Artificial Intelligence-Based Image Search Method, Device, Equipment and Medium

Abstract

The present invention discloses an image search method, device, equipment and medium based on artificial intelligence. The image search method based on artificial intelligence includes: obtaining an image search request, where the image search request includes a target user identifier, an original CT image corresponding to the target user identifier, and an original delineation drawing of a critical organ; inputting the original CT image and the original delineation drawing of the critical organ into a dose analysis model to generate an original dose distribution map corresponding to the target user identifier; performing registration processing on the original dose distribution map to obtain a standard dose distribution map; inputting the standard dose distribution map into an image search model to obtain a target feature vector corresponding to the target user identifier; querying a radiotherapy plan database based on the target feature vector to obtain a target dose distribution map that matches the target feature vector. The image search method based on artificial intelligence can improve the acquisition efficiency and accuracy of the target dose distribution map.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and particularly to an image search method, device, equipment and medium based on artificial intelligence. Background Art

[0002] Tumor radiotherapy (referred to as radiotherapy) is a local radiotherapy method for treating malignant tumors by using radiation such as α, β, γ rays generated by radioactive isotopes and X-rays, electron beams, proton beams and other particle beams generated by various X-ray radiotherapy machines or accelerators. During radiotherapy, a radiotherapy plan needs to be designed based on the dose distribution map so as to perform radiotherapy based on this radiotherapy plan. In the design of radiotherapy plans, the dose distribution map is usually manually operated by physicians and physicists. Due to the high similarity between the target organ and the surrounding tissues, the diversity of prescription dose levels, and the presence of many sensitive key structures near the target organ, it often takes a long time to determine. The quality of the radiotherapy plan depends on the clinical experience and medical knowledge of physicians and physicists. The production process has a large workload, low efficiency, strong subjectivity, and large differences.

[0003] With the development of computer technology, some scholars have proposed to use a model trained by deep learning to automatically generate a dose distribution map applicable to patients for radiotherapy. However, the generated dose distribution map often ignores the spatial position relationship between organs and has low accuracy, resulting in low reference value in clinical applications. Summary of the Invention

[0004] Embodiments of the present invention provide an image search method, device, equipment and medium based on artificial intelligence to solve the problem that the current efficiency of obtaining a radiotherapy dose distribution map is low or the accuracy is not high.

[0005] An image search method based on artificial intelligence includes:

[0006] Obtain an image search request, where the image search request includes a target user identifier, an original CT image corresponding to the target user identifier, and an original drawn image of the organ at risk;

[0007] Input the original CT image and the original drawn image of the organ at risk into a dose analysis model to generate an original dose distribution map corresponding to the target user identifier;

[0008] Perform registration processing on the original dose distribution map to obtain a standard dose distribution map;

[0009] Input the standard dose distribution map into an image search model to obtain a target feature vector corresponding to the target user identifier;

[0010] Query a radiotherapy plan database based on the target feature vector to obtain a target dose distribution map that matches the target feature vector.

[0011] An image search device based on artificial intelligence, comprising:

[0012] An image search request acquisition module, configured to acquire an image search request, where the image search request includes a target user identifier, an original CT image corresponding to the target user identifier, and an original delineation drawing of critical organs;

[0013] An original dose distribution map acquisition module, configured to input the original CT image and the original delineation drawing of critical organs into a dose analysis model to generate an original dose distribution map corresponding to the target user identifier;

[0014] A standard dose distribution map acquisition module, configured to perform registration processing on the original dose distribution map to acquire a standard dose distribution map;

[0015] A target feature vector acquisition module, configured to input the standard dose distribution map into an image search model to acquire a target feature vector corresponding to the target user identifier;

[0016] A target dose distribution map acquisition module, configured to query a radiotherapy plan database based on the target feature vector to acquire a target dose distribution map that matches the target feature vector.

[0017] A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, where when the processor executes the computer program, the steps of the above-mentioned artificial intelligence-based image search method are implemented.

[0018] A computer-readable storage medium storing a computer program, where when the computer program is executed by a processor, the steps of the above-mentioned artificial intelligence-based image search method are implemented.

