Method for reconstruction of spiral ct images based on neural radiance fields

Abstract

The application discloses a spiral CT image reconstruction method based on a neural radiation field, and belongs to the field of computer vision.The method comprises the following steps: determining an original projection image of a scanned object; the original projection image is obtained by a spiral CT device through a fan-beam scanning mode; according to a sampling angle corresponding to the original projection image, the original projection images with the same sampling angle in each cycle of a rotation process are classified and spliced through a rotation direction to obtain spliced projection images; and a spiral CT image is generated by calling an image reconstruction model based on the spliced projection images; wherein the image reconstruction model is constructed and trained based on a neural radiation field.The method reduces the radiation risk of related personnel.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of computer vision, and particularly relates to a spiral CT image reconstruction method based on neural radiance field. BACKGROUND

[0002] Computed Tomography (CT) is a method of obtaining the three-dimensional structure inside a target by acquiring two-dimensional ray transmission images through multi-angle projection and using mathematical reconstruction algorithms. CT is widely used in clinical diagnosis, industrial non-destructive testing, biological imaging and other scenarios. Its basic principle is to obtain projection by using the attenuation difference of X-rays in penetrating an object, and to reconstruct voxel-level tomographic images through algorithms such as Filtered Back Projection (FBP), Algebra Reconstruction Technique (ART) or Statistical Iterative Reconstruction (SIR).

[0003] At present, Spiral CT (Spiral CT) technology is widely used. Its working mode is to fix the X-ray source and the detector on the ring track of the Spiral CT device, rotate around the Z-axis of the Spiral CT device at a constant angular velocity, and at the same time, the scanning platform moves along the Z-axis at a constant speed, so that the X-rays pass through the scanned object placed on the scanning platform in a spiral path, and the original projection images are continuously acquired through the detector. Among them, the scanned object varies depending on the application field, such as clinical diagnosis, the scanned object is a human being; and for industrial non-destructive testing, the scanned object is a workpiece.

[0004] However, in the related art, the original projection images are generally acquired in a high angular density manner, thus the relevant personnel are at high risk of radiation. Among them, the angular density refers to the number of original projection images acquired per unit angle during the rotation of the detector around the scanned object; the relevant personnel include but are not limited to the operator of the Spiral CT device and the scanned object. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a Spiral CT image reconstruction method based on neural radiance field, which can solve the problem of how to reduce the radiation risk of relevant personnel.

[0006] In order to solve the above technical problems, the present application is implemented as follows:

[0007] In a first aspect, the embodiments of the present application provide a Spiral CT image reconstruction method based on neural radiance field, which comprises:

[0008] determine an original projection image of a scanned object; the original projection image is acquired by a spiral CT device using a fan beam scanning mode;

[0009] According to the sampling angle corresponding to the original projection image, the original projection images with the same sampling angle in each week of the rotation process are classified and spliced by the rotation direction to obtain a spliced projection image.

[0010] Based on the spliced projection image, an image reconstruction model is called to generate a spiral CT image.

[0011] The image reconstruction model is constructed and trained based on a neural radiation field.

[0012] In a second aspect, the embodiments of the present application provide a spiral CT image reconstruction device based on a neural radiation field, which comprises:

[0013] A determination module is configured to determine an original projection image of a scanned object; the original projection image is acquired by a spiral CT device using a fan beam scanning mode.

[0014] A splicing module is configured to classify and splice the original projection images with the same sampling angle in each week of the rotation process by the rotation direction according to the sampling angle corresponding to the original projection image to obtain a spliced projection image.

[0015] A calling module is configured to call an image reconstruction model based on the spliced projection image to generate a spiral CT image; the image reconstruction model is constructed and trained based on a neural radiation field.

[0016] In a third aspect, the embodiments of the present application provide a computer device, which comprises a processor, a memory, and a program or instruction stored on the memory and executable on the processor, and the program or instruction is executed by the processor to realize the steps of the method according to the first aspect.

[0017] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, which stores a program or instruction, and the program or instruction is executed by the processor to realize the steps of the method according to the first aspect.

[0018] In a fifth aspect, the embodiments of the present application provide a chip, which comprises a processor and a communication interface, the communication interface and the processor are coupled, and the processor is configured to run a program or instruction to realize the method according to the first aspect.

[0019] In the embodiments of the present application, a spiral CT image reconstruction method based on neural radiance field is provided. In the method, neural radiance field is used as an image reconstruction algorithm in the reconstruction of spiral CT images. Neural radiance field uses a neural network to model the density and color of each three-dimensional point in space, and can reconstruct a high-quality three-dimensional model from sparse input.

[0020] Therefore, in terms of image quality, the method utilizes the powerful ability of neural radiance field to model sparse viewing angles. The viewing angle refers to the relative angle between the detector and the scanned object when collecting image data. Specifically, neural radiance field can model the voxel density field through multi-angle ray projection data, and has a good spatial consistency expression ability in the reconstruction process to achieve high restoration of detail texture. Even in the context of sparse sampling of projection data, strong noise or slight artifact interference, the method can still reconstruct high-fidelity tomographic images, effectively preserving important information such as tissue boundaries and structural outlines.

