X-ray CT local imaging method, device and electronic equipment for expanding imaging field of view

Abstract

The present invention discloses an X-ray CT local imaging method, device and electronic device for expanding the imaging field of view, belonging to the field of radiation imaging. In the case of only local imaging scanning data, the present invention can effectively decouple the influence of multiple data pathologies, and combine the scanning data conditions of the inner and outer regions of the region of interest, and gradually expand the reconstructible area through multi-stage deep learning technology, so as to achieve high-quality reconstruction of the inner region of the region of interest, and effectively restore the structure of the outer region of the region of interest, thereby improving the image quality of X-ray CT local imaging. The present invention can be flexibly applied to different CT scanning modes and system architectures, and can be used in the fields of medical diagnosis, industrial non-destructive testing and security inspection.

Description

Technical Field

[0001] The present invention relates to the field of radiation imaging technology, and in particular to an X-ray CT local imaging method, device and electronic equipment for expanding the imaging field of view. Background Art

[0002] X-ray CT imaging systems are widely used in the fields of medical treatment, security inspection, industrial non-destructive testing, etc. According to CT reconstruction theory, the distribution of physical characteristic parameters of a section of an object can be calculated by the line integral value of the physical characteristic parameters in all directions. Therefore, in the existing CT scanning system, the ray source and the detector usually collect projections of the object in all directions according to a certain orbit, and the ray source orbit needs to cover enough scanning angles, and the detector can completely receive all rays passing through the object at each viewing angle. However, in some special application scenarios, in order to reduce the radiation dose or system hardware cost, the detector can only completely receive the rays covering the local region of interest (ROI) of the object. In local imaging, the reconstruction of the internal area of ​​the ROI has the problem of projection data truncation, while the reconstruction of the external area of ​​the ROI has the problem of serious insufficient projection angle sampling. If the traditional analytical reconstruction method is used, serious numerical deviations and artifacts (such as cupping artifacts, etc.) will be generated inside the ROI, and due to the serious lack of projection angle outside the ROI, it is difficult to effectively restore the structural information outside the ROI. At present, local ROI image reconstruction mainly relies on iterative reconstruction algorithms. Traditional iterative algorithms have problems such as unstable optimization. By introducing prior information about the reconstruction area (such as sparse structure, piecewise linearity, or known partial structural information, etc.) and pre-estimating the area outside the ROI, the reconstruction of the ROI area can be achieved more stably. However, this type of method has a large amount of computation and is difficult to restore high-frequency information inside the ROI, especially when the projection noise is large, and it is impossible to effectively restore the structural information of the area outside the ROI.

[0003] In recent years, with the development of deep learning technology, image reconstruction methods that extract prior information from large-scale data sets through neural networks have been widely used. In the field of ROI reconstruction, there are three main types of deep learning-related reconstruction methods: ① using neural networks to complete truncated projection data; ② using neural networks to remove artifacts of reconstructed images; ③ using neural network reconstruction results as prior information and combining them with traditional iterative reconstruction methods to obtain the final reconstruction results. Although the above methods can effectively remove artifacts caused by truncation of projection data, they mainly achieve reconstruction of the internal area of ​​ROI through projection extrapolation, that is, based on the existing local ROI projection extrapolation to estimate the missing projection outside the field of view. The projection data collected by the actual system is often accompanied by other non-ideal factors, such as low-dose scanning and sparse viewing angle sampling. The resulting projection noise, sparse sampling and other problems exist in the existing local ROI projection. It is difficult for existing methods to simultaneously deal with artifacts caused by such coupled pathological problems. In addition, in practical applications, the effective recovery of the structure of the external area of ​​ROI is of great significance in the positioning of the inspected object during scanning, the location identification of lesions or abnormal structures, and the expansion of the inspectable area. Existing reconstruction methods often have difficulty in achieving effective reconstruction and structural restoration of the area outside the ROI, and are unable to effectively evaluate the credibility of the restoration results of the area outside the ROI. Therefore, an improved imaging system and method are needed. Summary of the invention

[0004] The present invention provides an X-ray CT local imaging method and device for expanding the imaging field of view, so as to solve the problems that the existing image reconstruction method cannot effectively solve the ROI reconstruction problem of multiple artifact coupling and cannot effectively restore the ROI external area structure.

[0005] A first aspect of the present invention provides an X-ray CT local imaging method for expanding the imaging field of view, comprising the following steps:

[0006] Acquire local imaging projection data including the local area of ​​interest in the local area scan, and calculate the angle sampling coverage of each position of the scanned object according to the light source scanning trajectory of the local area scan, and obtain the spatial distribution of the local imaging angle sampling;

[0007] Decoupling the influence of multiple data pathologies in the local imaging projection data to obtain local imaging projection data with only data truncation, decoupling the projection features of the local region of interest and the external region, obtaining projection data containing only the local region of interest based on the decoupled local imaging projection data, and reconstructing using the decoupled projection data containing only the local region of interest to obtain a reconstruction result inside the local region of interest;

[0008] The angle sampling threshold and loss function weight are determined, and the reconstruction result with an expanded imaging field of view is obtained through multi-stage neural network reconstruction using only local imaging projection data with data truncation.

