An adaptive extended field of view radiation source positioning method

Abstract

The present invention discloses an adaptive extended-field-of-view radioactive source positioning method, including establishing a coded aperture imaging system, dividing and numbering its imaging field of view; using the Monte Carlo method to simulate so that each area obtains a sufficient amount of projection images; constructing a deep neural network, pre-processing the projection images as training samples to train and optimize the network; using the Monte Carlo method to simulate the system response matrix containing all pixels in the imaging field of view and the system response matrix containing all pixels in the complete coded field of view; during actual measurement, collecting on-site case data and inputting it into the model to identify the area where the radioactive source is located, and then using a statistical iterative algorithm to achieve image reconstruction based on the identification results. The adaptive extended-field-of-view radioactive source positioning method of the present invention can complete radiation imaging positioning with a large field of view and high resolution through one imaging, while being free from the interference of pseudo-reconstruction hotspots caused by partial coding effects, and greatly improving the positioning and search efficiency and accuracy of the radioactive source.

Description

Technical Field

[0001] The invention relates to a radiation source positioning method, in particular to an adaptive extended field of view radiation source positioning method. Background Art

[0002] Nuclear safety is related to the national economy and people's livelihood. With the commissioning of a large number of nuclear facilities in my country, rapid and accurate imaging and positioning of radioactive materials is of great significance to ensuring nuclear safety. Compared with traditional non-imaging equipment such as radiation dosimeters and spectrometers, gamma-ray imaging technology can provide spatial distribution and relative intensity information of radioactive sources at a long distance, which can effectively reduce the radiation absorption dose of relevant personnel and has obvious advantages.

[0003] Coded aperture imaging is a radiographic imaging technology based on mechanical collimation. It has the advantages of high angular resolution and high sensitivity, and has been widely used in the fields of nuclear industry, nuclear safety and security, environmental monitoring, public safety, etc. However, the imaging field of the coded aperture imaging system is very small, which not only affects the efficiency of searching for radiation hotspots over a large area, but is also likely to be affected by the partial coding effect caused by the radioactive source located in part of the coded field of view, resulting in pseudo hotspots in the reconstructed image, which in turn causes the radioactive source to be mislocated, and even causes physical damage to the relevant personnel due to excessive radiation exposure in a short period of time.

[0004] Explanation of imaging field of view, full encoding field of view and partial encoding field of view:

[0005] The imaging field of view of the existing coded aperture imaging system generally refers only to the complete coded field of view, without considering the partial coded field of view. However, in practical applications, the radiation source may be located in the partial coded field of view and interfere with the accuracy of the reconstruction result. In order to obtain accurate results, it is usually necessary to image the imaging system multiple times at different azimuth angles and then perform manual judgment, which is not only time-consuming and labor-intensive, but also difficult to ensure reliability.

[0006] Fully-coded Field-of-View, abbreviated as FCFOV, means that the photons emitted by the radiation source and detected in this area have been modulated by the collimator, so they can form a complete projection pattern on the detector array.

[0007] Partially-coded Field-of-View, abbreviated as PCFOV. Only a part of the photons emitted by the radiation source and detected in this area are modulated by the collimator, while the other part directly interacts with the detector, thus forming only a partial projection pattern on the detector array.

[0008] The imaging field of view described in the present invention is composed of a complete coding field of view and a partial coding field of view. In the entire imaging field of view, the partial coding field of view is an annular area obtained by subtracting the complete coding field of view from the imaging field of view. Summary of the invention

[0009] The purpose of the present invention is to provide a method for positioning a radiation source with an extended field of view that solves the above-mentioned problem, avoids the mispositioning of the radiation source due to pseudo hot spots generated by partial coding effects, and can quickly and accurately reconstruct a wide-field-of-view, high-resolution radiation source spatial distribution image.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is as follows: an adaptive extended field of view radiation source positioning method, comprising the following steps;

[0011] (1) establishing a coded aperture imaging system and determining its imaging field of view, wherein the imaging field of view is composed of a complete coded field of view located at the center thereof and a partial coded field of view located outside the complete coded field of view, wherein the complete coded field of view is taken as the central area, and the partial coded field of view is divided into 8 partial areas along the boundaries of the central area, and the 9 areas are numbered respectively;

[0012] (2) The Monte Carlo method is used to simulate the imaging process of the coded aperture imaging system on the radiation source located in the nine regions, so that each region can obtain a sufficient number of projection images;

[0013] (3) constructing a deep neural network, using the preprocessed projection image as a training sample, using the region number where the radiation source corresponding to the projection image is located as a label, training and optimizing the neural network model, and obtaining a deep neural network model;

[0014] (4) The Monte Carlo method is used to simulate the system response matrix A containing all pixels in the imaging field of view, and the system response matrix A containing all pixels in the complete encoding field of view is obtained. F ;

[0015] (5) During actual measurement, a coded aperture imaging system is used to collect event data in the radiation scene to form a projection image, which is then input into a deep neural network model after preprocessing to identify the area where the radiation source is located;

[0016] (6) Image reconstruction is achieved using a statistical iterative reconstruction algorithm;

[0017] There are one or more radiation sources. When the radiation sources are all located in the central area, the reconstruction process uses the system response matrix A F ; When a radioactive source is identified as being located in a partial area, the reconstruction process uses the system response matrix A.

