Graph construction method, computer device and storage medium

Abstract

The present application relates to a method, device, computer equipment and storage medium for constructing a map. The method includes: determining an image from the acquired group images as the first iterative image; the group images include images of multiple different objects; using the first iterative image to register other images in the group images, and determining a candidate map and a new first iterative image based on each registered image; based on the new first iterative image, returning to execute the step of using the first iterative image to register other images in the group images, and determining a candidate map and a new first iterative image based on each registered image; if the candidate maps in two adjacent iterative processes meet the preset iteration cutoff condition, the candidate map in the current iterative process is determined as the first target map. The use of this method can improve the quality of the constructed map.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a graph construction method, apparatus, computer equipment and storage medium. Background Art

[0002] Brain mapping is a core concept in the field of brain science. Through brain mapping, doctors or researchers can study the patient's brain structure, brain function and brain diseases. This can help them better understand the pathogenesis of brain diseases or discover brain diseases more quickly, so as to provide more accurate treatment for the patient's brain. Therefore, the construction of brain maps is particularly important.

[0003] In related technologies, the brain medical images of each person in a group of people are usually averaged, and the average image is used as the initial brain map; then, the brain medical image of each person can be aligned to the initial brain map, and the aligned brain medical images are further averaged to obtain a new brain map, and then the alignment is continued. After repeating this process several times, the final brain map is constructed.

[0004] However, when the above-mentioned technology is used to construct a brain map, the resulting brain map has low clarity and poor image quality. Summary of the invention

[0005] Based on this, it is necessary to provide a graph construction method, device, computer equipment and storage medium that can improve the quality of the constructed graph in response to the above technical problems.

[0006] A method for constructing a graph, the method comprising:

[0007] Determine an image from the acquired group of images as a first iteration image; the group of images includes images of multiple different objects;

[0008] Using the first iteration image to register other images in the group image, and determining a candidate atlas and a new first iteration image according to each registered image;

[0009] According to the new first iterative image, returning to the step of registering other images in the group image using the first iterative image, and determining the candidate atlas and the new first iterative image according to the registered images;

[0010] If the candidate spectra in two adjacent iterations satisfy a preset iteration cutoff condition, the candidate spectra in the current iteration will be determined as the first target spectra.

[0011] In one embodiment, the step of determining an image from the acquired group of images as the first iterative image includes:

[0012] Calculating similarity values ​​between each image in the group of images and other images, and determining at least one similarity value corresponding to each of the images;

[0013] A first iterative image is determined from the group of images according to at least one similarity value corresponding to each of the images.

[0014] In one embodiment, determining the first iterative image from the group of images according to at least one similarity value corresponding to each of the images comprises:

[0015] Determining the similarity and value corresponding to each of the images according to at least one similarity value corresponding to each of the images;

[0016] According to the similarities and values ​​corresponding to the above images, a first iterative image is determined from the group of images.

[0017] In one embodiment, determining the first iterative image from the group of images according to the similarities and values ​​corresponding to the images includes:

[0018] Sorting the similarities and values ​​corresponding to the above images to obtain a sorting result;

[0019] The image corresponding to the smallest similarity and value in the above sorting results is determined as the first iteration image.

[0020] In one embodiment, the above-mentioned using the first iterative image to register other images in the group image includes:

[0021] Using the first iterative image and a preset registration model to register other images in the group image, and determine the deformation field corresponding to each other image in the group image; wherein the registration model is obtained by training based on multiple real training image pairs, each real training image pair includes a real reference image and a real moving image;

[0022] The corresponding other images are converted according to the deformation fields corresponding to each other image, and the registration images corresponding to each other image are determined.

[0023] In one embodiment, the determining of candidate atlases based on the registered images includes:

[0024] A mean operation is performed on each of the above-mentioned registered images and the first iterative image to determine a candidate atlas.

[0025] In one embodiment, the step of determining a new first iterative image based on the registered images includes:

[0026] Performing a proportional operation on the deformation fields corresponding to the other images to determine the proportional deformation fields corresponding to the other images;

[0027] The above-mentioned converting the corresponding other images according to the deformation field corresponding to each other image to determine the registration image corresponding to each other image includes:

[0028] According to the proportional deformation fields corresponding to the other images, the corresponding other images are converted to determine the proportional registration images corresponding to the other images;

[0029] A new first iteration image is determined according to each of the above-mentioned proportional registration images.

[0030] In one embodiment, the step of determining a new first iterative image according to the registered images at each ratio includes:

[0031] Calculating the similarity value between each of the above-mentioned scale-registered images and other scale-registered images to obtain at least one similarity value corresponding to each scale-registered image;

[0032] A new first iteration image is determined from each other image according to at least one similarity value corresponding to each of the above-mentioned proportionally registered images.

[0033] In one embodiment, the step of determining a new first iteration image from each other image according to at least one similarity value corresponding to each scaled registered image comprises:

[0034] Determine the similarity and value corresponding to each proportional registered image according to at least one similarity value corresponding to each of the proportional registered images;

[0035] According to the similarities and values ​​corresponding to the above-mentioned proportional registration images, a new first iteration image is determined from the other images.

[0036] In one embodiment, the method of determining a new first iteration image from each other image according to the similarities and values ​​corresponding to each scale-registered image includes:

[0037] Sorting the similarities and values ​​corresponding to the above-mentioned proportional registration images to obtain a new sorting result;

[0038] The proportional registration image corresponding to the smallest similarity and value in the new sorting results is determined as a new first iteration image.

[0039] In one embodiment, the group image is a first age group image, and the method further includes:

[0040] Acquire a second age group image and a third age group image; the second age group is lower than the first age group, and the first age group is lower than the third age group;

[0041] According to the similarity values ​​between the first target atlas and each image in the second age group image, an image is determined from the second age group image as a second iteration image, and a second target atlas corresponding to the second age group image is determined according to the second iteration image;

[0042] According to the similarity values ​​between the first target map and each image in the third age group image, an image is determined from the third age group image as a third iterative image, and the third target map corresponding to the third age group image is determined according to the third iterative image.

[0043] In one embodiment, the training method of the preset registration model includes:

[0044] Obtain real training images and standard atlases for multiple objects;

[0045] Using the above standard atlas to register each real training image, and determine the initial registration training image corresponding to each real training image;

[0046] Randomly select two initial registration training images from the above initial registration training images as real training image pairs to obtain multiple real training image pairs;

[0047] The initial registration model is trained based on the multiple real training images to determine the registration model.

[0048] In one of the embodiments, the above iteration cutoff condition is that the similarity value between the candidate graphs in two adjacent iterations is less than a preset similarity threshold.