[0009] Optionally, in one embodiment of the present invention, the local imaging projection data p of the local region of interest IN (λ,u,v) is:

[0010]

[0011] Among them, λ∈[λ min ,λ max ] is the polar coordinate corresponding to the light source trajectory, λ min and λ max are the two endpoints of the light source trajectory, (u, v) represents the two-dimensional coordinates of the detector pixel, and p FULL (λ, u, v) is the complete projection data when the detector can completely receive all the rays from each light source position passing through the scanned object, M λ is the binary angle sampling mask, n p is the projection noise, u m With v m Indicates the maximum value of the row and column index of the detector unit, and is an indicative function used to limit the range of rays received by the detector.

[0012] Optionally, in one embodiment of the present invention, calculating the angle sampling coverage of each position of the scanned object according to the light source scanning trajectory of the local area scanning includes:

[0013] For any point in space Angle sampling range It is expressed as:

[0014]

[0015] in, is the light source trajectory, represents the unit vector parallel to the y-axis, and To pass through The two-dimensional coordinates of the projection point of the ray to the detector:

[0016]

[0017] Wherein, the detector pixel is represented by the two-dimensional coordinate (u, v), and are unit vectors along the u and v directions, is the unit vector in the direction of the perpendicular line from the light source to the detector plane, and D is the length of the perpendicular line.

[0018] Optionally, in one embodiment of the present invention, the angle sampling threshold and the loss function weight are determined, and the reconstruction result of the extended imaging field of view is obtained by multi-stage neural network reconstruction using only local imaging projection data with data truncation, including:

[0019] Use N stages to restore the imaging field of view and set the threshold θ of the angle sampling coverage in the i-th stage i , and according to any point in space Angle sampling range Identify key restoration areas

[0020]

[0021] Design loss function weights So as to satisfy: and exist The boundary area has a smooth transition;

[0022] Combining the outputs of the neural networks at each stage, the final reconstruction result of the extended imaging field of view is obtained:

[0023]

[0024] in, represents the neural network of the i-th stage, and the parameters are recorded as Reconstruction results according to each stage Design a supervised or unsupervised loss function Loss(·) and As a weighted function, the objective function of neural network parameter optimization at each stage is constructed:

[0025]

[0026] Optionally, in one embodiment of the present invention, decoupling the influence of multiple data pathologies in the local imaging projection data includes:

[0027] Use multiple neural networks or codecs to decouple multiple artifacts in parallel or in series.

[0028] Optionally, in one embodiment of the present invention, decoupling the local region of interest and the external region projection features includes:

[0029] A neural network is used to decouple the projection features of the areas inside and outside the local region of interest, and to estimate the contribution of the projection inside the region of interest and the projection outside the region of interest to the local imaging projection.

[0030] Optionally, in one embodiment of the present invention, the two decoupling processes are completed synchronously by one neural network or step by step by multiple different sub-networks.

[0031] Optionally, in one embodiment of the present invention, the neural networks used in the decoupling process and the reconstruction process of expanding the imaging field of view both meet the following conditions: having a sufficient receptive field and having the ability to encode spatial positions.

[0032] A second aspect of the present invention provides an X-ray CT local imaging device for expanding the imaging field of view, comprising:

[0033] An acquisition module is used to obtain local imaging projection data including a local area of ​​interest in a local area scan, and calculate the angle sampling coverage of each position of the scanned object according to the light source scanning trajectory of the local area scan, so as to obtain the spatial distribution of the local imaging angle sampling;

[0034] A decoupling module is used to decouple the influence of various data pathologies in the local imaging projection data, obtain local imaging projection data with only data truncation, decouple the projection features of the local region of interest and the external region, obtain projection data containing only the local region of interest based on the decoupled local imaging projection data, and reconstruct using the decoupled projection data containing only the local region of interest to obtain a reconstruction result inside the local region of interest;

[0035] The reconstruction module is used to determine the angle sampling threshold and the loss function weight, and utilizes only the local imaging projection data with data truncation to obtain the reconstruction result with expanded imaging field of view through multi-stage neural network reconstruction.

[0036] A third aspect of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the X-ray CT local imaging method for expanding the imaging field of view as described in the above embodiment.

[0037] The X-ray CT local imaging method, device and electronic device for expanding the imaging field of view according to the embodiments of the present invention have the following beneficial effects:

[0038] 1. The present invention can effectively handle the ROI reconstruction problem under various data pathological coupling conditions, making the scanning method of local imaging more flexible, and helping to combine local imaging with sparse sampling, low-dose scanning and other methods to further reduce the radiation dose.

[0039] 2. The present invention constructs a new approach to solving ROI reconstruction by decoupling the projection features of the internal and external regions of the ROI, which can effectively reduce the impact of data truncation on the reconstruction of the internal region of the ROI and help to achieve structural recovery of the external region of the ROI.