[0019] The above artificial intelligence-based image search method, apparatus, device and medium obtain the original CT image and the original delineation drawing of the critical organ in the image search request, input the original CT image and the original delineation drawing of the critical organ into a dose analysis model to generate an original dose distribution map, providing technical support for image search. Perform registration processing on the original dose distribution map to obtain a standard dose distribution map to eliminate differences between individuals, thereby achieving the purpose of information fusion. Input the standard dose distribution map into an image search model to obtain a target feature vector corresponding to the target user identifier, and query a radiotherapy plan database based on the target feature vector to obtain a target dose distribution map that matches the target feature vector, so that the server can search for the associated stored historical radiotherapy plan in the target radiotherapy plan database and send it to the client, so that the clinician can formulate a target radiotherapy plan based on the historical radiotherapy plan, improving the formulation efficiency and accuracy of the target radiotherapy plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments of the present invention. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 is a schematic diagram of an application environment of an artificial intelligence-based image search method in an embodiment of the present invention;

[0022] Figure 2 is a flowchart of an artificial intelligence-based image search method in an embodiment of the present invention;

[0023] Figure 3 is another flowchart of an artificial intelligence-based image search method in an embodiment of the present invention;

[0024] Figure 4 is another flowchart of an artificial intelligence-based image search method in an embodiment of the present invention;

[0025] Figure 5 is another flowchart of an artificial intelligence-based image search method in an embodiment of the present invention;

[0026] Figure 6 is another flowchart of an artificial intelligence-based image search method in an embodiment of the present invention;

[0027] Figure 7 is another flowchart of an artificial intelligence-based image search method in an embodiment of the present invention;

[0028] Figure 8 It is another flowchart of the artificial intelligence-based image search method in an embodiment of the present invention;

[0029] Figure 9 It is a schematic diagram of an artificial intelligence-based image search device in an embodiment of the present invention;

[0030] Figure 10 It is a schematic diagram of a computer device in an embodiment of the present invention. Specific Embodiments

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0032] The artificial intelligence-based image search method provided by the embodiments of the present invention can be applied in the application environment as shown in Figure 1 Specifically, the artificial intelligence-based image search method is applied in an image search system, and the image search system includes a client and a server as shown in Figure 1 The client and the server communicate through a network, and are used to process the generation of the original CT image and the original delineation drawing of the critical organ corresponding to the target user identifier, generate a standard dose distribution map, and quickly search for a historical registration distribution map similar to the standard dose distribution map as the target dose distribution map through an image search model, so as to improve the acquisition efficiency and accuracy of the target dose distribution map, thereby providing a reference for clinicians to formulate a target radiotherapy plan. The target radiotherapy plan is a radiotherapy plan formulated for the target user. Among them, the client, also known as the user end, refers to a program that provides local services corresponding to the server. The client can be installed on but not limited to various personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0033] In one embodiment, as shown in Figure 2 A kind of artificial intelligence-based image search method is provided, and taking the method applied in the Figure 1 server as an example for illustration, it includes the following steps:

[0034] S201: Obtain an image search request, where the image search request includes a target user identifier, an original CT image corresponding to the target user identifier, and an original delineation drawing of a critical organ.

[0035] Among them, the target user refers to the user who undergoes detection to determine the tumor distribution, so that clinicians can formulate a target radiotherapy plan. The target radiotherapy plan is a radiotherapy plan formulated for the target user, and one user corresponds to one radiotherapy plan. The target user identifier is an identifier used to uniquely identify the target user. For example, the target user identifier can be the name of the target user and the ID number of the target user, etc.

[0036] The original CT image is an image obtained by the target user through CT scanning. CT (Computed Tomography, that is, computed tomography) is to scan the layer where the part of the human body to be examined is located with an X-ray beam. The detector receives the X-rays passing through this layer, converts them into visible light, then into electrical signals through photoelectric conversion, and then into digital signals through an analog / digital converter. After the computer processes these digital signals, the cross-sectional or three-dimensional image of the part of the human body to be examined obtained is the original CT image, so as to use this original CT image to detect small lesions in the part to be examined. It can be understood that the part to be examined includes the diseased part and the non-diseased part. For example, the part to be examined is the lung, the diseased part is the right lung, and the non-diseased parts are the left lung and the organs around the left lung, etc.

[0037] The organs at risk refer to the non-diseased important tissues or organs within the radiation range of the radiotherapy rays. The original organs-at-risk contour map refers to the map obtained by contouring the non-diseased important tissues or organs within the radiation range of the radiotherapy rays in the original CT image. The original organs-at-risk contour map corresponds to the target user.