[0049] A graph construction device, comprising:

[0050] A first iterative image determination module is used to determine an image from the acquired group images as a first iterative image; the group images include images of multiple different objects;

[0051] A registration module, used to register other images in the group image using the first iteration image, and determine a candidate atlas and a new first iteration image according to each registered image;

[0052] An iteration module, used for returning to the step of registering other images in the group image using the first iteration image according to the new first iteration image, and determining a candidate atlas and a new first iteration image according to each registered image;

[0053] The target spectrum determination module is used to determine the candidate spectrum in the current iteration process as the first target spectrum when the candidate spectrum in two adjacent iteration processes meets a preset iteration cutoff condition.

[0054] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0055] Determine an image from the acquired group of images as a first iteration image; the group of images includes images of multiple different objects;

[0056] Using the first iteration image to register other images in the group image, and determining a candidate atlas and a new first iteration image according to each registered image;

[0057] According to the new first iterative image, returning to the step of registering other images in the group image using the first iterative image, and determining the candidate atlas and the new first iterative image according to the registered images;

[0058] If the candidate spectra in two adjacent iterations satisfy a preset iteration cutoff condition, the candidate spectra in the current iteration will be determined as the first target spectra.

[0059] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0060] Determine an image from the acquired group of images as a first iteration image; the group of images includes images of multiple different objects;

[0061] Using the first iteration image to register other images in the group image, and determining a candidate atlas and a new first iteration image according to each registered image;

[0062] According to the new first iterative image, returning to the step of registering other images in the group image using the first iterative image, and determining the candidate atlas and the new first iterative image according to the registered images;

[0063] If the candidate spectra in two adjacent iterations satisfy a preset iteration cutoff condition, the candidate spectra in the current iteration will be determined as the first target spectra.

[0064] The above-mentioned atlas construction method, device, computer equipment and storage medium, by determining a real image from the acquired group images as the first iteration image, using the first iteration image to align other images in the group image, and determining the candidate atlas and the new first iteration image according to each aligned image, and then using the new first iteration image, cyclically executing the steps of aligning and determining the candidate atlas and the new first iteration image, until the candidate atlases in two adjacent iterations meet the iteration cutoff condition, the iteration stops, and the candidate atlas in the current iteration process is determined as the first target atlas; wherein the group image includes multiple real images of different objects. In this method, since when determining the iteration initial image, a real image determined from the real images of each object in the group image is used as the iteration initial image, the real image is relatively clear, real and accurate, so when the initial iteration image is subsequently used for alignment and determination of the new iteration image, the alignment result is more accurate, and the determined new iteration image is also more accurate, so that the accuracy of the target atlas constructed by the iteration process is higher, that is, the quality of the target atlas finally constructed is also the best. In addition, since the constructed target map is constructed based on images of multiple objects (ie, group images), the constructed target map is also more representative. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is an internal structure diagram of a computer device in one embodiment;

[0066] Figure 2 A schematic diagram of a flow chart of a method for constructing a graph in one embodiment;

[0067] Figure 2a A schematic diagram of an image iteration process in atlas construction in one embodiment;

[0068] Figure 3 A schematic diagram of a flow chart of a method for constructing a map in another embodiment;

[0069] Figure 4 A schematic diagram of a flow chart of a method for constructing a map in another embodiment;

[0070] Figure 4a An example diagram of preprocessing an image in another embodiment;

[0071] Figure 4b An example diagram of a network structure of a registration model in another embodiment;

[0072] Figure 4c is an example diagram of a registration result obtained by using a registration model in another embodiment;

[0073] Figure 5A schematic diagram of a flow chart of a method for constructing a map in another embodiment;

[0074] Figure 6 A schematic diagram of a flow chart of a method for constructing a map in another embodiment;

[0075] Figure 6a An example diagram of target map transmission in another embodiment;

[0076] Figure 6b An example diagram of a target map for each age group constructed in another embodiment;

[0077] Figure 7 It is a structural block diagram of a graph construction device in one embodiment. DETAILED DESCRIPTION

[0078] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0079] At present, when constructing a brain map template, two methods are usually used. One is to scan the same person multiple times, and average the multiple brain images obtained to obtain an average brain image as a brain map template; although this method is relatively simple when constructing a brain map template, and a clearer brain map template can be obtained, it only uses the brain image of one person, so the constructed brain map template is not representative. Another method is to scan each person based on a group of people, average the brain images of each person, obtain an initial brain map template, and then align all the brain images of all people to this initial brain map template, and average the results of the alignment to obtain a new template, and repeat this process several times in sequence to obtain the final brain map template; although this method comprehensively considers the group images, the alignment template obtained based on the mean of all images will be very fuzzy, especially in the first few iterations, the alignment template is a fuzzy non-real image, which will result in the final constructed brain map template. The clarity is low, that is, the quality is poor. Based on this, the embodiment of the present application provides a method, device, computer equipment and storage medium for atlas construction, which can solve the above technical problems.

[0080] The atlas construction method provided in the embodiment of the present application can be applied to a medical device, which includes a scanning device and a computer device that are interconnected. The scanning device can scan the detection object and send the obtained scanning data to the computer device, so that the computer device performs data processing, image reconstruction, and atlas construction according to the obtained scanning data.

[0081] The computer device may be a terminal device or a server. Taking the computer device as a terminal device as an example, its internal structure diagram may be as follows: Figure 1 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used 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 operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for constructing a graph is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0082] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0083] It should be noted that the execution subject of the embodiments of the present application can be a medical device, a computer device in a medical device, or a map construction device, and the map construction device can be implemented as part or all of the computer device through software, hardware, or a combination of software and hardware. The following method embodiments are described by taking the execution subject as a computer device as an example.

[0084] In one embodiment, Figure 2 As shown, a method for constructing a graph is provided, which may include the following steps:

[0085] S202, determining an image from the acquired group images as a first iterative image; the group image includes images of multiple different objects.

[0086] The multiple different objects here may be multiple objects of different ages, different genders, and / or different heights. In this embodiment, the group image here may be an image of different objects in the same age group with age as the main consideration, for example, multiple images of different objects of 50±3 years old (i.e., 47-53 years old) may be used.

[0087] In addition, the image here can be an image of any part of the object, for example, it can be a brain image, an abdominal image, a chest image, a stomach image, a lung image, etc. It should be noted that the image of each object in the group image here is an image of the same part of each object, for example, each image in the group image is a brain image of each object. The image of each object here is a real and clear image, so the first iterative image selected therefrom must also be a real and clear image. In addition, the image here can be a CT (computed tomography) image, an MR (magnetic resonance) image, a DR (digital X-ray) image, etc.

[0088] Specifically, the first iteration image here can also be called the initial iteration image. When determining an image from the group of images as the first iteration image, any image in the group of images (such as the first image, the middle image, etc.) can be used as the first iteration image; of course, it can also be analyzed by the similarity between each image in the group of images and other images, and an image with the highest similarity with other images can be selected as the first iteration image; of course, it can also be calculated that the Euclidean distance between each image in the group of images and other images, and an image with the shortest Euclidean distance can be selected as the first iteration image; or, it can also be selected by other methods. In short, an image can be selected from the group as the first iteration image.