[0040] 3. The present invention constructs a method for analyzing the spatial distribution characteristics of local imaging scan sampling, and based on this, realizes the structural recovery of the ROI external area guided by data conditions, expands the imaging field of local imaging, and improves the efficiency of ray utilization.

[0041] 4. The present invention is applicable to a variety of scanning trajectories and can be used to solve a variety of data truncation problems, including the lateral data truncation problem in two-dimensional scanning, the longitudinal data truncation problem in three-dimensional scanning, etc.

[0042] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0044] Figure 1 A schematic flow chart of an X-ray CT local imaging method for expanding the imaging field of view provided by an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of scanning geometry for CT local imaging according to an embodiment of the present invention;

[0046] Figure 3 A technical circuit diagram of an X-ray CT local imaging method for expanding the imaging field of view provided by an embodiment of the present invention;

[0047] Figure 4 (a) is a schematic diagram of local imaging geometry of an embodiment of the present invention;

[0048] Figure 4 (b) is a schematic diagram of the spatial distribution of local imaging angle sampling according to an embodiment of the present invention;

[0049] Figure 5 Schematic diagram of a decoupled neural network according to an embodiment of the present invention, wherein: Figure 5 (a) is the “single encoder + multiple decoders” decoupling; Figure 5 (b) is series decoupling; Figure 5 (c) is parallel decoupling; Figure 5 (d) is hybrid decoupling;

[0050] Figure 6 A schematic diagram of a framework of a method for expanding a reconstructable field of view using a multi-stage neural network according to an embodiment of the present invention;

[0051] Figure 7An exemplary diagram of an X-ray CT local imaging device for expanding the imaging field of view according to an embodiment of the present invention;

[0052] Figure 8 A schematic diagram of the structure of an electronic device provided by an embodiment of the invention. DETAILED DESCRIPTION

[0053] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0054] It is understandable that in local imaging, the projection of the outer area of ​​ROI is truncated, and the entire projection of the inner area of ​​ROI and the partial projection of the outer area together constitute the projection data obtained by scanning, and are usually accompanied by other non-ideal data acquisition factors. Existing image reconstruction methods mainly focus on completing the truncated projection data of the external area and reducing its impact on the reconstruction of the inner area of ​​ROI. Therefore, it is impossible to solve the local imaging problem of multiple pathological data conditions coupled at the same time, and it is difficult to achieve high-quality recovery of the outer area of ​​ROI. In view of the above problems, the present invention proposes an X-ray CT local imaging method that expands the imaging field of view.

[0055] Specifically, Figure 1 The present invention provides a flowchart of an X-ray CT local imaging method for expanding the imaging field of view provided by an embodiment of the present invention.

[0056] like Figure 1 As shown, the X-ray CT local imaging method for expanding the imaging field of view comprises the following steps:

[0057] In step S101, local imaging projection data including a local area of ​​interest in a local area scan is obtained, and the angle sampling coverage of each position of the scanned object is calculated according to the light source scanning trajectory of the local area scan to obtain the spatial distribution of the local imaging angle sampling.

[0058] like Figure 2 As shown, the light source trajectory can be obtained by Represents, where λ∈[λ min ,λ max ] is the polar coordinate corresponding to the light source trajectory, λ min and λ max are the two endpoints of the light source trajectory. The detector pixel can be represented by a two-dimensional coordinate (u, v). The unit vector along the u and v directions is and The unit vector in the direction of the perpendicular line from the light source to the detector plane is The length of the vertical line is D. Any point of the scanned object The linear attenuation coefficient at The boundary of the object is The ROI boundary is If the detector can completely receive all rays from each light source position passing through the scanned object, a complete projection can be obtained, which is denoted as p FULL (λ,u,v):

[0059]

[0060] in, represents the unit vector along the direction of the ray, and t represents the distance between a point on the ray and the ray source:

[0061]

[0062] In local imaging, the detector size or the light source emission range is limited, and only the ROI area can be completely covered by the rays. In actual systems, projection is often accompanied by other non-ideal factors (such as noise n p Or the sampling angle is incomplete). Therefore, the local imaging projection data p of the local area of ​​interest IN (λ, u, v) can be expressed as:

[0063]

[0064] Among them, M λ is the binary angle sampling mask, which takes the value of 1 for the angle covered by the sampling, otherwise it is 0, n p is the projection noise, u m With v m Indicates the maximum value of the detector unit index, which is determined by the detector size. and is an indicative function used to limit the range of rays received by the detector. Is the indicative function:

[0065]

[0066] To analyze the sampling characteristics of local imaging, define any point in space Angle sampling range It can be expressed as:

[0067]

[0068] in, represents the unit vector parallel to the y-axis, and To pass through The two-dimensional coordinates of the projection point of the ray to the detector:

[0069]

[0070] For any scanning trajectory, the angle sampling coverage of each position of the scanned object can be calculated by formula (5), thereby obtaining the spatial distribution of local imaging angle sampling, which is used to guide the truncated projection recovery and imaging field of view expansion in subsequent steps.