[0038] Specifically, the target user goes to the hospital for CT scanning to obtain the original CT image of the target user, and outlines the organs at risk based on the original CT image to obtain the original organs-at-risk contour map. The image data such as the original CT image and the original organs-at-risk contour map of the target user are associated with the target user identifier and stored in the image database. It can be understood that the image data of each user corresponds one-to-one with the user identifier for management. Among them, the image data includes but is not limited to CT images and organs-at-risk contour maps. The image database is a database used to store the image data of all users.

[0039] As an example, the clinician generates an image search request with the target user identifier, the original CT image corresponding to the target user identifier, and the original organs-at-risk contour map by clicking the image search button on the client, and sends the image search request to the server so that the server can obtain the image search request.

[0040] S202: Input the original CT image and the original organs-at-risk contour map into the dose analysis model to generate the original dose distribution map corresponding to the target user identifier.

[0041] Among them, the dose analysis model is a model used to generate a predicted dose distribution map. The dose analysis model is a model generated based on the training of a deep neural network. The dose analysis model is a model formed by training a training sample with a deep neural network. The training sample includes a training CT image, a training contour drawing of an organ at risk, and a corresponding training dose distribution map corresponding to the same user identification.

[0042] The original dose distribution map is the radiation dose predicted by the dose analysis model, that is, when performing radiotherapy on the target user, it is the predicted radiation dose required for the diseased part of the target user. Since the original dose distribution map is generated by the dose analysis model, the spatial position between different organs may be ignored in the original dose distribution map, and the accuracy is not high, which cannot meet the clinical standards. Therefore, clinicians cannot directly generate a target radiotherapy plan from the original dose distribution map.

[0043] Specifically, the original CT image and the original contour drawing of the organ at risk are input into the dose analysis model, so as to quickly generate the original dose distribution map corresponding to the target user identification, providing technical support for image search.

[0044] S203: Perform registration processing on the original dose distribution map to obtain a standard dose distribution map.

[0045] Among them, registration processing is used to compare or fuse images obtained by different users under different conditions, so as to accurately search for images subsequently. It can be understood that due to differences in the body shapes of different users, the sizes or spatial positions of the organs of different users are different. By finding a spatial transformation to map the dose distribution maps or CT images of different users to another image, the same images of different users are made to correspond to the points at the same position in space one by one, so as to eliminate the differences between different individuals.

[0046] Specifically, an image registration algorithm is used to register the original CT image with the standard CT image to obtain standard registration parameters. Based on the standard registration parameters, the original dose distribution map is transformed to obtain a standard dose distribution map, ensuring that the organs of the original CT image corresponding to the target user identification are in corresponding positions with the organs of the standard CT image, which can exclude the influence of factors such as the size and spatial position of the organs of different individuals on image search, ensure that similar images can be searched subsequently, and improve the accuracy of image search. The standard CT image refers to a general CT image template.

[0047] S204: Input the standard dose distribution map into the image search model to obtain the target feature vector corresponding to the target user identification.

[0048] Among them, the image search model refers to a pre-trained model for identifying dose distribution maps to output feature vectors. Specifically, the image search model is a model generated by training a convolutional neural network based on a triplet loss function, which can ensure that the distances of feature vectors generated by the image search model for similar images are small. Specifically, it makes the distances of corresponding feature vectors generated by the image search model for similar registration distribution maps small, and the distances of feature vectors generated by the image search model for dissimilar images are large. Specifically, it makes the distances of corresponding feature vectors generated by the image search model for dissimilar registration distribution maps large, that is, the image search model ensures that the distance of the feature vector generated by the image search model for a similar image is less than the distance of the feature vector generated by the image search model for a dissimilar image, so as to accurately obtain a historical registration distribution map similar to the standard dose distribution map subsequently, and thus determine the target dose distribution map. Specifically, the standard dose distribution map is input into the image search model to obtain the target feature vector corresponding to the target user identifier, providing a basis for subsequent searching for a map similar to the standard dose distribution map.

[0049] S205: Query the radiotherapy plan database based on the target feature vector to obtain the target dose distribution map that matches the target feature vector.

[0050] Among them, the radiotherapy plan database refers to a database for storing user data corresponding to historical users after radiotherapy. The user data includes the associated historical user identifier, historical CT image, historical dose distribution map, historical radiotherapy plan, and the historical feature vector corresponding to the historical dose distribution map.