[0021] Therefore, compared with high-angle density acquisition of original projection images, the method realizes high-quality reconstruction under low-angle density, i.e., reducing the number of viewing angles, thereby significantly reducing the number of X-ray exposures and the total radiation dose, and further reducing the radiation risk of relevant personnel. Especially in the field of clinical diagnosis, the method effectively alleviates the radiation risk of patients receiving CT examination while ensuring image quality, and has important clinical safety value. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a flowchart of a spiral CT image reconstruction method based on neural radiance field provided by some embodiments of the present application;

[0023] Figure 2 is a flowchart of model training provided by some embodiments of the present application;

[0024] Figure 3a is a spliced projection image without content deduplication processing provided by some embodiments of the present application;

[0025] Figure 3b is another spliced projection image after content deduplication processing provided by some embodiments of the present application;

[0026] Figure 4a is a comparison chart of rotated projection images with sampling angles of 0° and 360° without viewing angle correction provided by some embodiments of the present application;

[0027] Figure 4b is another comparison chart of rotated projection images with sampling angles of 0° and 360° after viewing angle correction provided by some embodiments of the present application;

[0028] Figure 5a is an all-section view of a three-dimensional reconstruction result obtained by an image reconstruction model in a case where no view angle correction is performed on the projection images after rotation, provided by some embodiments of the present application;

[0029] Figure 5b is another all-section view of a three-dimensional reconstruction result obtained by an image reconstruction model after view angle correction is performed on the projection images after rotation, provided by some embodiments of the present application;

[0030] Figure 6 is a structural block diagram of a neural-radiance-field-based helical CT image reconstruction device, provided by some embodiments of the present application;

[0031] Figure 7 is an internal structural diagram of a computer device, provided by some embodiments of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0033] The terms "first", "second", and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein. In addition, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / ", generally indicates that the front and rear associated objects are in an "or" relationship.

[0034] In order to facilitate understanding, the technical terms that may be involved in the embodiments of the present application are explained as follows:

[0035] Fan Beam: a beam shape of an X-ray source. X-rays are diverged along a two-dimensional plane (with a certain thickness), and a line array detector is used to rotate around the scanned object to collect projection data.

[0036] Cone Beam: a beam shape of an X-ray source. X-rays are diverged from a point source to a three-dimensional space, and a surface array detector is used for collection, which can cover a larger volume area in one scan.

[0037] Among them, fan-beam or cone-beam can be used in clinical diagnosis, industrial non-destructive testing, biological imaging and other scenarios to scan the scanned object. Compared with fan-beam, cone-beam can achieve higher spatial resolution and faster scanning speed, so it is often used in oral CT, industrial CT and other scenarios, while fan-beam is often used in clinical diagnosis.

[0038] Viewing angle height: during the scanning process, the detector rotates horizontally while there is also a vertical displacement relative to the scanned object. Due to the limited size of the detector, after the scanned object is displaced, its position in the field of view of the detector also changes, i.e. the position of the scanned object in the field of view of the detector in the vertical direction changes.

[0039] Viewing angle structure: the position of the scanned object in the field of view of the detector during rotation.

[0040] Rotation plane: the plane in which the X-ray source and the detector rotate on the ring track.

[0041] Vertical direction: the direction perpendicular to the rotation plane, which is also the direction of movement of the scanning platform, also known as the Z-axis direction.

[0042] Spiral CT data: the projection obtained by the detector of the spiral CT device in a spiral acquisition path.

[0043] Spiral CT image: the tomographic image obtained by reconstructing the above projection.

[0044] Each revolution of the rotation process: one revolution of the X-ray source and the detector on the ring track is 360°.

[0045] Sampling angle: the X-ray source on the spiral CT device emits X-rays once every fixed angle, and the detector receives X-rays once at the same time, completing sampling.

[0046] Angular density: in fan-beam CT, the number of X-ray photons received by the detector in a unit angle.

[0047] Unit angle: the angular interval during sampling by the detector. It can be 1°, 2°, 3°, etc. For example, if the unit angle is 1°, the X-ray source emits X-rays once every 1°, and correspondingly, the detector receives X-rays once.

[0048] Initial pose of horizontal rotation of 360°: it includes the pose of the detector at each sampling angle during one revolution.

[0049] Rotation direction: the direction of the detector and the X-ray source during rotation, including counterclockwise rotation and clockwise rotation.

[0050] The spiral CT image reconstruction method based on neural radiance field provided by the embodiments of the present application will be described in detail below with reference to the specific embodiments and application scenarios thereof in combination with the accompanying drawings.

[0051] It should be noted that, as mentioned in the background, the original projection images are generally collected in a high angular density manner, so there is a high radiation risk for relevant personnel. The high radiation risk is caused by collecting the original projection images in a high angular density manner, and therefore, to reduce the radiation risk, the angular density needs to be reduced.

[0052] To solve the above problems, the applicant has found that high angular density means that more original projection images are collected within a unit angle, so that the quality of the reconstructed spiral CT image is higher and the details are richer, but at the same time, the scanning time and radiation dose are increased; low angular density means that fewer original projection images are collected within a unit angle, so that the quality of the reconstructed spiral CT image is reduced and the details are lost, but the scanning time and radiation dose are also reduced accordingly.

[0053] Therefore, increasing the angular density aims to improve the quality, while reducing the angular density will result in a decrease in quality, which is not worth the loss. The reason is that the image reconstruction algorithm currently used for spiral CT image reconstruction based on neural radiance field needs high angular density to ensure the reconstruction quality.