[0071] In step S102, the influence of various data pathologies in the local imaging projection data is decoupled to obtain local imaging projection data with only data truncation, the projection features of the local region of interest and the external region are decoupled, and projection data containing only the local region of interest is obtained based on the decoupled local imaging projection data. Reconstruction is performed using the decoupled projection data containing only the local region of interest to obtain a reconstruction result inside the local region of interest.

[0072] Specifically, the influence of various data pathological factors (such as noise, sparse angle, data truncation, etc.) in the local imaging projection data is decoupled to obtain local imaging projection data with only data truncation factors (no influence of other factors or the influence of other factors can be ignored), and then the projection features of the internal and external areas of the ROI are decoupled to obtain the projection data containing only the internal area of ​​the ROI for reconstruction, and the reconstruction result of the internal area of ​​the ROI is obtained.

[0073] In an embodiment of the present invention, decoupling the local region of interest and the external region projection features includes:

[0074] A neural network is used to decouple the projection features of the areas inside and outside the local region of interest, and to estimate the contribution of the projection inside the region of interest and the projection outside the region of interest to the local imaging projection.

[0075] In the embodiment of the present invention, if the projection data collected by the detector only passes through the inner area of ​​ROI (denoted as p ROI (λ, u, v)), high-quality ROI reconstruction can be achieved through traditional reconstruction methods.

[0076]

[0077] However, p cannot be directly obtained in local imaging. ROI (λ, u, v) data, we need to decouple the following two sets of coupling relationships in order to obtain IN (λ, u, v) to obtain p ROI (λ, u, v), achieving high-quality reconstruction.

[0078] First, decouple the effects of various data pathologies. In addition to data truncation, the projection data p of local imaging IN(λ, u, v) usually also includes the influence of various data pathologies (such as formula (3)). In order to separate the influence of different data pathologies, multiple neural networks (for example, commonly used U-Net, Transformer and other architectures) or codecs can be introduced to achieve parallel or (and) series decoupling of various artifacts, gradually solving different types of pathological problems and obtaining projection data with only data truncation.

[0079]

[0080] Secondly, the projection features of the internal and external areas of the ROI are decoupled. In local imaging, the ray passes through the internal and external areas of the ROI at the same time, and the collected projection data also includes the projection p of the internal area of ​​the ROI. ROI The “background” projection of the area outside the ROI (with p BG (λ, u, v) means):

[0081]

[0082] It is worth noting that in In , there is no data truncation in pROI, but there is data truncation in pBG. Therefore, a neural network for background projection estimation can be introduced to extract The global projection features of ROI are decoupled, and the projection features of the internal and external regions of ROI are separated. There is no data truncation component in the ROI, and the projection p containing only the internal area of ​​the ROI is obtained. ROI (λ, u, v). The background projection is the part of the ROI external area projection that falls within the local imaging projection area.

[0083] Since the above two sets of coupling relationships exist simultaneously in the local scan projection, the decoupling process can be completed synchronously by a neural network or step by step by multiple different sub-networks. ROI Afterwards, high-quality reconstruction of the internal area of ​​ROI can be achieved through traditional analytical reconstruction methods or iterative reconstruction methods.

[0084] In step S103, the angle sampling threshold and the loss function weight are determined, and the local imaging projection data with only data truncation is used to obtain a reconstruction result with an expanded imaging field of view through multi-stage neural network reconstruction.

[0085] Through the sampling feature analysis in step S101, the angular sampling range of each region of the scanned object in the local imaging can be obtained. Generally speaking, the scanning protocol of local imaging will ensure that the sampling angles of the internal area of ​​ROI are complete, while the sampling angles of the external area of ​​ROI are missing to varying degrees. If a neural network is introduced to directly restore the external area of ​​ROI, it will be difficult for the neural network to adapt to the differences in global sampling conditions and it will be difficult to achieve high-quality image restoration. Therefore, a multi-stage neural network can be introduced to achieve the gradual expansion of the imaging field of view and structural restoration. Assuming that N stages are used to restore the imaging field of view, the parameters and weights of the neural network used in each stage can be shared or different. For the i-th stage, set the angle sampling threshold θ i , and combined with Determine the areas that need to be restored during this phase

[0086]

[0087] To guide neural network training, pay more attention to The restoration of the area can design appropriate loss function weights At the same time: ① ② exist When defining the network loss function, g i Combined with the traditional loss function, the focus area of ​​this stage is restored. Finally, the output of the neural network in all stages is combined to obtain the final reconstruction result of the extended imaging field of view:

[0088]

[0089] in, represents the neural network of the i-th stage, and the parameters are recorded as Reconstruction results according to each stage Design a supervised or unsupervised loss function Loss(·) and As a weighted function, the objective function of neural network parameter optimization at each stage is constructed:

[0090]

[0091] Generally speaking, in order to make the network performance more stable, in a multi-stage neural network, the early stage gives priority to restoring the area with better data conditions, while the later stage restores the part with poorer data conditions, that is, θ i+1 ≤θ i (i=1,...,N-1).