[0051] Among them, the historical user refers to a user who has undergone radiotherapy. The historical user identifier is an identifier used to uniquely identify the historical user. The historical CT image is an image obtained by the historical user through CT scanning. The historical dose distribution map is a dose distribution map formed by the historical user during radiotherapy. The historical radiotherapy plan is a radiotherapy plan collected by the historical user during radiotherapy. The historical feature vector is a feature vector obtained by inputting the historical dose distribution map into the image search model. The target dose distribution map is a historical registration distribution map similar to the standard dose distribution map.

[0052] The historical feature vector is obtained through the image search model, which can ensure that the distances of feature vectors of similar historical dose distribution maps are small, and the distances of feature vectors of dissimilar historical dose distribution maps are large, so as to accurately obtain a historical registration distribution map similar to the standard dose distribution map subsequently, and thus determine the target dose distribution map based on the similar historical dose distribution map.

[0053] Specifically, the similarity between the target feature vector and the historical feature vector corresponding to any historical user identifier in the radiotherapy plan database is calculated using a similarity calculation formula, and the top M (M is a positive integer) historical feature vectors with the largest similarity are obtained. The historical dose distribution maps corresponding to these historical feature vectors are used as the target dose distribution maps, so that the server can search for the associated historical radiotherapy plans stored in the target radiotherapy plan database based on the target dose distribution maps and send them to the client, so that clinicians can formulate the target radiotherapy plan according to the historical radiotherapy plans, improving the formulation efficiency and accuracy of the target radiotherapy plan. It can be understood that since the historical radiotherapy plan is the radiotherapy plan that the historical user has already undergone, the historical radiotherapy plan has strong reference value, which can shorten the time required for clinicians to formulate the target radiotherapy plan, shorten the radiotherapy cycle of the target user, and provide timely radiotherapy for the target user.

[0054] The artificial intelligence-based image search method provided in this embodiment obtains the original CT image and the original critical organ delineation map in the image search request, inputs the original CT image and the original critical organ delineation map into the dose analysis model to generate the original dose distribution map, providing technical support for image search. The original dose distribution map is subjected to registration processing to obtain the standard dose distribution map to eliminate the differences between different individuals, thereby achieving the purpose of information fusion. The standard dose distribution map is input into the image search model to obtain the target feature vector corresponding to the target user identifier. Based on the target feature vector, the radiotherapy plan database is queried to obtain the target dose distribution map that matches the target feature vector, so that the server can search for the associated historical radiotherapy plans stored in the target radiotherapy plan database based on the target dose distribution map and send them to the client, so that clinicians can formulate the target radiotherapy plan according to the historical radiotherapy plans, improving the formulation efficiency and accuracy of the target radiotherapy plan.

[0055] In one embodiment, as shown in FIG. 3, before step S204, that is, before inputting the standard dose distribution map into the image search model to obtain the target feature vector corresponding to the target user identifier, the artificial intelligence-based image search method further includes:

[0056] S301: Obtain the historical user image data of the first historical user identifier, where the historical user image data includes the historical CT image, the historical critical organ delineation map, and the historical dose distribution map.

[0057] Among them, the first historical user identifier is the identifier of a historical user in the image database. The historical user image data is the image data associated with the first historical user identifier and stored in the image database. The historical user image data includes, but is not limited to, the historical CT image, the historical critical organ delineation map, and the historical dose distribution map. Among them, the historical critical organ delineation map refers to the map obtained by delineating the non-diseased important tissues or organs within the radiation range of the radiotherapy rays in the historical CT image of the same historical user.

[0058] S302: Input the historical CT image and the historical delineation of critical organs corresponding to the first historical user identifier into the dose analysis model to obtain the analysis dose distribution map corresponding to the first historical user identifier.

[0059] Among them, the analysis dose distribution map is the predicted dose distribution map obtained by predicting the historical CT image and the historical delineation of critical organs corresponding to the first historical user identifier through the dose distribution model, and the analysis dose distribution map is used as the training data for training the image search model.

[0060] S303: Based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier, obtain the historical registration distribution map and the analysis registration distribution map.

[0061] Among them, the historical registration distribution map refers to the map obtained after performing image registration processing on the historical dose distribution map. The analysis registration distribution map refers to the image obtained after performing image registration processing on the analysis dose distribution map.