[0054] Therefore, in an exemplary embodiment, the present application proposes a spiral CT image reconstruction method based on neural radiance field, which uses neural radiance field (NeRF) as an image reconstruction algorithm when reconstructing a spiral CT image. Neural radiance field uses neural network to model the density and color of each three-dimensional point in space, and can realize the reconstruction of high-quality three-dimensional model from sparse input.

[0055] Therefore, in terms of image quality, this method takes advantage of the powerful ability of neural radiance field to model sparse viewing angles. Among them, the viewing angle refers to the relative angle between the detector and the scanned object when collecting image data. Specifically, neural radiance field can model the voxel density field through multi-angle ray projection data, and realize high restoration of detail texture in the reconstruction process, with good spatial consistency expression ability. Even in the context of sparse sampling of projection data, strong noise or slight artifact interference, this method can still reconstruct high-fidelity tomographic images, effectively preserving important information such as tissue boundaries and structural outlines.

[0056] Therefore, compared with high-angle-density acquisition of original projection images, the method realizes high-quality reconstruction under low-angle-density, i.e., a reduced number of viewing angles, thereby significantly reducing the number of X-ray exposures and the total radiation dose, and further reducing the radiation risk of related personnel. Especially in the field of clinical diagnosis, the method effectively alleviates the radiation risk of patients receiving CT examination while ensuring image quality, and has important clinical safety value.

[0057] However, the applicant found that the data collected by the spiral CT device is irregularly sampled in space, i.e., the projection is constantly moving in the Z-axis direction and the angle of each projection image changes continuously, resulting in inconsistent viewing angle height between projections, which brings certain difficulties to the use of reconstruction algorithms. Among them, the inconsistent viewing angle height refers to the inconsistent position of the scanned object in the detector field of view in different projection images. In addition, since the relative motion trajectory of the detector and the scanned object is a spiral line, the projection data collected is an approximate one-dimensional sinusoidal curve (sinogram), i.e., the width of the projection is much larger than the height.

[0058] However, the neural radiance field usually uses a two-dimensional image with uniform viewing angles as input, aiming to ensure the geometric consistency and ray consistency in the training process of the neural radiance field model. The reconstruction quality depends largely on the spatial uniformity of the input viewing angle and the stability of the light density distribution. Among them, uniform viewing angle refers to equal-interval rotation acquisition in the same horizontal plane, such as a plane perpendicular to the Z-axis, i.e., consistent viewing angle height, i.e., consistent viewing angle structure.

[0059] It should be noted that uniform viewing angles can ensure the spatial uniformity of the viewing angle and the stability of the light density distribution, i.e., geometric consistency and ray consistency.

[0060] Therefore, the current neural radiance field is difficult to be directly applied to the three-dimensional reconstruction task of spiral CT images.

[0061] In summary, due to the inconsistent viewing angle height between projections and the approximately one-dimensional data of the projection, there is a natural difference between spiral CT data and neural radiance field methods in terms of data organization and viewing angle structure, which makes spiral CT data unable to meet the requirements of neural radiance field for input data, causing them to work together, and ultimately leading to the current neural radiance field being difficult to be directly applied to the three-dimensional reconstruction task of spiral CT images.

[0062] To solve the above problems, the embodiment provides a scheme for effectively converting the data structure of spiral CT data into a projection form suitable for neural radiation field modeling, that is, for each original projection image having the same sampling angle in each week of the rotation process, the original projection images are classified and spliced by the rotation direction according to the sampling angle corresponding to the original projection image, to obtain a spliced projection image; and based on the spliced projection image, an image reconstruction model is called to generate a spiral CT image. Thus, the above scheme can realize compatibility of the data acquisition mode of spiral CT data and full play of the neural radiation field reconstruction capability.

[0063] Specifically, the scheme first realizes reconstructive conversion of spiral CT data (i.e., original projection images) collected by a spiral CT device using a fan beam scanning mode at the data structure level. That is, by splicing the projection data in each week of the spiral path, the original projection images are effectively converted into multi-angle projection images of an approximate cone beam. That is, the splicing is performed in the height dimension of the original projection image, that is, the width of the projection is much greater than the height, so that the width of the projection is improved to a certain extent, the spiral CT data is converted from one dimension to two dimensions, and the difference between the spiral CT data and the neural radiation field method in the data organization mode is eliminated. Moreover, the viewing angle height of each original projection image collected on the spiral path changes along the Z-axis direction, so that the scanning of the scanned object is completed, and each spliced projection image is a complete image of the scanned object corresponding to each sampling angle, that is, the viewing angle structure of the spiral CT data is converted into a viewing angle structure meeting the requirements of the neural radiation field for input data. Thus, a spatial structure meeting the modeling requirements of the neural radiation field is constructed. This conversion not only improves the compatibility between the original projection image and the neural radiation field, but also significantly widens the application range of the neural radiation field in the CT field, and breaks through the technical barrier between the spiral scanning mechanism and the neural reconstruction network.

[0064] The following takes a spiral CT applied to clinical diagnosis and using a fan beam scanning as an example to describe each process step of the method.

[0065] With reference to Figure 1 The method includes steps 102-106. Wherein:

[0066] Step 102, determine the original projection image of the scanned object; the original projection image is collected by a spiral CT device using a fan beam scanning mode.