[0089] S204, using the first iterative image to register other images in the group image, and determining a candidate atlas and a new first iterative image according to each registered image.

[0090] In this step, after obtaining the initial iterative image, i.e., the first iterative image, the first iterative image can be used to sequentially register the remaining images in the group of images. During the registration, the first iterative image and the other images can be registered using a preset registration model, or the first iterative image and the other images can be registered using a registration method based on mathematical calculations. Of course, other methods can also be used for registration, which is not specifically limited in this embodiment.

[0091] During the registration process, the registered image can be obtained after each other image is registered to the first iteration image. Afterwards, the candidate atlas can be obtained by taking the average of the first iteration image and each other image after registration; or, it can be obtained by obtaining the proportional deformation field between the first iteration image and each other image, registering each other image using the proportional deformation field, and then taking the average of the registered images (including the first iteration image). For the first iteration image, the ratio deformation field between the first iteration image and each other image may be obtained, and each other image may be registered using the ratio deformation field, and then the image with the smallest similarity and value may be taken as the first iteration image for the registered images (including the first iteration image); or, the ratio deformation field between the first iteration image and each other image may be obtained, and each other image may be registered using the ratio deformation field, and then the average value may be taken as the first iteration image for the registered images (including the first iteration image); or, the ratio deformation field between the first iteration image and each other image may be obtained, and each other image may be registered using the ratio deformation field, and then the image with the smallest similarity and value may be taken as the first iteration image for the first several iterations, and then the average value may be taken as the first iteration image for the registered images (including the first iteration image) for the latter several iterations. Of course, other methods may be used to obtain the candidate atlas and the new first iteration image, which is not specifically limited in this embodiment.

[0092] In addition, since the candidate atlas and the new first-iteration image may be different images, two names are used here to distinguish them.

[0093] S206, based on the new first iterative image, return to the step of using the first iterative image to register other images in the group image, and determine the candidate atlas and the new first iterative image based on the registered images.

[0094] In this step, after obtaining a new first iteration image, the new first iteration image can be used to continue to register the remaining images in the group image, and the candidate atlas and the new first iteration image can be determined according to the registration result. The determination process can refer to step S204. After that, step S204 can be executed repeatedly until the cycle or iteration process ends, and the candidate atlas in the current cycle or iteration process is obtained.

[0095] For example, see Figure 2a As shown, an example diagram of an image iteration process is provided. Taking the group image including subjects 1-N as an example, the group image is the brain image of the subject, and the first iteration image is selected by the principle of minimum MSE (mean square error), for example, it needs to be iterated K times, and finally a candidate map can be obtained. It should be noted that Figure 2aIt is just an example and does not affect the essential content of the embodiments of the present application.

[0096] S208: If the candidate graphs in two adjacent iterations satisfy a preset iteration cutoff condition, the candidate graph in the current iteration is determined as the first target graph.

[0097] In this step, optionally, the iteration cutoff condition is that the similarity value between the candidate spectra in two adjacent iterations is less than a preset similarity threshold. That is to say, in the iteration process, a candidate spectra can be obtained in each iteration, and the similarity value between the candidate spectra in each two adjacent iterations can be calculated, and the similarity value calculated each time is compared with the similarity threshold. When the similarity value calculated in a certain time is greater than or equal to the similarity threshold, it means that the candidate spectra in the iteration process corresponding to the calculation does not meet the requirements and needs to continue to iterate; when the similarity value calculated in a certain time is less than the similarity threshold, it means that the candidate spectra in the iteration process corresponding to the calculation is already a candidate spectra that meets the requirements, then the candidate spectra in the iteration process can be used as the first target spectra. That is, the target spectra corresponding to the above-mentioned group images are obtained, and the spectra template of the group of images is constructed.

[0098] In addition, the similarity threshold can be set according to actual conditions, for example, it can be 0.05, 0.1, 0.2, etc.

[0099] In the above-mentioned atlas construction method, a real image is determined from the acquired group images as the first iteration image, the first iteration image is used to align other images in the group images, and the candidate atlas and the new first iteration image are determined according to the aligned images, and then the new first iteration image is used to cyclically execute the steps of alignment and determination of the candidate atlas and the new first iteration image until the candidate atlases in two adjacent iterations meet the iteration cutoff condition, the iteration stops, and the candidate atlas in the current iteration is determined as the first target atlas; wherein the group images include real images of multiple different objects. In this method, since a real image is determined from the real images of each object in the group images as the iteration initial image when determining the iteration initial image, the real image is relatively clear, real and accurate, so when the initial iteration image is subsequently used for alignment and determination of the new iteration image, the alignment result is more accurate, and the determined new iteration image is also more accurate, so that the accuracy of the target atlas constructed by the iteration process is higher, that is, the quality of the target atlas finally constructed is also the best. In addition, since the constructed target map is constructed based on images of multiple objects (ie, group images), the constructed target map is also more representative.

[0100] In another embodiment, Figure 3As shown, another method for constructing a graph is provided. Based on the above embodiment, the above S202 may include the following steps:

[0101] S302, calculating the similarity value between each image in the group of images and other images, and determining at least one similarity value corresponding to each of the images.

[0102] The similarity value here can be calculated by MSE (mean square error), or by Euclidean distance. Of course, other methods can also be used for calculation. The following embodiments of this application mainly introduce the calculation of similarity value by MSE as an example.

[0103] Specifically, after obtaining the group image, the mean square error between each image in the group image and other images may be calculated to obtain at least one mean square error corresponding to each image, that is, to obtain at least one similarity value corresponding to each image.

[0104] For example, assuming that there are 5 images in the group image, namely image 1, image 2, image 3, image 4, and image 5, for image 1, the mean square error between it and images 2, 3, 4, and 5 can be calculated to obtain 4 mean square errors; for images 2-5, the calculation can also be performed in this way to obtain their respective 4 mean square errors.

[0105] S304: Determine a first iterative image from the group of images according to at least one similarity value corresponding to each of the images.

[0106] In this step, after obtaining at least one similarity value corresponding to each image, optionally, the first iterative image may be determined in the following steps A1-A2:

[0107] Step A1, determining the similarity and value corresponding to each of the above images according to at least one similarity value corresponding to each of the above images.

[0108] Here, at least one similarity value corresponding to each image can be summed to obtain the similarity and value corresponding to each image. Taking the five images in S302 as an example, the four mean square errors of each image can be summed to obtain the mean square error sum of the five images, that is, each image obtains a similarity value, and a total of five similarities and values ​​are obtained.

[0109] Here, if each image has only one similarity value, then the corresponding similarity value is the one similarity value.

[0110] Step A2: determining a first iterative image from the group of images according to the similarities and values ​​corresponding to the above images.

[0111] When determining the first iterative image by the similarities and values ​​in this step, optionally, the similarities and values ​​corresponding to the above images are sorted to obtain a sorting result; and the image corresponding to the smallest similarity and value in the above sorting result is determined as the first iterative image.

