instructions are executed by the processor, the processor can implement the X-ray static fast multi-contrast signal analysis method based on a single projection as described in any one of claims 1-8.

Images