[0067] After the patient is positioned on the scanning platform, the X-ray source and detector rotate at a constant speed around the Z-axis of the spiral CT system, while the scanning platform translates at a constant speed along the Z-axis, achieving continuous spiral scanning and acquiring a large number of fan-beam raw projection images. The acquired raw projection images are X-ray projection images in DICOM format. The detector size is H x W, meaning the raw projection image size is H x W. Here, H is the height of the raw projection image, and W is the width.

[0068] For example, the original projection images are stored in a DICOM file set arranged in the order of acquisition time, denoted as ,in, Indicates the k Assuming that the number of images collected per rotation in spiral scanning is N, the total number of images L should be M×N, where M represents a total of M rotations.

[0069] Step 104 , according to the sampling angle corresponding to the original projection image, the original projection images with the same sampling angle in each rotation process are classified and spliced ​​according to the rotation direction to obtain a spliced ​​projection image.

[0070] The classification splicing can be achieved by the following formula 1:

[0071] (one)

[0072] in, is the projected image after stitching; j is the number of the sampling angle; N is the number of the original projection images collected during one rotation, M is the number of rotations; For the M The number of the sampling angle in the circle is j The original projected image of For the The number of the sampling angle in the circle is j The original projected image of The number of the sampling angle in the first circle is j The original projected image of I The subscript of is the unique number of the angle of the original projection image during the rotation process.

[0073] For example, the unit angle is 1°, 360 original projection images (N is 360) are collected per week, that is, the detector collects an original projection image every 1° rotation; the number of rotations is 100, and a total of 360*100 original projection images are collected during the rotation. Based on this, taking the sampling angle of 1° as an example, the first image (first week) is taken first, then the 361st image (second week) is taken, and so on, until the 100 original projection images corresponding to the sampling angle of 1° in the sampling process are taken, and they are spliced into the spliced projection image corresponding to the sampling angle. After that, for each sampling angle, the splicing in the above manner is performed, so as to obtain the spliced projection images corresponding to all sampling angles, that is, 360 spliced projection images are obtained.

[0074] In step 106, based on the spliced projection image, an image reconstruction model is called to generate a spiral CT image; wherein the image reconstruction model is obtained based on a neural radiation field.

[0075] Wherein, the image reconstruction model can process the input spliced projection image and the input horizontal ring around 360° initialization pose, so as to output the spiral CT image, that is, the CT reconstruction slice image, and realize the complete volume CT image reconstruction.

[0076] Wherein, the initialization pose is mainly used to provide geometric constraints and view angle related information for the reconstruction of three-dimensional scene.

[0077] In the training process of the image reconstruction model:

[0078] Referring to Figure 2 , the training process is: based on a volume rendering pipeline, first, pose and scene modeling are performed, and a three-dimensional density field is generated through a decoding network; then the three-dimensional density is line integrated to obtain a rendering projection; finally, the rendering projection and the input projection are calculated to obtain a loss, and the network is updated until the model training end requirement is met, and the training is ended.

[0079] The training process specifically includes the following steps 1 to 4:

[0080] Step 1, scene modeling and light sampling.

[0081] That is, computer three-dimensional modeling is performed; and the light emitted by the light source is simulated, and the light is sampled on the light.

[0082] Specifically, for a given pixel point p of a specific pose view angle, the observation pixel value is represented as formula two:

[0083] (II)

[0084] Wherein, is the pixel value of p point; is the light intensity of the light source; Equation of the light ray that is incident on the pixel point p, and are the start and end points of the light ray that is integrated in the reconstructed scene.

[0085] The discrete representation of Equation Two is Equation Three:

[0086] (Three)

[0087] where, denotes the number of sampling points on the light path; i i

[0088] Step 2, Obtain the neural network features of the sampling points.

[0089] Convert the global coordinates of the sampling points into corresponding local coordinates inside the node, further extract the features of the local coordinate positions , and perform trilinear interpolation with the corresponding network weights through Equation Four to obtain the neural feature vector of the sampling points

[0090] (Four)

[0091] where, is the corresponding trilinear interpolation weight.

[0092] Step 3, Decode network training.

[0093] Through the global decoding network shared by all sampling points and Equation Five, convert the neural feature vector of the sampling points into the required output quantity (i.e., density

[0094] (Five)

[0095] Exemplarily, the global decoding network may be a 3-layer multilayer perceptron (MLP) with 64 neurons, which has 4081 learnable parameters.

[0096] wherein the SiLU activation function is used inside the MLP, and the SoftPlus activation function is used in the last layer. Finally, the density is multiplied by the weight

[0097] ​​​​​​Step 4, calculate the loss function.

[0098] In order to optimize the parameters of the neural feature and the decoding network, this step uses the mean square error of the estimated pixel value and the real pixel value of the target view to calculate the loss, and the mean square error is expressed as formula six:

[0099] (six)

[0100] wherein, is the real pixel value of the pixel point p, is the estimated pixel value of the pixel point p.

[0101] Further, referring to Figure 3a wherein, the dashed box is an enlarged view, after the spliced projection image is obtained by splicing according to the above method, the applicant finds that the continuity of the splicing part in the spliced projection image is poor, and this problem causes the spiral CT image obtained by subsequent reconstruction to have large artifacts, thereby affecting the resolution and clarity of the reconstruction.