[0112] Here, when sorting the similarities and values, they can be sorted from large to small or from small to large. When sorting from large to small, the last similarity and value in the sorting result is the smallest similarity and value; when sorting from small to large, the first similarity and value in the sorting result is the smallest similarity and value.

[0113] In short, after the minimum similarity and value are determined, the image corresponding to the minimum similarity and value can be determined as the first iteration image.

[0114] In this embodiment, by calculating the similarity values ​​between each image in the group image, and determining the first iterative image from the calculated similarity values, this selection method is relatively simple and accurate, so the efficiency and accuracy of selecting the first iterative image can be improved. In addition, the first iterative image is determined by the similarity and value of each image, so that the similarity between each image and each other image can be comprehensively considered, so that the determined first iterative image is more accurate. Further, by sorting the similarities and values, and determining the image corresponding to the smallest similarity and value in the sorting result as the first iterative image, the first iterative image selected in this way is the most similar to other images, so that the accuracy of the selected first iterative image can be further improved, and then the subsequent iteration process through the first iterative image can be made more accurate, and the target map finally obtained is also more accurate.

[0115] In another embodiment, Figure 4 As shown, another method for constructing a graph is provided. Based on the above embodiment, the above S204 may include the following steps:

[0116] S402, using the above-mentioned first iterative image and a preset registration model to align other images in the group image, and determine the deformation field corresponding to each other image in the group image; wherein the above-mentioned registration model is obtained by training based on multiple real training image pairs, and each real training image pair includes a real reference image and a real moving image.

[0117] In this step, the registration model can be a rigid registration model or a nonlinear registration model. The rigid registration model mainly performs rapid registration of the general shape in the image, and the nonlinear registration model mainly performs fine registration of the details in the image. Before using the registration model for registration, the registration model can be trained in advance. Optionally, the training method of the registration model can include the following steps B1-B4:

[0118] Step B1, obtaining real training images and standard atlases of multiple objects.

[0119] Step B2: register each real training image using the standard atlas to determine an initial registered training image corresponding to each real training image.

[0120] Step B3: arbitrarily select two initial registration training images from the above initial registration training images as real training image pairs to obtain multiple real training image pairs.

[0121] Step B4: training the initial registration model based on the above-mentioned multiple real training images to determine the registration model.

[0122] In steps B1-B4, see Figure 4a As shown, after obtaining real training images and standard atlases (here the standard atlases may be MNI152 images) of multiple objects, each real image may be preprocessed. The real training images here are images obtained by scanning the real objects.

[0123] Taking the real training image obtained as a brain MR image as an example, the multiple brain MR images obtained here are all acquired based on the same 3DT1 magnetic resonance sequence. Therefore, the multiple MR images obtained are kept in the same coordinate system and origin coordinates, and the spatial resolution is the same, all of which are 1×1×1mm. 3 All acquired brain MR images were preprocessed as follows: first, the original images were corrected for deviation fields to eliminate the uneven grayscale of the magnetic resonance images; then, the neck and skull of each image were removed to avoid interference of the neck, skull and other skull images on the registration task.

[0124] Afterwards, the MNI152 image can be used to register each real training image, and rigid body registration with 6 degrees of freedom (including rotation and translation in the three directions of X, Y, and Z) can be performed. Afterwards, using MNI152 as a template, the grayscale values ​​of each registered real training image are histogram matched to obtain the initial registration training image corresponding to each real training image. Finally, two initial registration training images are arbitrarily selected from each initial registration training image as a real training image pair, and the real training image pair includes a reference image and a moving image, so that multiple real training image pairs can be obtained. Afterwards, the reference image in each real training image pair can be used as a template to perform rigid body registration with 9 degrees of freedom (including rotation, translation, and scaling in the three directions of X, Y, and Z) on the moving image in each training image pair, and finally a large number of paired, real reference images and moving images are obtained. The real training image pairs, which are composed of moving images, can be trained for the registration model.

[0125] Of course, before training, cropping operations can also be performed on each image in the real training image pair. The patch size of the registration model input can be, for example, [100, 100, 100], and its grayscale value is normalized, for example, to [0, 255]. This can save the computing power of the computer device and improve the computing power of the computer device.

[0126] For example, assuming that there are 600 real training images of an object, and 520 pairs of real training images need to be formed, then after preprocessing the 600 real training images, 2 images can be selected each time as a real training image pair. The operation is performed 520 times, and finally 520 pairs of real training image pairs are obtained. It should be noted that the two images included in each real training image pair are not images of the same object, that is, they are images of two objects. At the same time, the real training images of these 600 objects include training images of objects of different age groups, but when extracting 520 pairs of real training image pairs, generally, the proportion of the real training image pairs with a relatively small age difference between the two images is high, and the proportion of the real training image pairs with a relatively large age difference between the two images is low.

[0127] After preprocessing the real training images, the registration model can be trained. During training, see Figure 4b As shown in the figure, a network structure diagram of a registration model is provided. The registration model (Reg3D-Deform-Net) consists of two downsampling feature extraction modules and two upsampling image mapping modules. The input module is responsible for extracting primary features, and the output module is responsible for feature fusion of the mapped feature maps. In the main network structure, each downsampling module contains two convolution operations with kernel size = 3 and stride = 1 and one downsampling operation with kernel size = 2 and stride = 2 to complete the feature extraction of the image. Each upsampling module contains a transposed convolution operation with kernel size = 2 and stride = 2 and two convolution operations with kernel size = 3 and stride = 1 to complete the mapping of the image. A skip connection is constructed between the downsampling and upsampling modules of the network structure, and the image of the downsampling module is made to have the same size as the image of the upsampling module through cropping operation.

[0128] Specifically, the loss between the output image of the registration model and the reference image, the loss of the deformation field, etc. can be calculated as the total loss to train the initial registration model. At the same time, the Adam optimizer that can adaptively adjust the learning rate is used to optimize the network parameters, and finally the non-rigid registration network model is trained. The downsampling segment uses a 3×3×3 convolution kernel to extract more global abstract features of the image as the number of layers increases. At the same time, the convolution operation with stride=2 makes the image resolution halved layer by layer, and the number of feature maps doubled layer by layer. The upsampling segment uses transposed convolution to improve the resolution of feature mapping, and uses a 3×3×3 convolution kernel to establish the mapping relationship between the original input image and the output image. Skip-layer connections are constructed in the downsampling and upsampling segments of the network to avoid the problem of gradient disappearance during training, and to add auxiliary information to the image mapping. Finally, a trained registration model is obtained.

[0129] The above-mentioned output image means that after each pair of real training images is input into the initial registration model, a predicted deformation field will be output. The moving image in the real training image pair is transformed by the predicted deformation field, and the obtained image can be recorded as the output image.

[0130] When calculating the total loss, the MSE (mean square error) loss between the output image and the reference image corresponding to each real training image pair and the Gradient loss of the predicted deformation field corresponding to each real training image pair can be calculated, and the weighted sum of these two types of losses can be used to obtain the total loss of each real training image pair.