[0092] In one embodiment of the present invention, a neural network is introduced in the decoupling process and the reconstruction process of expanding the imaging field of view to achieve composite artifact decoupling and imaging field of view expansion. Common neural network structures such as U-shaped networks and Transformers can be used. Since the "background" projection contribution p of the ROI external area in local imaging is BG With p ROI The data conditions sampled in the entire imaging field of view and outside the ROI vary with position. To achieve optimal performance, the neural network needs to meet two conditions: ① sufficient receptive field; ② the ability to encode spatial position. Therefore, the receptive field can be expanded by combining dilated convolution and frequency domain learning, as well as position encoding technology, to integrate imaging geometry into the operation of the neural network, so that the performance can be optimally designed.

[0093] The X-ray CT local imaging method for expanding the imaging field of view of the present invention innovatively designs a reconstruction framework based on the decoupling concept, which can efficiently realize the decoupling of projection data of the internal and external areas of the ROI, as well as the decoupling of various artifact features, and can flexibly and efficiently solve the influence of various coupling factors and the internal and external structural features of the ROI in actual scanning. Combined with the sampling feature analysis of local imaging, a multi-stage deep learning network is innovatively introduced to achieve progressive recovery of the external area of ​​the ROI, effectively expanding the imaging field of view, and can be flexibly applied to different CT scanning modes and system architectures, and can be used in medical diagnosis, industrial non-destructive testing and security inspection fields.

[0094] The following uses two-dimensional circular orbit fan beam scanning as an example to introduce the implementation details of the present invention.

[0095] like Figure 3 As shown, the technical route of the X-ray CT local imaging method for expanding the imaging field of view is demonstrated.

[0096] First, the sampling characteristics of local imaging are analyzed and described: In some embodiments, the ray source and the detector move along a circular orbit around the scanned object (such as Figure 4 (a)), the orbit radius is D S , the light source trajectory can be obtained by The detector direction can be expressed by Indicates that the normal direction of the detector The distance from the ray source to the detector is represented by D. The radius of the scanned object is represented by L, and the region of interest is represented by the area with a central radius of R. In order to make the detector completely cover the scanned object, the detector half length Any point in space Angle sampling range It can be expressed as:

[0097]

[0098] Pass through The one-dimensional coordinate of the projection point of the ray to the detector. If we consider λ min = 0 and λ max =2π, that is, the ray source rotates around the scanned object once, and the spatial distribution of the sampling angle shown in formula (13) is as follows: Figure 4 As shown in (b), the gray value of each point in the figure represents the size of the sampling angle. From the figure, we can see that the sampling angle coverage of the inner area of ​​ROI is greater than 180° (the central white circular area), while the sampling angle coverage of the outer area of ​​ROI is less than 180°. Since the sampling trajectory is circular, the spatial distribution of the sampling angle is circularly symmetrical, and the farther away from the center of the circle along the radial direction, the more serious the sampling angle loss.

[0099] Secondly, the ROI reconstruction method based on the decoupling idea: In the embodiment of the present invention, local imaging and sparse viewing angle sampling (the sampling viewing angle mask is M λ ), low-dose scanning (projection noise is n p ) can further reduce the radiation dose. In local imaging, the rays pass through the inner and outer areas of the ROI at the same time, and the collected projection data also includes the projection data of the inner area of ​​the ROI. ROI The “background” projection p of the area outside the ROI BG Therefore, an appropriate decoupling strategy can be used to separate the influence of various artifacts and decouple the projection features of different regions.

[0100] For example, a "single encoder + multiple decoders" approach can be used (such as Figure 5 (a)) encodes the features of local imaging projection data through a shared encoder, and then decodes different artifacts or projection features through different decoders, and finally obtains p ROI ,The reconstruction result of the internal area of ​​ROI is obtained through the traditional analytical reconstruction algorithm.

[0101] For another example, the “series decoupling” method can be used (such as Figure 5 (b)) Through multiple cascaded neural networks, different artifacts or projection features are decoupled one by one. A typical approach is to first decouple the projection noise, then complete the missing projection angles to decouple the influence of sparse angle sampling, and then decouple the "background" projection p BG Finally, we get p ROI ,The reconstruction result of the internal area of ​​ROI is obtained through the traditional iterative reconstruction algorithm.

[0102] For another example, a “parallel decoupling” method can be used (such as Figure 5 (c)), through multiple parallel neural networks, different artifacts or projection features are simultaneously decoupled, and finally p is obtained by synthesis. ROI,The reconstruction result of the internal area of ​​ROI is obtained through the traditional iterative reconstruction algorithm.

[0103] For another example, the “parallel-parallel decoupling” method can be used (such as Figure 5 (d)) through multiple parallel and serial neural networks to decouple one or more artifacts or projection features step by step. A typical approach is to first complete the missing projection angles and decouple the projection noise through parallel decoupling, and then decouple the “background” projection p BG Finally, we get p ROI ,The reconstruction result of the internal area of ​​ROI is obtained through the traditional iterative reconstruction algorithm.