[0102] In order to solve this problem, the applicant finds that the reason for this problem is that the pictures captured by the detector will overlap during the process of the scanning bed translating along the Z axis. Therefore, in this embodiment, the original projection images are subjected to content deduplication processing to obtain deduplicated projection images; then the deduplicated projection images are classified and spliced, that is, the deduplicated projection images having the same sampling angle in each week of the rotation process are classified and spliced according to the rotation direction to obtain a spliced projection image, thereby solving the problem of poor continuity of the splicing part in the spliced projection image.

[0103] In an embodiment, the content deduplication processing can be realized by pixel point matching and overlap detection on two adjacent original projection images. That is, for every two adjacent images in the original projection images, the pixel values of the pixel points at the same coordinate position are used for overlap detection (such as whether the difference between the pixel values is within a preset threshold, if yes, it is determined that there is overlap), and the overlapping part is the part that needs to be deduplicated.

[0104] In another embodiment, the content deduplication processing is cropping.

[0105] Specifically, for all original projection images, determine the overlapping pictures in two adjacent original projection images in the acquisition sequence; for the overlapping pictures, crop each original projection image; and the remaining part of the original projection image after cropping the overlapping pictures is taken as a deduplicated projection image.

[0106] For example, for a certain scanning process, the number of pixels in the above-mentioned overlapping picture is fixed, assuming that the overlapping picture is the first w columns of the original projection picture starting from the second picture, then for each original projection picture , the following clipping is performed by formula seven:

[0107] (seven)

[0108] That is, the first w columns of the original projection picture are taken, and then the de-duplicated projection picture is obtained.

[0109] Correspondingly, for each de-duplicated projection picture having the same sampling angle in each rotation process according to the sampling angle corresponding to the de-duplicated projection picture, the step of classifying and splicing to obtain a spliced projection picture by the rotation direction is as follows:

[0110] For each angle number j of the sampling angle , a new spliced projection picture can be constructed.

[0111] (eight)

[0112] , wherein is the spliced projection picture, that is, the spliced "columnar image", each row of which represents the projection of the same angle in multiple circles at a certain position; j is the number of the sampling angle; N is the number of the original projection pictures collected in one rotation, M is the number of rotations; is the de-duplicated projection picture in which the number of the sampling angle in the first M rotation is j ; is the de-duplicated projection picture in which the number of the sampling angle in the first rotation is j ; is the de-duplicated projection picture in which the number of the sampling angle in the first rotation is j ; I , wherein the subscript is the unique number of the angle of the de-duplicated projection picture in the rotation process.

[0113] Referring to Figure 3b , Figure 3b is the de-duplicated (corrected splicing width) spliced projection picture. The dashed box is an enlarged view, and it can be seen that the continuity of the spliced projection picture after the correction of the splicing width is greatly improved.

[0114] Exemplarily, since the rotation axis direction of the spiral CT data is different from the rotation axis direction of the standard input format of the neural radiation field, the former is a horizontal direction, and the latter is a vertical direction. In order to improve the efficiency of the neural radiation field model in processing the spliced projection image, in the embodiment, the spliced projection image is rotated by 90° to obtain a rotated projection image; based on the rotated projection image, an image reconstruction model is called to generate a spiral CT image.

[0115] wherein the rotation operation can be implemented by formula nine:

[0116] (nine)

[0117] After the above rotation operation, the dimension of the rotated projection image becomes so that the rotated projection image conforms to the standard input format of the neural radiation field.

[0118] Further, the applicant has found that, referring to Figure 4a Since the scanning bed is always in uniform motion during the acquisition process of the spiral CT data, the starting position of the rotated projection image of each sampling view angle exists a slight displacement in the Z-axis direction, which reduces the reconstruction quality of the model. That is, due to the vertical motion of the spiral motion trajectory, the first row of the rotated projection image when the sampling angle is 0° and the first row of the rotated projection image when the sampling angle is 1° will exist a slight displacement deviation in the vertical direction. With the increase of the sampling angle, the displacement deviation increases accordingly, so there will be a large displacement deviation between the first rotated projection image and the last rotated projection image in one rotation.

[0119] In order to solve the above problem, the embodiment proposes a scheme of performing view angle correction on each rotated projection image, so as to eliminate the displacement deviation. That is, the rotated projection image is subjected to view angle correction to obtain a corrected projection image; based on the corrected projection image, an image reconstruction model is called to generate a spiral CT image.

[0120] Referring to Figure 5a which is a cross-sectional view of a three-dimensional reconstruction result reconstructed by the image reconstruction model without view angle correction of the rotated projection image, the artifact degree in the cross-sectional view is higher; referring to Figure 5b which is a cross-sectional view of a three-dimensional reconstruction result reconstructed by the image reconstruction model after view angle correction of the rotated projection image, the artifact degree in the cross-sectional view is reduced to a certain extent. Therefore, the embodiment reduces the artifact degree, thereby improving the resolution and clarity of the three-dimensional reconstruction result, and further improving the resolution and clarity of the spiral CT image.

[0121] For the above-mentioned view correction operation, it includes but is not limited to the following operation modes:

[0122] Operation mode one:

[0123] Firstly, the pixel offset of the rotated projection image in the Z-axis direction of the spiral CT device is determined. That is, the Z-axis offset modeling is performed.