[0131] Of course, the total loss can also be the weighted sum of the NCC (correlation) loss between the output image and the reference image corresponding to each real training image pair and the Gradient (gradient) loss of the predicted deformation field corresponding to each real training image pair; or, the total loss can also be the weighted sum of the MI (mutual information) loss between the output image and the reference image corresponding to each real training image pair and the Gradient (gradient) loss of the predicted deformation field corresponding to each real training image pair; or, the total loss can also be the MSE (mean square error) loss between the output image and the reference image corresponding to each real training image pair, the Gradient (gradient) loss of the predicted deformation field corresponding to each real training image pair, and the JacobAntiFoldi ng (Jacob) loss; or, the total loss may also be the weighted sum of the NCC (correlation) loss between the output image corresponding to each real training image pair and the reference image, the Gradient (gradient) loss of the predicted deformation field corresponding to each real training image pair, and the JacobAntiFolding (Jacob) loss; or, the total loss may also be the weighted sum of the MI (mutual information) loss between the output image corresponding to each real training image pair and the reference image, the Gradient (gradient) loss of the predicted deformation field corresponding to each real training image pair, and the JacobAntiFolding (Jacob) loss; or, it may also be other forms of losses, which are not specifically limited in this embodiment.

[0132] Of course, during the training process of the above registration model, the tissue segmentation map of each real training image can also be integrated into the registration model for training.

[0133] After the registration model is trained, the trained registration model can be used to register the first iteration image and each other image in turn to obtain the corresponding deformation field between each other image and the first iteration image, that is, each other image will obtain a deformation field between it and the first iteration image. The registration results can be seen in Figure 4c As shown, Figure 4c The left image in the figure is a moving image, the middle image is a reference target image, and the right image is the output image registered by the registration model.

[0134] It should be noted that the above Figure 4a , Figure 4b , Figure 4c It is just an example and does not affect the essential content of the embodiments of the present application.

[0135] S404: Perform conversion processing on the corresponding other images according to the deformation field corresponding to each other image, and determine the registration image corresponding to each other image.

[0136] In this step, after obtaining the deformation field between each other image and the first iterative image, each other image can be directly transformed using the deformation field corresponding to each other image to obtain other images that are completely aligned; of course, it is also possible to perform proportional calculations on the deformation field corresponding to each other image, and use the obtained proportional deformation field to transform each other image to obtain other images that are proportionally aligned.

[0137] After obtaining the registered images corresponding to each other image, the candidate atlas can be determined. Optionally, determining the candidate atlas based on each registered image in S204 may include: performing a mean operation on each of the registered images and the first iterative image to determine the candidate atlas.

[0138] Here, the candidate map can be obtained by taking the average of other images that are fully registered and the first iterative image, or by taking the average of other images that are proportionally registered and the first iterative image. In this embodiment, the candidate map is mainly obtained by taking the average of other images that are fully registered and the first iterative image, so that the candidate map obtained in each iteration process is more concentrated and clearer.

[0139] Similarly, after obtaining the registered images corresponding to each other image, a new first iterative image can be determined. Optionally, determining the new first iterative image based on the registered images in S204 may include: performing proportional operations on the deformation fields corresponding to each other image to determine the proportional deformation fields corresponding to each other image; accordingly, S404 may include: performing conversion processing on the corresponding other images according to the proportional deformation fields corresponding to each other image to determine the proportional registered images corresponding to each other image; and then determining the new first iterative image based on the proportional registered images.

[0140] The proportional calculation here can be performed according to a preset ratio, and the preset ratio can be set according to actual conditions, for example, it can be 1 / 5, 1 / 4, 1 / 3, etc. Then, after obtaining the deformation fields corresponding to each other image, each deformation field can be processed according to the preset ratio to obtain each proportional deformation field, and each other image can be registered through each proportional deformation field, and the obtained registered image is recorded as the proportional registered image. After that, a new first iterative image can be determined through each proportional registered image. The method for determining the new first iterative image here can be the same as the method for determining the first iterative image in S202 above. In addition, by selecting a new first iterative image through each proportional registered image here, the distance between the iterative image and the candidate atlas in the iterative process can be refined, so that the candidate atlas obtained in the final iterative process is more accurate and true.

[0141] In this embodiment, the above-mentioned first iterative image and the preset registration model are used to register other images in the group image, the deformation field corresponding to each other image in the group image is determined, and the deformation field is used to register each other image to obtain a registered image. Here, the registration model is used for image registration, which can efficiently complete one-to-one image registration, improve the efficiency of image registration and the accuracy of the registration result. In addition, by training the registration model, the registration process using the registration model can be made more accurate. Further, the average of the registered image and the first iterative image is taken to determine the candidate atlas, which can improve the efficiency and accuracy of obtaining the candidate atlas. Furthermore, the new first iterative image is determined using the proportional registration deformation field and the proportional registration image, which can refine the distance between the iterative image and the candidate atlas during the iteration process, so that the candidate atlas obtained in the final iteration process is more accurate and true.

[0142] In another embodiment, Figure 5 As shown, another atlas construction method is provided. Based on the above embodiment, determining a new first iterative image according to each scale registration image in the above specific step S404 may include the following steps:

[0143] S502: Calculate the similarity value between each of the above-mentioned proportionally registered images and other proportionally registered images to obtain at least one similarity value corresponding to each of the proportionally registered images.

[0144] The explanation of this step can refer to the explanation of S302 above, which will not be repeated here.

[0145] S504: Determine a new first iteration image from each other image according to at least one similarity value corresponding to each of the above-mentioned proportional registration images.

[0146] Similarly, optionally, the similarity and value corresponding to each of the scaled registered images can be determined according to at least one similarity value corresponding to each of the scaled registered images; a new first iterative image can be determined from each of the other images according to the similarity and value corresponding to each of the scaled registered images. Optionally, the similarities and values ​​corresponding to each of the scaled registered images can be sorted to obtain a new sorting result; the scaled registered image corresponding to the smallest similarity and value in the new sorting result is determined as the new first iterative image.

[0147] The explanation here can also refer to the explanation of S304 above, which will not be repeated here.

[0148] Of course, in the iterative process, the first several iterations may adopt the principle of similarity and minimum value to calculate a new first iteration image based on the scaled registered image; the next several iterations may calculate a new first iteration image by averaging the scaled registered images of this iteration; for example, there are 100 iterations in total, the first 80 iterations may adopt the principle of similarity and minimum value to calculate a new first iteration image based on 1 / 5 scaled registered images, and the next 20 iterations may calculate a new first iteration image by averaging 1 / 5 scaled registered images in each iteration. Of course, other methods are also possible, which are not specifically limited here.