[0104] Thirdly, the imaging field of view expansion method based on multi-stage neural network: For example, a three-stage method can be used to gradually expand the imaging field of view of local imaging (such as Figure 6 ). According to the image reconstruction theory, the combination of neural network can effectively restore the area with sampling angle above 90°, making it valuable for detection or diagnosis. Therefore, for the three stages, the angle sampling threshold θ is set respectively. i 150°, 120° and 90°. Key restoration areas at each stage Defined by formula (10), combined with formula (13) and the threshold of each stage, the loss function weight is defined for each stage of training. Used to guide the neural network to pay more attention to the area that meets the angle sampling threshold conditions:

[0105]

[0106] In order to enhance the generalization of the neural network, the neural networks at each stage share parameters, and the loss function used in their training is defined as the reconstruction result and truth value The L2 norm between:

[0107]

[0108] Finally, construct an image reconstruction neural network: In an embodiment of the present invention, a classic U-shaped network can be used as the basic network architecture. In order to expand the receptive field of the network, combined with the sizes of the local imaging projection data in the two dimensions of the detector and the sampling angle, a dilated convolution mechanism is introduced for the convolution layer between two adjacent down / upsampling of the U-shaped network. In this embodiment, there are 3 convolution layers between two adjacent down / upsampling of the U-shaped network, and the dilated convolution expansion rates of each convolution layer in the detector dimension are set to 1, 2, and 5, respectively, and the dilated convolution expansion rates of the sampling angle dimension are set to 1, 1, and 1, so that the overall receptive field of the network can completely cover the projection data and adapt to the needs of high-resolution detector sampling. In addition, in order to enable the neural network to make full use of the projection geometric features of the local scan, the U-shaped network will receive three-channel input: projection data, detector direction position code PE(u) and angle direction position code PE(λ):

[0109]

[0110] Among them, N λ With N u Represent the number of sampling angles and the number of detector pixels respectively.

[0111] The X-ray CT local imaging method for expanding the imaging field of view proposed in the embodiment of the present invention focuses on the image reconstruction problem of local imaging, and proposes a set of new multi-stage ROI reconstruction methods that can expand the imaging field of view based on the analysis of data sampling conditions of different regions in local scanning and combined with deep learning technology. In the case of only scanning data of the local ROI area, the influence of various data pathologies can be effectively decoupled, and combined with the scanning data conditions of the areas inside and outside the ROI, the reconstructable area can be gradually expanded through multi-stage deep learning technology to achieve high-quality reconstruction of the internal area of ​​the ROI, and effectively restore the structure of the external area of ​​the ROI, thereby improving the image quality of X-ray CT local imaging. The present invention can be applied to medical, security inspection, industrial inspection and other fields for perspective imaging inspection of objects or human bodies.

[0112] Next, an X-ray CT local imaging device for expanding the imaging field of view according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0113] Figure 7 An exemplary diagram of an X-ray CT local imaging device for expanding the imaging field of view according to an embodiment of the present invention.

[0114] like Figure 7 As shown, the X-ray CT local imaging device 10 for expanding the imaging field of view includes: an acquisition module 100 , a decoupling module 200 and a reconstruction module 300 .

[0115] Among them, the acquisition module 100 is used to obtain local imaging projection data containing a local region of interest in the local area scan, and calculate the angle sampling coverage of each position of the scanned object according to the light source scanning trajectory of the local area scan, and obtain the spatial distribution of local imaging angle sampling. The decoupling module 200 is used to decouple the influence of various data pathologies in the local imaging projection data, obtain the projection data of local imaging with only data truncation, decouple the projection features of the local region of interest and the external region, obtain the projection data containing only the local region of interest according to the decoupled local imaging projection data, and reconstruct the reconstruction result inside the local region of interest using the decoupled projection data containing only the local region of interest. The reconstruction module 300 is used to determine the angle sampling threshold and the loss function weight, and obtain the reconstruction result of the expanded imaging field of view by using the local imaging projection data with only data truncation through multi-stage neural network reconstruction.

[0116] Optionally, in one embodiment of the present invention, the local imaging projection data p of the local region of interest IN (λ,u,v) is:

[0117]

[0118] Among them, λ∈[λ min ,λ max ] is the polar coordinate corresponding to the light source trajectory, λ min and λ max are the two endpoints of the light source trajectory, (u, v) represents the two-dimensional coordinates of the detector pixel, and p FULL (λ, u, v) is the complete projection data when the detector can completely receive all the rays from each light source position passing through the scanned object, M λ is the binary angle sampling mask, n p is the projection noise, u m With v m Indicates the maximum value of the detector unit index, and is an indicative function used to limit the range of rays received by the detector.

[0119] Optionally, in one embodiment of the present invention, calculating the angle sampling coverage of each position of the scanned object according to the light source scanning trajectory of the local area scanning includes:

[0120] For any point in space Angle sampling range It is expressed as:

[0121]

[0122] in, is the light source trajectory, represents the unit vector parallel to the y-axis, and To pass through The two-dimensional coordinates of the projection point of the ray to the detector:

[0123]

[0124] Wherein, the detector pixel is represented by the two-dimensional coordinate (u, v), and are unit vectors along the u and v directions, is the unit vector in the direction of the perpendicular line from the light source to the detector plane, and D is the length of the perpendicular line.