[0124] Specifically, in order to simulate the offset of the detector in the Z-axis direction during the spiral scanning process, it is assumed that the maximum offset value of the rotated projection image in the Z-axis direction is when the sampling angle position is 0° and the sampling angle position is 360°. Pixel, the first i The offset of the i-th rotated projection image can be calculated by formula ten:

[0125] (Ten)

[0126] Wherein, .

[0127] Secondly:

[0128] In an embodiment, the correspondence between the pixel offset and each pixel point in each rotated projection image is established, and the correspondence is input into the image reconstruction model during the process of training the image reconstruction model for learning, so that the trained image reconstruction model can automatically correct the input rotated projection image when reconstructing.

[0129] In another embodiment, a mapping relationship between the original pixel coordinates and the target pixel coordinates is established, and based on the mapping relationship, the pixel value at the original pixel coordinates is interpolated to the target pixel coordinates to obtain a corrected projection image.

[0130] Wherein, the mapping relationship between the original pixel coordinates and the target pixel coordinates is established, including:

[0131] For each column j in the rotated projection image, the original pixel coordinates are recorded as:

[0132] (Eleven)

[0133] Wherein, is the transpose of the matrix formed by the original pixel coordinates of each pixel point in each column j in the rotated projection image.

[0134] Then the aligned target pixel coordinates can be represented by formula twelve:

[0135] (Twelve)

[0136] The range of the longitudinal coordinate (Z-axis direction) of the target pixel coordinate is clipped to a legal interval by Equation Thirteen:

[0137] (Thirteen)

[0138] The range of the longitudinal coordinate of the aligned target pixel coordinate can be limited by Equation Thirteen, i.e., the longitudinal coordinate of the target pixel coordinate is ensured to be within the legal interval [1, H], so as to avoid calculation errors caused by exceeding the boundary.

[0139] The mapping relationship is that The pixel offset and the mapping relationship are aligned.

[0140] After obtaining the mapping relationship, the pixel points of the rotated projection image are re-sampled column by column by interpolation. That is, for each column of pixel points in the rotated projection image j , the pixel value of each column in the rotated projection image is interpolated to the aligned by the mapping relationship and an interpolation method (such as one-dimensional linear interpolation, nearest-neighbor interpolation, etc.). For example, if the interpolation method is one-dimensional linear interpolation, then according to Equation Fourteen, the pixel value of each column in the rotated projection image can be interpolated to the aligned :

[0141] (Fourteen)

[0142] wherein is a one-dimensional linear interpolation formula.

[0143] In summary, the projection image after Z-axis translation alignment (i.e., the corrected projection image) is .

[0144] Referring to Figure 4b , Figure 4b is the rotated projection image after view angle correction. As can be seen from the dashed line position in Figure 4a , before view angle correction, the displacement deviation of the rotated projection image at the sampling angle of 0° and the sampling angle of 360° in the height direction is large; as can be seen from the dashed line position in Figure 4b , after view angle correction, the displacement deviation is eliminated.

[0145] Operation Mode Two:

[0146] ​The pixel coordinates of each pixel point in the rotated post-projection image at a sampling angle of 0° are taken as the reference coordinates, the pixel correspondence between the rotated post-projection image corresponding to each sampling angle and the rotated post-projection image at a sampling angle of 0° is established, and the pixel correspondence is input into the image reconstruction model during the process of training the image reconstruction model, so as to enable the trained image reconstruction model to automatically correct the viewing angle of the input rotated post-projection image during reconstruction.

[0147] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0148] Based on the same inventive concept, the embodiments of the present application also provide a neural radiation field based spiral CT image reconstruction device for implementing the above-mentioned neural radiation field based spiral CT image reconstruction method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more neural radiation field based spiral CT image reconstruction device embodiments provided below can refer to the limitations of the neural radiation field based spiral CT image reconstruction method in the foregoing, which will not be repeated here.

[0149] In one exemplary embodiment, as shown in Figure 6 A neural radiation field based spiral CT image reconstruction device is provided, comprising: a determination module 100, a splicing module 200 and a calling module 300, wherein:

[0150] The determination module 100 is configured to determine an original projection image of a scanned object, wherein the original projection image is acquired by a spiral CT device using a fan beam scanning mode.

[0151] The splicing module 200 is configured to classify and splice the original projection images having the same sampling angle in each cycle of the rotation process by the rotation direction according to the sampling angle corresponding to the original projection image, to obtain a spliced post-projection image.

[0152] The calling module 300 is configured to call an image reconstruction model based on the spliced projection image to generate a spiral CT image, wherein the image reconstruction model is constructed and trained based on a neural radiation field.

[0153] Optionally, the splicing module 200 is specifically configured to:

[0154] perform content deduplication processing on the original projection image to obtain a deduplicated projection image;

[0155] According to the sampling angle corresponding to the deduplicated projection image, the deduplicated projection images having the same sampling angle in each cycle of the rotation process are classified and spliced through the rotation direction to obtain a spliced projection image.

[0156] Optionally, the splicing module 200 is further configured to:

[0157] For all original projection images, determine the overlapping picture in two original projection images adjacent in the acquisition sequence;

[0158] For the overlapping picture, each original projection image is cropped;

[0159] The remaining part of the original projection image after cropping the overlapping picture is taken as a deduplicated projection image.