[0149] In this embodiment, by calculating the similarity values ​​between the registered images of each ratio, and determining the new first iterative image therefrom by the calculated similarity values, such a selection method is relatively simple and accurate, and thus the efficiency and accuracy of selecting the new first iterative image can be improved. In addition, by determining the new first iterative image by the similarities and values ​​of the registered images of each ratio, the similarities between the registered images of each ratio can be comprehensively considered, so that the determined new first iterative image is more accurate. Furthermore, by sorting the similarities and values, and determining the image corresponding to the smallest similarity and value in the new sorting result as the new first iterative image, the selected new first iterative image is the most similar to each other image, so that the accuracy of the selected new first iterative image can be further improved, and then the subsequent iteration process through the new first iterative image can be made more accurate, and the target atlas finally obtained is also more accurate.

[0150] In another embodiment, Figure 6 As shown, another atlas construction method is provided. Based on the above embodiment, the above group image is a first age group image. The above method may further include the following steps:

[0151] S602, obtaining a second age group group image and a third age group group image; the second age group is lower than the first age group, and the first age group is lower than the third age group.

[0152] In the process of actually applying the target map, it is often necessary to construct target maps for different age groups. Therefore, this embodiment mainly describes the construction of target maps for different age groups.

[0153] After obtaining the images of the plurality of subjects, the images of the respective subjects are grouped according to their ages, that is, obtaining the first age group image, the second age group image, and the third age group image. Figure 6aAs shown in the figure, if the span is from 20 to 80 years old, every 5 years is divided into a group, and the target map is constructed for each group of images. In order to ensure the time continuity of the longitudinal data structure, the middle age group (here 50 years old) is the center, and the target map is transmitted to both sides by the middle age group according to the principle of highest similarity.

[0154] The first age group here is the central age group. Generally, a target map for this central age group will be pre-built. The construction process can be found in the above Figure 2-Figure 5 Here, the second age group is the age group located on the left side of the first age group, and the third age group is the age group located on the right side of the first age group.

[0155] S604, determining an image from the second age group image as a second iterative image based on the similarity values ​​between the first target map and each image in the second age group image, and determining a second target map corresponding to the second age group image based on the second iterative image.

[0156] In this step, after constructing the first target map corresponding to the first age group (i.e., the central age group), the similarity between the first target map and each image in the second age group group image can be calculated, and the image with the highest similarity to the first target map is selected from the similarity results as the initial iteration image of the second age group group image, which is recorded as the second iteration image. The second iteration image here is similar to the first iteration image in the first age group group image. Afterwards, the above method can be used to Figure 2-Figure 5 In this way, a corresponding target map is constructed for the second age group image, that is, a corresponding second target map is obtained.

[0157] S606, determining an image from the third age group image as a third iterative image based on the similarity values ​​between the first target map and each image in the third age group image, and determining a third target map corresponding to the third age group image based on the third iterative image.

[0158] Similarly, in this step, after constructing the first target map corresponding to the first age group (i.e., the central age group), the similarity between the first target map and each image in the third age group group image can be calculated, and the image with the highest similarity to the first target map can be selected from the similarity results as the initial iteration image of the third age group group image, which is recorded as the third iteration image. The third iteration image here is similar to the first iteration image in the first age group group image. Afterwards, the above method can be used. Figure 2-Figure 5 In this way, a corresponding target map is constructed for the third age group image, that is, a corresponding third target map is obtained.

[0159] For example, continue with the above span from 20 to 80 years old, divide every 5 years into a group, and construct the target atlas for each group of images as an example. Then, you can first construct a target atlas for 50 years old, and pass the 50-year-old target atlas to the 45-year-old on the left and the 55-year-old on the right, to obtain the target atlas for 45 years old and the target atlas for 55 years old. Then, you can continue to pass the 45-year-old target atlas to the 40-year-old on the left, and continue to pass the 55-year-old target atlas to the 60-year-old on the right. Repeat these processes in sequence, and finally obtain the target atlas for each age group between 20 and 80 years old. The target atlas finally obtained can be seen in Figure 6b shown.

[0160] In this embodiment, the target map constructed for the central age group is transferred to the group images of other age groups on both sides, so that when constructing the target maps of other surrounding age groups, the target maps of adjacent age groups can be combined. This ensures the time continuity of the constructed target map in the vertical direction, so that the target maps of each age group finally constructed have higher accuracy and a wider range of applicability.

[0161] It should be understood that although Figure 2 , 3 The steps in the flowcharts of , 4, 5, and 6 are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2 , 3 At least part of the steps in 4, 5, and 6 may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily one by one, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0162] In one embodiment, Figure 7 As shown, a map construction device is provided, comprising: a first iterative image determination module 10, a registration module 11, an iteration module 12 and a target map determination module 13, wherein:

[0163] A first iterative image determination module 10 is used to determine an image from the acquired group images as a first iterative image; the group images include images of multiple different objects;

[0164] A registration module 11, used to register other images in the group image using the first iteration image, and determine a candidate atlas and a new first iteration image according to each registered image;

[0165] The iteration module 12 is used to return to the step of registering other images in the group image using the first iteration image according to the new first iteration image, and determining the candidate atlas and the new first iteration image according to the registered images;

[0166] The target spectrum determination module 13 is used to determine the candidate spectrum in the current iteration process as the first target spectrum when the candidate spectrum in two adjacent iteration processes meets a preset iteration cutoff condition.

[0167] Optionally, the above iteration cutoff condition is that the similarity value between the candidate graphs in two adjacent iterations is less than a preset similarity threshold.

[0168] For the specific limitations on the atlas construction device, please refer to the limitations on the atlas construction method above, which will not be repeated here.

[0169] In another embodiment, another atlas construction device is provided. Based on the above embodiment, the first iterative image determination module 10 may include:

[0170] A calculation unit, used to calculate the similarity value between each image in the group of images and other images, and determine at least one similarity value corresponding to each of the images;

[0171] The first iterative image determining unit is used to determine a first iterative image from the group of images according to at least one similarity value corresponding to each of the images.

[0172] Optionally, the first iterative image determining unit may include:

[0173] a similarity and value determination subunit, configured to determine the similarity and value corresponding to each of the images according to at least one similarity value corresponding to each of the images;

[0174] The first iterative image determining subunit is used to determine a first iterative image from the group of images according to the similarities and values ​​corresponding to the above images.

[0175] Optionally, the first iterative image determination subunit is specifically used to sort the similarities and values ​​corresponding to the above images to obtain a sorting result; and determine the image corresponding to the smallest similarity and value in the sorting result as the first iterative image.

[0176] In another embodiment, another atlas construction device is provided. Based on the above embodiment, the registration module 11 may include:

[0177] A registration unit, used to register other images in the group of images using the first iterative image and a preset registration model, and determine the deformation field corresponding to each other image in the group of images; wherein the registration model is obtained by training based on a plurality of real training image pairs, each real training image pair includes a real reference image and a real moving image;

[0178] The conversion unit is used to perform conversion processing on the corresponding other images according to the deformation field corresponding to each other image, and determine the registration image corresponding to each other image.