[0125] Optionally, in one embodiment of the present invention, the reconstruction module further comprises:

[0126] Use N stages to restore the imaging field of view and set the threshold θ of the angle sampling coverage in the i-th stage i , and according to any point in space Angle sampling range Identify key restoration areas

[0127]

[0128] Design loss function weights So as to satisfy: and exist The boundary area has a smooth transition;

[0129] Combining the outputs of the neural networks at each stage, the final reconstruction result of the extended imaging field of view is obtained:

[0130]

[0131] in, represents the neural network of the i-th stage, and the parameters are recorded as Reconstruction results according to each stage Design a supervised or unsupervised loss function Loss(·) and As a weighted function, the objective function of neural network parameter optimization at each stage is constructed:

[0132]

[0133] Optionally, in one embodiment of the present invention, the decoupling module is further used to perform parallel or series decoupling of multiple artifacts using multiple neural networks or codecs.

[0134] Optionally, in one embodiment of the present invention, the decoupling module is further used to decouple projection features of the internal and external areas of the local region of interest using a background projection estimation neural network.

[0135] Optionally, in one embodiment of the present invention, the two decoupling processes are completed synchronously by one neural network or step by step by multiple different sub-networks.

[0136] Optionally, in one embodiment of the present invention, the neural networks used in the decoupling process and the reconstruction process of expanding the imaging field of view both meet the following conditions: having a sufficient receptive field and having the ability to encode spatial positions.

[0137] It should be noted that the above explanation of the embodiment of the X-ray CT local imaging method for expanding the imaging field of view is also applicable to the X-ray CT local imaging device for expanding the imaging field of view of this embodiment, and will not be repeated here.

[0138] According to the X-ray CT local imaging device for expanding the imaging field of view proposed in the embodiment of the present invention, the image reconstruction problem of local imaging is focused on, and based on the analysis of the data sampling conditions of different regions in the local scan, combined with the deep learning technology, a new multi-stage ROI reconstruction method for expandable imaging field of view is proposed. In the case of only the scanning data of the local ROI region, the influence of various data pathologies can be effectively decoupled, and the scanning data conditions of the inner and outer regions of the ROI are combined, and the reconstructible area is gradually expanded through the multi-stage deep learning technology to achieve high-quality reconstruction of the inner region of the ROI, and effectively restore the structure of the outer region of the ROI, thereby improving the image quality of X-ray CT local imaging. The present invention can be applied to the fields of medical treatment, security inspection, and industrial inspection, and is used for perspective imaging inspection of objects or human bodies.

[0139] Figure 8 A schematic diagram of the structure of an electronic device provided by an embodiment of the invention. The electronic device may include:

[0140] A memory 801 , a processor 802 , and a computer program stored in the memory 801 and executable on the processor 802 .

[0141] When the processor 802 executes the program, the X-ray CT local imaging method for expanding the imaging field of view provided in the above embodiment is implemented.

[0142] Furthermore, the electronic device further comprises:

[0143] The communication interface 803 is used for communication between the memory 801 and the processor 802 .

[0144] The memory 801 is used to store computer programs that can be executed on the processor 802 .

[0145] The memory 801 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0146] If the memory 801, the processor 802 and the communication interface 803 are implemented independently, the communication interface 803, the memory 801 and the processor 802 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0147] Optionally, in a specific implementation, if the memory 801, the processor 802 and the communication interface 803 are integrated on a chip, the memory 801, the processor 802 and the communication interface 803 can communicate with each other through an internal interface.

[0148] The processor 802 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0149] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the above-mentioned X-ray CT local imaging method for expanding the imaging field of view is implemented.

[0150] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0151] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0152] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present invention belong.

[0153] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0154] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