[0160] Optionally, the classified splicing

[0161]

[0162] wherein, is a spliced projection image; j is the number of the sampling angle; N is the number of the original projection images collected in a rotation cycle, M is the number of rotation cycles; is the deduplicated projection image whose number of the sampling angle is M in the first cycle; j is the deduplicated projection image whose number of the sampling angle is in the first cycle; is the deduplicated projection image whose number of the sampling angle is j in the first cycle; is the deduplicated projection image whose number of the sampling angle is j in the first cycle; I is the subscript of the deduplicated projection image in the rotation process.

[0163] Optionally, the calling module 300 is specifically configured to:

[0164] rotating the spliced projection image by 90° to obtain a rotated projection image;

[0165] based on the rotated projection image, calling an image reconstruction model to generate a spiral CT image.

[0166] Optionally, the calling module 300 is further configured to:

[0167] performing view angle correction on the rotated projection image to obtain a corrected projection image;

[0168] based on the corrected projection image, calling an image reconstruction model to generate a spiral CT image.

[0169] Optionally, the calling module 300 is further configured to:

[0170] determining a pixel offset of the rotated projection image in the Z-axis direction of the spiral CT device;

[0171] based on the pixel offset, performing coordinate adjustment on original coordinates of pixels in the rotated projection image to obtain a corrected projection image.

[0172] Optionally, the calling module 300 is further configured to:

[0173] based on the pixel offset, establishing a mapping relationship between original pixel coordinates and target pixel coordinates;

[0174] based on the mapping relationship, interpolating pixel values at the original pixel coordinates to the target pixel coordinates to obtain a corrected projection image.

[0175] The above various modules in the spiral CT image reconstruction device based on neural radiation field can be all or partially realized by software, hardware and combinations thereof. The above various modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the above various modules.

[0176] In one exemplary embodiment, a computer device is provided, which can be a terminal, and an internal structure diagram thereof can be as shown in Figure 7The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to perform wired or wireless communication with external terminals. The wireless communication can be achieved through WIFI, mobile cellular network, near field communication (NFC) or other technologies. The computer program is executed by the processor to implement a spiral CT image reconstruction method based on a neural radiation field. The display unit of the computer device is configured to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0177] Those skilled in the art can understand that, Figure 7 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0178] In one embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the following steps:

[0179] determining an original projection image of a scanned object; the original projection image is acquired by a spiral CT device using a fan beam scanning mode;

[0180] According to the sampling angle corresponding to the original projection image, the original projection images with the same sampling angle in each cycle of the rotation process are classified and spliced by the rotation direction to obtain a spliced projection image;

[0181] Based on the spliced projection image, an image reconstruction model is called to generate a spiral CT image;

[0182] The image reconstruction model is constructed and trained based on a neural radiance field.

[0183] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0184] The original projection images are subjected to content deduplication processing to obtain deduplicated projection images.

[0185] According to the sampling angle corresponding to the deduplicated projection image, the deduplicated projection images having the same sampling angle in each cycle of the rotation process are classified and spliced by the rotation direction to obtain spliced projection images.

[0186] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0187] For all original projection images, determine the overlapping picture in the two original projection images adjacent in the acquisition sequence;

[0188] For the overlapping picture, each original projection image is cropped;

[0189] The remaining part of the original projection image after cropping the overlapping picture is taken as a deduplicated projection image.

[0190] In one embodiment, the classified and spliced

[0191]

[0192] wherein, is a spliced projection image; j is the number of the sampling angle; N is the number of the original projection images collected in a rotation cycle, M is the number of rotation cycles; is the deduplicated projection image whose number of the sampling angle is M in the first cycle; j is the deduplicated projection image whose number of the sampling angle is in the first cycle; is the deduplicated projection image whose number of the sampling angle is j in the first cycle; is the deduplicated projection image whose number of the sampling angle is j in the first cycle; I is the subscript of the angle of the deduplicated projection image in the rotation process.

[0193] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0194] The spliced projection image is rotated by 90° to obtain a rotated projection image.

[0195] based on the rotated projection image, calling an image reconstruction model to generate a spiral CT image.

[0196] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0197] angle correction is performed on the rotated projection image to obtain a corrected projection image;

[0198] based on the corrected projection image, calling an image reconstruction model to generate a spiral CT image.

[0199] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0200] determining a pixel offset of the rotated projection image in the Z-axis direction of the spiral CT device;

[0201] based on the pixel offset, adjusting the original coordinates of the pixels in the rotated projection image to obtain a corrected projection image.

[0202] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0203] based on the pixel offset, establishing a mapping relationship between the original pixel coordinates and the target pixel coordinates;

[0204] based on the mapping relationship, interpolating the pixel values at the original pixel coordinates to the target pixel coordinates to obtain a corrected projection image.

[0205] In one embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the following steps:

[0206] determining an original projection image of a scanned object; the original projection image is obtained by a spiral CT device using a fan-beam scanning method;

[0207] According to the sampling angle corresponding to the original projection image, the original projection images with the same sampling angle in each cycle of the rotation process are classified and spliced by the rotation direction to obtain a spliced projection image;

[0208] based on the spliced projection image, calling an image reconstruction model to generate a spiral CT image;

[0209] The image reconstruction model is constructed and trained based on a neural radiation field.