[0018] Preferably, the coded aperture imaging system comprises a coded collimator and a detector array, wherein the detector array is a detector distributed in a matrix.

[0019] Preferably, the encoding modes of the encoding collimator include random array, non-redundant array, uniform redundant array, modified uniform redundant array, and self-supporting array.

[0020] Preferably, in step (1), the nine regions are numbered specifically as follows: R1-R9 in order from top to bottom and from left to right in the imaging field of view, wherein the mark corresponding to the central region is R5.

[0021] Preferably, in step (2), the imaging process of simulating the radiation source is to simulate the radiation sources of different energies, activities, quantities, shapes, positions and the conditions of adding noises of different levels and distributions.

[0022] Preferably, the structure of the deep neural network adopts one or more of a fully connected neural network and a convolutional neural network.

[0023] Preferably, the preprocessing is one or more of normalization, standardization, filling, dimensionality reduction, and principal component analysis.

[0024] Preferably, the label in step (3) is encoded using one-hot encoding.

[0025] Preferably, the statistical iterative reconstruction algorithm includes a maximum likelihood expectation maximization algorithm and a compressed sensing algorithm.

[0026] As a preferred method: obtain the system response matrix A containing all pixels in the complete coded field of view F Specifically, it can be extracted from the system response matrix A, or simulated separately using the Monte Carlo method, the system response matrix A and the system response matrix A F The pixel division sizes are the same or different.

[0027] Compared with the prior art, the advantages of the present invention are:

[0028] The partial coded field of view is divided into several partial areas and a central area to form 9 areas. The Monte Carlo method is used to simulate the imaging process of the coded aperture imaging system for the radioactive source in each area, so as to obtain projection images with different characteristics corresponding to the 9 areas, which are input into the deep neural network model for training and optimization. The radioactive sources located in different partial areas will form special patterns related to the area on the detector array. Therefore, this area division method is conducive to accelerating the convergence process of deep neural network model training, improving the model's recognition accuracy of the radioactive source located in the partial coded field of view, and accurately identifying the location of the radioactive source in subsequent actual measurements.

[0029] The present invention combines the deep learning method with the statistical iterative reconstruction algorithm to expand the field of view of the traditional coded aperture imaging system that can accurately image from the original complete coded field of view to the partial coded field of view. The present invention can accurately identify whether there is a radiation source in the partial coded field of view, and realize wide field of view and high resolution imaging of the adaptively expanded field of view according to the identification result, which can effectively avoid the pseudo hotspots in the reconstructed image caused by the partial coding effect. The method is highly portable and suitable for devices with different geometric structures and different materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is the overall flow chart of the present invention;

[0031] Figure 2a A schematic diagram of a method for dividing the field of view of a coded aperture imaging system;

[0032] Figure 2b for Figure 2a Main view of the middle imaging field of view;

[0033] Figure 3 This is a flowchart of image reconstruction using the maximum likelihood expectation maximization algorithm during actual measurement;

[0034] Figure 4a Schematic diagram of the position of the radiation source when the radiation source exists in only part of the coded field of view;

[0035] Figure 4b For existing technology Figure 4a The image reconstructed by cross-correlation method under the condition;

[0036] Figure 4c For existing technology Figure 4a The image reconstructed using the traditional maximum likelihood expectation maximization algorithm under the condition;

[0037] Figure 4d For the present invention Figure 4a The image reconstructed under the condition;

[0038] Figure 5a It is a schematic diagram of the position of the radiation source under the condition that the radiation source exists in the complete coded field of view and the partial coded field of view at the same time;

[0039] Figure 5b For existing technology Figure 5a The image reconstructed by cross-correlation method under the condition;

[0040] Figure 5c For existing technology Figure 5a The image reconstructed using the traditional maximum likelihood expectation maximization algorithm under the condition;

[0041] Figure 5d For the present invention Figure 5a The reconstructed image under the condition.

[0042] In the figure: 1. Encoded collimator; 2. Detector array; 3. Imaging field of view. DETAILED DESCRIPTION

[0043] The present invention will be further described below in conjunction with the accompanying drawings.