[0179] Optionally, the above device may further include a model training module, the model training module including:

[0180] A training image acquisition unit, used to acquire real training images and standard atlases of multiple objects;

[0181] A training image registration unit, used to register each real training image using the above standard atlas, and determine an initial registration training image corresponding to each real training image;

[0182] An image pair selection unit, used to select any two initial registration training images from the above initial registration training images as real training image pairs to obtain multiple real training image pairs;

[0183] The training unit is used to train the initial registration model based on the multiple real training images to determine the registration model.

[0184] Optionally, the registration module 11 may further include:

[0185] The candidate atlas determination unit is used to perform a mean operation on each of the above-mentioned registered images and the first iterative image to determine the candidate atlas.

[0186] Optionally, the registration module 11 may further include:

[0187] A proportional calculation unit, used for performing proportional calculation on the deformation fields corresponding to the other images to determine the proportional deformation fields corresponding to the other images;

[0188] Accordingly, the conversion unit is specifically used to perform conversion processing on the corresponding other images according to the proportional deformation field corresponding to each of the other images, and determine the proportional registration image corresponding to each of the other images;

[0189] The new first iterative image determining unit is used to determine a new first iterative image according to each of the above-mentioned proportional registration images.

[0190] In another embodiment, another atlas construction device is provided. Based on the above embodiment, the new first iterative image determination unit may include:

[0191] A similarity value calculation subunit is used to calculate the similarity value between each of the above-mentioned proportional registration images and other proportional registration images, and obtain at least one similarity value corresponding to each proportional registration image;

[0192] The determination subunit is used to determine a new first iterative image from each other image according to at least one similarity value corresponding to each of the above-mentioned proportional registration images.

[0193] Optionally, the above-mentioned determination subunit is specifically used to determine the similarity and value corresponding to each of the above-mentioned proportionally registered images according to at least one similarity value corresponding to each of the above-mentioned proportionally registered images; and determine a new first iterative image from each other image according to the similarity and value corresponding to each of the above-mentioned proportionally registered images.

[0194] Optionally, the determination subunit is specifically used to sort the similarities and values ​​corresponding to the above-mentioned proportional registration images to obtain a new sorting result; and determine the proportional registration image corresponding to the smallest similarity and value in the above-mentioned new sorting result as a new first iterative image.

[0195] In another embodiment, another atlas construction device is provided. Based on the above embodiment, the group image is a first age group image, and the device may further include:

[0196] A multi-group image acquisition module, used to acquire a second age group image and a third age group image; the second age group is lower than the first age group, and the first age group is lower than the third age group;

[0197] a second target map determining unit, configured to determine an image from the second age group image as a second iteration image according to similarity values ​​between the first target map and each image in the second age group image, and determine a second target map corresponding to the second age group image according to the second iteration image;

[0198] A third target map determining unit is used to determine an image from the third age group group image as a third iterative image based on the similarity values ​​between the first target map and each image in the third age group group image, and determine a third target map corresponding to the third age group group image based on the third iterative image.

[0199] For the specific limitations on the atlas construction device, please refer to the limitations on the atlas construction method above, which will not be repeated here.

[0200] Each module in the above-mentioned graph construction device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0201] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0202] An image is determined from the acquired group images as the first iteration image; the group image includes images of multiple different objects; the first iteration image is used to align other images in the group image, and a candidate map and a new first iteration image are determined according to each of the images after the alignment; according to the new first iteration image, the step of using the first iteration image to align other images in the group image, and determining a candidate map and a new first iteration image according to each of the images after the alignment is returned to be executed; if the candidate maps in two adjacent iteration processes meet a preset iteration cutoff condition, the candidate map in the current iteration process is determined as the first target map.

[0203] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0204] The similarity value between each image in the group of images and other images is calculated to determine at least one similarity value corresponding to each of the images; and the first iterative image is determined from the group of images according to the at least one similarity value corresponding to each of the images.

[0205] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0206] According to at least one similarity value corresponding to each of the above images, the similarity and value corresponding to each of the above images are determined; according to the similarity and value corresponding to each of the above images, a first iterative image is determined from the group of images.

[0207] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0208] The similarities and values ​​corresponding to the above images are sorted to obtain a sorting result; and the image corresponding to the smallest similarity and value in the sorting result is determined as the first iterative image.

[0209] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0210] The first iterative image and the preset registration model are used to register other images in the group image to determine the deformation field corresponding to each other image in the group image; wherein the registration model is obtained by training based on multiple real training image pairs, each real training image pair includes a real reference image and a real moving image; according to the deformation field corresponding to each other image, the corresponding other images are converted to determine the registration image corresponding to each other image.

[0211] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0212] A mean operation is performed on each of the above-mentioned registered images and the first iterative image to determine a candidate atlas.

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

[0214] Performing proportional operations on the deformation fields corresponding to the other images to determine the proportional deformation fields corresponding to the other images; performing conversion processing on the corresponding other images according to the proportional deformation fields corresponding to the other images to determine the proportional registration images corresponding to the other images; and determining a new first iterative image according to the proportional registration images.

[0215] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0216] Calculate the similarity value between each of the above-mentioned scale-registered images and other scale-registered images to obtain at least one similarity value corresponding to each of the scale-registered images; and determine a new first iterative image from each of the other images according to the at least one similarity value corresponding to each of the above-mentioned scale-registered images.

[0217] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0218] According to at least one similarity value corresponding to each of the above-mentioned proportionally registered images, the similarity and value corresponding to each of the proportionally registered images are determined; according to the similarity and value corresponding to each of the above-mentioned proportionally registered images, a new first iterative image is determined from each of the other images.

[0219] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0220] The similarities and values ​​corresponding to the above-mentioned proportional registration images are sorted to obtain a new sorting result; and the proportional registration image corresponding to the smallest similarity and value in the above-mentioned new sorting result is determined as a new first iteration image.

[0221] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0222] A second age group group image and a third age group group image are obtained; the second age group is lower than the first age group, and the first age group is lower than the third age group; according to the similarity value between the first target map and each image in the second age group group image, an image is determined from the second age group group image as a second iteration image, and a second target map corresponding to the second age group group image is determined according to the second iteration image; according to the similarity value between the first target map and each image in the third age group group image, an image is determined from the third age group group image as a third iteration image, and a third target map corresponding to the third age group group image is determined according to the third iteration image.

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

[0224] Acquire real training images and standard atlases of multiple objects; use the above standard atlas to align each real training image to determine the initial registration training image corresponding to each real training image; arbitrarily select two initial registration training images from each of the above initial registration training images as real training image pairs to obtain multiple real training image pairs; train the initial registration model based on the above multiple real training images to determine the registration model.

[0225] In one embodiment, the above iteration cutoff condition is that the similarity value between the candidate graphs in two adjacent iterations is less than a preset similarity threshold.