Claims

1. An X-ray CT local imaging method for expanding the imaging field of view, characterized in that: The following steps are involved: Obtain local imaging projection data p containing the local area of ​​interest in the local area scan IN (λ,u,v): Among them, λ∈[λ min ,λ max ] is the polar coordinate corresponding to the light source trajectory, λ min and λ max are the two endpoints of the light source trajectory, (u, v) represents the two-dimensional coordinates of the detector pixel, and p FULL (λ,u,v) is the complete projection data when the detector can completely receive all the rays from each light source position passing through the scanned object, M λ is the binary angle sampling mask, n p is the projection noise, u m With v m Indicates the maximum value of the row and column index of the detector unit, and is an indicative function used to limit the range of rays received by the detector; The angle sampling coverage of each position of the scanned object is calculated according to the light source scanning trajectory of the local area scanning, and the spatial distribution of the local imaging angle sampling is obtained, wherein the angle sampling coverage of each position of the scanned object is calculated according to the light source scanning trajectory of the local area scanning, including: For any point in space Angle sampling range It is expressed as: in, is the light source trajectory, represents the unit vector parallel to the y-axis, and To pass through The two-dimensional coordinates of the projection point of the ray to the detector: Here, the detector pixel is represented by a two-dimensional coordinate (u,v), and are unit vectors along the u and v directions, is the unit vector in the direction of the perpendicular line from the light source to the detector plane, and D is the length of the perpendicular line; Decoupling the influence of multiple data pathologies in the local imaging projection data to obtain local imaging projection data with only data truncation, decoupling the projection features of the local region of interest and the external region, obtaining projection data containing only the local region of interest based on the decoupled local imaging projection data, and reconstructing using the decoupled projection data containing only the local region of interest to obtain a reconstruction result inside the local region of interest; Determine the angle sampling threshold and loss function weight, use the local imaging projection data with only data truncation, and reconstruct the image through multi-stage neural network to obtain the reconstruction result of expanding the imaging field of view; specifically include: Use N stages to restore the imaging field of view and set the threshold θ of the angle sampling coverage in the i-th stage i , and according to any point in space Angle sampling range Identify key restoration areas Design loss function weights So as to satisfy: and exist The boundary area has a smooth transition; Combining the outputs of the neural networks at each stage, the final reconstruction result of the extended imaging field of view is obtained: in, represents the neural network of the i-th stage, and the parameters are recorded as Reconstruction results at each stage Design a supervised or unsupervised loss function Loss(·) and As a weighted function, the objective function of neural network parameter optimization at each stage is constructed:

2. The method according to claim 1, characterized in that Decoupling the effects of various data pathologies in the local imaging projection data, including: Use multiple neural networks or codecs to decouple multiple artifacts in parallel or in series.

3. The method according to claim 1, characterized in that Decouple the local region of interest from the external region projection features, including: A neural network is used to decouple the projection features of the areas inside and outside the local region of interest, and to estimate the contribution of the projection inside the region of interest and the projection outside the region of interest to the local imaging projection.

4. The method according to claim 2 or 3, characterized in that: The two decoupling processes are completed synchronously by one neural network or step by step by multiple different sub-networks.

5. The method according to claim 4, characterized in that The neural networks used in the decoupling process and the reconstruction process of expanding the imaging field of view all meet the following conditions: having sufficient receptive field and the ability to encode spatial position.

6. An X-ray CT local imaging device with an expanded imaging field of view, characterized in that: include: The acquisition module is used to obtain the local imaging projection data including the local area of ​​interest in the local area scanning, and calculate the angle sampling coverage of each position of the scanned object according to the light source scanning trajectory of the local area scanning, and obtain the spatial distribution of the local imaging angle sampling; wherein, the local imaging projection data p of the local area of ​​interest IN (λ,u,v) is: Among them, λ∈[λ min ,λ max ] is the polar coordinate corresponding to the light source trajectory, λ min and λ max are the two endpoints of the light source trajectory, (u, v) represents the two-dimensional coordinates of the detector pixel, and p FULL (λ,u,v) is the complete projection data when the detector can completely receive all the rays from each light source position passing through the scanned object, M λ is the binary angle sampling mask, n p is the projection noise, u m With v m Indicates the maximum value of the detector unit index, and is an indicative function used to limit the range of rays received by the detector; The angle sampling coverage of each position of the scanned object is calculated based on the light source scanning trajectory of the local area scan, including: For any point in space Angle sampling range It is expressed as: in, is the light source trajectory, represents the unit vector parallel to the y-axis, and To pass through The two-dimensional coordinates of the projection point of the ray to the detector: Here, the detector pixel is represented by a two-dimensional coordinate (u,v), and are unit vectors along the u and v directions, is the unit vector in the direction of the perpendicular line from the light source to the detector plane, and D is the length of the perpendicular line; A decoupling module is used to decouple the influence of various data pathologies in the local imaging projection data, obtain local imaging projection data with only data truncation, decouple the projection features of the local region of interest and the external region, obtain projection data containing only the local region of interest based on the decoupled local imaging projection data, and reconstruct using the decoupled projection data containing only the local region of interest to obtain a reconstruction result inside the local region of interest; The reconstruction module is used to determine the angle sampling threshold and the loss function weight, and obtain the reconstruction result of expanding the imaging field of view by using the local imaging projection data with only data truncation through multi-stage neural network reconstruction; The reconstruction module is further used to: Use N stages to restore the imaging field of view and set the threshold θ of the angle sampling coverage in the i-th stage i , and according to any point in space Angle sampling range Identify key restoration areas Design loss function weights So as to satisfy: and exist The boundary area has a smooth transition; Combining the outputs of the neural networks at each stage, the final reconstruction result of the extended imaging field of view is obtained: in, represents the neural network of the i-th stage, and the parameters are recorded as Reconstruction results according to each stage Design a supervised or unsupervised loss function Loss(·) and As a weighted function, the objective function of neural network parameter optimization at each stage is constructed:

7. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the X-ray CT local imaging method for expanding the imaging field of view as described in any one of claims 1 to 5.

Images