[0044] Example 1: See Figure 1-2b ,An adaptive extended field of view radiation source positioning method, comprising the following steps;

[0045] (1) Establishing a coded aperture imaging system, determining its imaging field of view 3, wherein the imaging field of view 3 is composed of a complete coded field of view located at the center thereof and a partial coded field of view located outside the complete coded field of view, wherein the complete coded field of view is taken as the central area, and the partial coded field of view is divided into 8 partial areas along the boundary of the central area, and the 9 areas are numbered respectively; in this embodiment, the 9 areas are numbered respectively as follows: in the imaging field of view 3, they are numbered R1 to R9 in order from top to bottom and from left to right, wherein the mark corresponding to the central area is R5;

[0046] (2) using the Monte Carlo method to simulate the imaging process of the coded aperture imaging system on the radioactive sources located in the nine regions, so that each region obtains a sufficient number of projection images; in this embodiment, the imaging process of the simulated radioactive source is to simulate the conditions of radioactive sources of different energies, activities, quantities, shapes, and positions, and adding noises of different levels and distributions;

[0047] (3) constructing a deep neural network, using the preprocessed projection image as a training sample, using the region number where the radiation source corresponding to the projection image is located as a label, training and optimizing the neural network model, and obtaining a deep neural network model;

[0048] (4) The Monte Carlo method is used to simulate the system response matrix A containing all pixels in the imaging field of view 3, and the system response matrix A containing all pixels in the complete encoding field of view is obtained. F ;

[0049] (5) During actual measurement, a coded aperture imaging system is used to collect event data in the radiation scene to form a projection image, which is then input into a deep neural network model after preprocessing to identify the area where the radiation source is located;

[0050] (6) Image reconstruction is achieved using a statistical iterative reconstruction algorithm;

[0051] There are one or more radiation sources. When the radiation sources are all located in the central area, the reconstruction process uses the system response matrix A F; When a radioactive source is identified as being located in a partial area, the reconstruction process uses the system response matrix A.

[0052] In this embodiment: the coded aperture imaging system comprises a coded collimator 1 and a detector array 2, wherein the detector array 2 is a detector distributed in a matrix.

[0053] The coding methods of the coding collimator 1 include random array, non-redundant array, uniform redundant array, modified uniform redundant array, and self-supporting array. The structure of the deep neural network adopts one or more of a fully connected neural network and a convolutional neural network.

[0054] The preprocessing is: one or more of normalization, standardization, filling, dimension reduction, and principal component analysis.

[0055] The labels in step (3) are encoded using one-hot encoding.

[0056] The statistical iterative reconstruction algorithm includes a maximum likelihood expectation maximization algorithm and a compressed sensing algorithm.

[0057] Get the system response matrix A containing all pixels in the complete encoding field of view F Specifically, it can be extracted from the system response matrix A, or simulated separately using the Monte Carlo method, the system response matrix A and the system response matrix A F The pixel division sizes are the same or different.

[0058] Example 2: See Figure 1-Figure 5d , based on Example 1, we further define.

[0059] Wherein, in step (1), when establishing a coded aperture imaging system, the detector array 2 of the coded aperture imaging system is a 19×19 BGO crystal strip, and the distance between the coded collimator 1 and the detector array 2 of the coded aperture imaging system is set to 59 mm. The numbering of the central area and the partial area in the imaging field of view 3 is shown in Figure 2a and Figure 2b , R1-R9 in turn, with the central area being R5. The FCFOV size corresponding to the R5 area is 40°×40°, and the overall imaging field of view 3 composed of FCFOV and PCFOV is 72°×72°.

[0060] In actual measurement, step (5) is first performed to collect case data in the radiation scene using a coded aperture imaging system to form a projection image, and the projection image is input into a deep neural network model to perform regional identification of the radiation source;

[0061] Then, in step (6), the image is reconstructed by using a statistical iterative reconstruction algorithm. The reconstruction method includes but is not limited to the maximum likelihood expectation maximization algorithm and the compressed sensing algorithm. However, in order to better illustrate the method of the present invention, the maximum likelihood expectation maximization algorithm is used in this embodiment. For details, see Figure 3 :

[0062] When a radioactive source is identified in part of the coded field of view, the system response matrix A containing all pixels is imported; when a radioactive source is not identified in part of the coded field of view, the system response matrix A containing only pixels of the complete coded field of view is extracted from the system response matrix A, or the system response matrix A containing only pixels of the complete coded field of view is simulated and imported separately using the Monte Carlo method. F , and the maximum likelihood expectation maximization algorithm is used to achieve image reconstruction, where the iterative formula is as follows:

[0063]

[0064] Where I is the total number of detectors in the detector array 2, and J is A or A F The total number of pixels in the corresponding image space;

[0065] k is the number of iterations, and are the reconstructed images after the kth and k+1th iterations respectively;

[0066] a ij is the basic element of the imported system response matrix, representing the probability that a photon is emitted from the jth pixel and detected by the i-th detector, p i is the count value of the i-th detector pixel, and i=1~I, j=1~J.