[0226] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0227] An image is determined from the acquired group images as the first iteration image; the group image includes images of multiple different objects; the first iteration image is used to align other images in the group image, and a candidate map and a new first iteration image are determined according to each of the images after the alignment; according to the new first iteration image, the step of using the first iteration image to align other images in the group image, and determining a candidate map and a new first iteration image according to each of the images after the alignment is returned to be executed; if the candidate maps in two adjacent iteration processes meet a preset iteration cutoff condition, the candidate map in the current iteration process is determined as the first target map.

[0228] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0229] The similarity value between each image in the group of images and other images is calculated to determine at least one similarity value corresponding to each of the images; and the first iterative image is determined from the group of images according to the at least one similarity value corresponding to each of the images.

[0230] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0231] According to at least one similarity value corresponding to each of the above images, the similarity and value corresponding to each of the above images are determined; according to the similarity and value corresponding to each of the above images, a first iterative image is determined from the group of images.

[0232] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0233] The similarities and values ​​corresponding to the above images are sorted to obtain a sorting result; and the image corresponding to the smallest similarity and value in the sorting result is determined as the first iterative image.

[0234] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0235] The first iterative image and the preset registration model are used to register other images in the group image to determine the deformation field corresponding to each other image in the group image; wherein the registration model is obtained by training based on multiple real training image pairs, each real training image pair includes a real reference image and a real moving image; according to the deformation field corresponding to each other image, the corresponding other images are converted to determine the registration image corresponding to each other image.

[0236] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0237] A mean operation is performed on each of the above-mentioned registered images and the first iterative image to determine a candidate atlas.

[0238] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0239] Performing proportional operations on the deformation fields corresponding to the other images to determine the proportional deformation fields corresponding to the other images; performing conversion processing on the corresponding other images according to the proportional deformation fields corresponding to the other images to determine the proportional registration images corresponding to the other images; and determining a new first iterative image according to the proportional registration images.

[0240] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0241] Calculate the similarity value between each of the above-mentioned scale-registered images and other scale-registered images to obtain at least one similarity value corresponding to each of the scale-registered images; and determine a new first iterative image from each of the other images according to the at least one similarity value corresponding to each of the above-mentioned scale-registered images.

[0242] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0243] According to at least one similarity value corresponding to each of the above-mentioned proportionally registered images, the similarity and value corresponding to each of the proportionally registered images are determined; according to the similarity and value corresponding to each of the above-mentioned proportionally registered images, a new first iterative image is determined from each of the other images.

[0244] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0245] The similarities and values ​​corresponding to the above-mentioned proportional registration images are sorted to obtain a new sorting result; and the proportional registration image corresponding to the smallest similarity and value in the above-mentioned new sorting result is determined as a new first iteration image.

[0246] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0247] A second age group group image and a third age group group image are obtained; the second age group is lower than the first age group, and the first age group is lower than the third age group; according to the similarity value between the first target map and each image in the second age group group image, an image is determined from the second age group group image as a second iteration image, and a second target map corresponding to the second age group group image is determined according to the second iteration image; according to the similarity value between the first target map and each image in the third age group group image, an image is determined from the third age group group image as a third iteration image, and a third target map corresponding to the third age group group image is determined according to the third iteration image.

[0248] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0249] Acquire real training images and standard atlases of multiple objects; use the above standard atlas to align each real training image to determine the initial registration training image corresponding to each real training image; arbitrarily select two initial registration training images from each of the above initial registration training images as real training image pairs to obtain multiple real training image pairs; train the initial registration model based on the above multiple real training images to determine the registration model.

[0250] In one embodiment, the above iteration cutoff condition is that the similarity value between the candidate graphs in two adjacent iterations is less than a preset similarity threshold.

[0251] Those of ordinary skill 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, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0252] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0253] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for constructing a graph, characterized in that: The method comprises: Determine an image from the acquired group of images as a first iteration image; the group of images includes images of multiple different objects; Using the first iterative image to register other images in the group of images, and performing a mean operation on each registered image and the first iterative image to determine a candidate atlas, and determining a new first iterative image according to each registered image; According to the new first iterative image, returning to the step of using the first iterative image to register other images in the group of images, performing a mean operation on each registered image and the first iterative image to determine a candidate atlas, and determining a new first iterative image according to each registered image; If the candidate spectra in two adjacent iterations satisfy a preset iteration cutoff condition, the candidate spectra in the current iteration will be determined as the first target spectra.

2. The method according to claim 1, characterized in that The step of determining an image from the acquired group images as a first iterative image comprises: Calculating similarity values ​​between each image in the group of images and other images, and determining at least one similarity value corresponding to each of the images; The first iterative image is determined from the group of images according to at least one similarity value corresponding to each of the images.

3. The method according to any one of claims 1 to 2, characterized in that: The using the first iterative image to register other images in the group of images includes: Using the first iterative image and a preset registration model to register other images in the group of images, and determine the deformation field corresponding to each other image in the group of images; wherein the registration model is obtained by training based on multiple real training image pairs, each real training image pair includes a real reference image and a real moving image; The corresponding other images are converted according to the deformation fields corresponding to each other image, and the registration images corresponding to each other image are determined.

4. The method according to claim 3, characterized in that The step of determining the candidate atlas according to the registered images includes: A mean operation is performed on each of the registered images and the first iterative image to determine the candidate atlas.

5. The method according to claim 3, characterized in that: The step of determining a new first iterative image according to each registered image comprises: Performing a proportional operation on the deformation field corresponding to each of the other images to determine the proportional deformation field corresponding to each of the other images; The converting process is performed on the corresponding other images according to the deformation field corresponding to each other image to determine the registration image corresponding to each other image, including: Performing conversion processing on the corresponding other images according to the proportional deformation fields corresponding to the other images, and determining proportional registration images corresponding to the other images; A new first iteration image is determined according to each of the scale-registered images.

6. The method according to claim 5, characterized in that Determining a new first iterative image according to each of the proportionally registered images comprises: Calculating a similarity value between each of the scale-registered images and other scale-registered images to obtain at least one similarity value corresponding to each of the scale-registered images; The new first iteration image is determined from each of the other images according to at least one similarity value corresponding to each of the scale-registered images.

7. The method according to any one of claims 1-2, characterized in that: The group image is a first age group image, and the method further includes: Acquire a second age group image and a third age group image; the second age group is lower than the first age group, and the first age group is lower than the third age group; Determining an image from the second age group image as a second iteration image according to similarity values ​​between the first target atlas and each image in the second age group image, and determining a second target atlas corresponding to the second age group image according to the second iteration image; According to the similarity values ​​between the first target atlas and each image in the third age group image, an image is determined from the third age group image as a third iterative image, and a third target atlas corresponding to the third age group image is determined according to the third iterative image.

8. The method according to claim 3, characterized in that The training method of the preset registration model includes: Obtain real training images and standard atlases for multiple objects; Using the standard atlas to register each of the real training images, and determining an initial registration training image corresponding to each of the real training images; Randomly selecting two initial registration training images from each of the initial registration training images as real training image pairs to obtain the multiple real training image pairs; The initial registration model is trained based on the multiple real training images to determine the registration model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images