[0210] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0211] performing content de-duplication processing on the original projection images to obtain de-duplicated projection images;

[0212] According to the sampling angle corresponding to the de-duplicated projection image, the de-duplicated projection images having the same sampling angle in each week of the rotation process are classified and spliced through the rotation direction to obtain spliced projection images.

[0213] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0214] For all original projection images, determine the overlapping picture in two original projection images adjacent in the acquisition sequence;

[0215] For the overlapping picture, crop each original projection image;

[0216] The remaining part of the original projection image after cropping the overlapping picture is taken as a de-duplicated projection image.

[0217] In one embodiment, the classified and spliced

[0218]

[0219] wherein, is a spliced projection image; j is the number of the sampling angle; N is the number of the original projection images collected in one rotation, M is the number of rotations; is the number of the sampling angle in the first M rotation, which is j de-duplicated projection image; is the number of the sampling angle in the first rotation, which is j de-duplicated projection image; is the number of the sampling angle in the first rotation, which is j de-duplicated projection image; I is the unique number of the angle of the de-duplicated projection image in the rotation process.

[0220] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0221] rotating the spliced projection image by 90° to obtain a rotated projection image;

[0222] Based on the rotated projection image, calling an image reconstruction model to generate a spiral CT image.

[0223] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0224] perspective correction is performed on the rotated projection image to obtain a corrected projection image;

[0225] Based on the corrected projection image, an image reconstruction model is called to generate a spiral CT image.

[0226] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0227] determining a pixel offset of the rotated projection image in the Z-axis direction of the spiral CT device;

[0228] Based on the pixel offset, the original coordinates of the pixels in the rotated projection image are adjusted to obtain a corrected projection image.

[0229] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0230] Based on the pixel offset, a mapping relationship between the original pixel coordinates and the target pixel coordinates is established;

[0231] Based on the mapping relationship, the pixel value at the original pixel coordinates is interpolated to the target pixel coordinates to obtain a corrected projection image.

[0232] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0233] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0234] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A spiral CT image reconstruction method based on neural radiation field, characterized in that: The spiral CT image reconstruction method based on the nerve radiation field includes: Determining an original projection image of the scanned object; the original projection image is acquired by a spiral CT device using a fan beam scanning method; According to the sampling angle corresponding to the original projection image, the original projection images with the same sampling angle in each rotation process are classified and spliced ​​according to the rotation direction to obtain a spliced ​​projection image; Based on the spliced ​​projection image, calling the image reconstruction model to generate a spiral CT image; Wherein, the image reconstruction model is constructed and trained based on the neural radiation field; The method further comprises: performing content deduplication processing on the original projection image to obtain a deduplicated projection image, and classifying and splicing the original projection images with the same sampling angle in each week of the rotation process according to the sampling angle corresponding to the original projection image according to the rotation direction to obtain a spliced ​​projection image; The classification splicing is achieved by the following formula: ,in, is the projected image after stitching; j is the number of the sampling angle; N is the number of the original projection images collected during one rotation, M is the number of rotations; For the M The number of the sampling angle in the circle is j The deduplicated projected image; For the The number of the sampling angle in the circle is j The deduplicated projected image; The number of the sampling angle in the first circle is j The deduplicated projected image; I The subscript of is the unique number of the angle of the deduplicated projected image during the rotation process.

2. The method for spiral CT image reconstruction based on neural radiation field according to claim 1, characterized in that: The performing content deduplication processing on the original projection image to obtain a deduplicated projection image includes: For all original projection images, determining overlapping frames in two original projection images that are adjacent in acquisition order; For the overlapping images, cropping each original projection image; The remaining portion of the original projection image after the overlapping frames are cut out is used as a deduplicated projection image.

3. The method for spiral CT image reconstruction based on neural radiation field according to any one of claims 1 to 2, characterized in that: The step of calling an image reconstruction model based on the spliced ​​projection image to generate a spiral CT image includes: Rotating the spliced ​​projected image by 90° to obtain a rotated projected image; Based on the rotated projection image, an image reconstruction model is called to generate a spiral CT image.

4. The method for spiral CT image reconstruction based on neural radiation field according to claim 3, characterized in that: The step of calling an image reconstruction model based on the rotated projection image to generate a spiral CT image includes: Performing visual angle correction on the rotated projected image to obtain a corrected projected image; Based on the corrected projection image, an image reconstruction model is called to generate a spiral CT image.

5. The method for spiral CT image reconstruction based on neural radiation field according to claim 4, characterized in that: The performing perspective correction on the rotated projected image to obtain a corrected projected image includes: Determining a pixel offset of the rotated projection image in a Z-axis direction of the spiral CT device; Based on the pixel offset, coordinate adjustment is performed on the original coordinates of the pixels in the rotated projection image to obtain a corrected projection image.

6. The method for spiral CT image reconstruction based on neural radiation field according to claim 5, characterized in that: The step of adjusting original coordinates of pixels in the rotated projected image based on the pixel offset to obtain a corrected projected image includes: Based on the pixel offset, a mapping relationship between the original pixel coordinates and the target pixel coordinates is established; Based on the mapping relationship, the pixel values ​​at the original pixel coordinates are interpolated to the target pixel coordinates to obtain a corrected projected image.

7. A computer device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the spiral CT image reconstruction method based on the neural radiation field as described in any one of claims 1 to 6.

8. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the spiral CT image reconstruction method based on the neural radiation field are implemented as described in any one of claims 1 to 6.

Images