[0067] Verification: In order to verify the effectiveness of the method, experimental tests were conducted on radiation scenes with radiation sources in both the complete coded field of view and the partial coded field of view, and the cross-correlation method, the traditional maximum likelihood expectation maximization algorithm and the method of the present invention were compared.

[0068] See also Figure 4a , activity of 1mCi 137 The Cs point source is 2 meters away from the coded aperture imaging system and is located about 1.3 meters to the left of the center of the imaging field of view 3. The present invention uses the cross-correlation method, the traditional maximum likelihood expectation maximization algorithm, and the image reconstructed by the method of the present invention. Figure 4b-4d As can be seen from the figure, the images reconstructed by the cross-correlation method and the traditional maximum likelihood expectation maximization algorithm both have pseudo hot spots on the right side of the center, while the method of the present invention can accurately reconstruct the radiation source on the left side located in the partial encoding field of view.

[0069] See also Figure 5a , by converting the above activity of 1mCi 137The point source of Cs is measured at the center of the imaging field of view 3 and 1.2 meters to the left of the center of the imaging field of view 3, and the projection data collected twice are added to form a radiation scene with a complete coded field of view and a partial coded field of view with a radiation source at the same time. The images reconstructed by the cross-correlation method, the traditional maximum likelihood expectation maximization algorithm, and the method of the present invention are shown in FIG. Figure 5b-5d As can be seen from the figure, the images reconstructed using the cross-correlation method and the traditional maximum likelihood expectation maximization algorithm can only accurately reconstruct the radiation hotspots of the complete and partial coded fields of view, but the radiation source located in the partial coded field of view is incorrectly reconstructed on the right side of the image. The method of the present invention can accurately reconstruct the radiation sources of the complete and partial coded fields of view at the same time.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An adaptive extended field of view radiation source positioning method, characterized in that: The steps include: (1) establishing a coded aperture imaging system and determining its imaging field of view, wherein the imaging field of view is composed of a complete coded field of view located at the center thereof and a partial coded field of view located outside the complete coded field of view, wherein the complete coded field of view is taken as the central area, and the partial coded field of view is divided into 8 partial areas along the boundaries of the central area, and the 9 areas are numbered respectively; (2) The Monte Carlo method is used to simulate the imaging process of the coded aperture imaging system on the radiation source located in the nine regions, so that each region can obtain a sufficient number of projection images; (3) constructing a deep neural network, using the preprocessed projection image as a training sample, using the region number where the radiation source corresponding to the projection image is located as a label, training and optimizing the neural network model, and obtaining a deep neural network model; (4) The Monte Carlo method is used to simulate the system response matrix A containing all pixels in the imaging field of view, and the system response matrix A containing all pixels in the complete encoding field of view is obtained. F ; (5) During actual measurement, a coded aperture imaging system is used to collect event data in the radiation scene to form a projection image, which is then input into a deep neural network model after preprocessing to identify the area where the radiation source is located; (6) Image reconstruction is achieved using a statistical iterative reconstruction algorithm; There are one or more radiation sources. When the radiation sources are all located in the central area, the reconstruction process uses the system response matrix A F ; When a radioactive source is identified as being located in a partial area, the reconstruction process uses the system response matrix A.

2. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: The coded aperture imaging system comprises a coded collimator and a detector array, wherein the detector array is a detector distributed in a matrix.

3. The adaptive extended field of view radiation source positioning method according to claim 2, characterized in that: The encoding modes of the encoding collimator include random array, non-redundant array, uniform redundant array, modified uniform redundant array, and self-supporting array.

4. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: In step (1), the nine regions are numbered specifically as follows: in the imaging field of view, they are numbered R1-R9 in order from top to bottom and from left to right, wherein the mark corresponding to the central region is R5.

5. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: In step (2), the imaging process of simulating the radiation source is to simulate the conditions of radiation sources of different energies, activities, quantities, shapes, and positions, and adding noises of different levels and distributions.

6. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: The structure of the deep neural network adopts one or more of a fully connected neural network and a convolutional neural network.

7. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: The preprocessing is: one or more of normalization, standardization, filling, dimension reduction, and principal component analysis.

8. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: The labels in step (3) are encoded using one-hot encoding.

9. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: The statistical iterative reconstruction algorithm includes a maximum likelihood expectation maximization algorithm and a compressed sensing algorithm.

10. The adaptive extended field of view radiation source positioning method according to claim 1, characterized in that: Get the system response matrix A containing all pixels in the complete encoding field of view F Specifically, it can be extracted from the system response matrix A, or simulated separately using the Monte Carlo method, the system response matrix A and the system response matrix A F The pixel division sizes are the same or different.

Images