[0062] Specifically, an image registration algorithm can be used to register the historical CT image of the first historical user identifier with the standard CT image to obtain the registration parameters corresponding to the first historical user identifier, and perform image registration processing on the historical dose distribution map based on the registration parameters to obtain the historical registration distribution map, and perform image registration processing on the analysis dose distribution map based on the registration parameters to obtain the analysis registration distribution map. Obtaining the historical registration distribution map and the analysis registration distribution map through registration processing can ensure that the trained model is more accurate.

[0063] S304: Query the image database, and determine the comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers.

[0064] Specifically, obtain the historical dose distribution maps corresponding to other historical user identifiers except the first historical user identifier from the image database, and perform registration processing on the historical dose distribution maps corresponding to other historical user identifiers to generate a comparison dose distribution map to eliminate the differences between the images of different users and ensure the accuracy of training the image search model. It can be understood that in order to obtain more training samples, the historical dose distribution maps corresponding to multiple other historical user identifiers except the first historical user identifier can be registered to obtain a comparison dose distribution map to obtain a sufficient number of training samples.

[0065] S305: Use the historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier as training samples.

[0066] Specifically, the historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier are used as training samples. It can be understood that since the historical registration distribution map and the analysis registration distribution map are image data of the same historical user, the similarity is relatively high. The comparison dose distribution map is a dose distribution map obtained by registering the historical dose distribution maps other than the first historical user identifier, and it is a dose distribution map that is not similar to the historical registration distribution map and the analysis registration distribution map corresponding to the first historical user identifier, so as to ensure that the generated image search model can make the distance between the feature vectors corresponding to dissimilar images large, specifically, to make the distance between the feature vectors corresponding to dissimilar registration distribution maps large, thereby ensuring the accuracy of subsequent image searches.

[0067] S306: Input the training samples into a convolutional neural network based on a triplet loss function for model training to obtain an image search model.

[0068] Specifically, input the historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier into a weight-sharing convolutional neural network based on a triplet loss function for training. When the loss is less than the function convergence value, it means that the training of the image search model is completed. The function convergence value is a pre-set value used to evaluate whether the loss function reaches the convergence requirement and can be zero. Among them, the triplet loss function is M (M is a positive integer) represents the number of training samples, i (i is a positive integer, i ≤ M) represents the i-th group of training samples, x a represents the vector corresponding to the historical registration distribution map, x p represents the vector corresponding to the analysis registration distribution map, x n represents the vector corresponding to the comparison dose distribution map, represents the Euclidean distance metric between the historical registration distribution map and the analysis registration distribution map, represents the Euclidean distance metric between the historical registration distribution map and the comparison dose distribution map, and α refers to the minimum interval between the distance between x a and x n and the distance between x a and x p When the value in [] is greater than the function convergence value, take the value in [] as the loss; when the value in [] is less than the function convergence value, the training of the image search model is completed.

[0069] The image search method based on artificial intelligence provided in this embodiment inputs the historical CT image and the historical delineation map of the critical organ corresponding to the first historical user identifier into the dose analysis model to quickly obtain the analysis dose distribution map corresponding to the first historical user identifier. Based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier, a historical registration distribution map and an analysis registration distribution map are obtained to eliminate the influence of the differences between images of different users on model training. The image database is queried, and based on the historical dose distribution maps of other historical user identifiers, the comparison dose distribution map corresponding to the first historical user identifier is determined. The historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier are used as training samples, thereby ensuring the accuracy of subsequent image search. The training samples are input into a convolutional neural network based on a triplet loss function for model training to quickly obtain an image search model, so as to ensure that the generated image search model can make the distance between the feature vectors corresponding to dissimilar images large, specifically, to make the distance between the feature vectors corresponding to dissimilar registration distribution maps large.

[0070] In one embodiment, as Figure 4 shown, step S303, that is, based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier, obtaining a historical registration distribution map and an analysis registration distribution map, includes:

[0071] S401: Use an image registration algorithm to register the historical CT image corresponding to the first historical user identifier with the standard CT image to obtain historical registration parameters.

[0072] Specifically, preprocess the historical CT image to provide a basis for registration. This preprocessing process includes: performing noise elimination processing on the historical CT image to eliminate interference factors; when the pixel sizes of the historical CT image and the standard CT image are different, adjust the size of the historical CT image so that the pixel sizes of the historical CT image and the standard CT image match, so that the features of the historical CT image and the standard CT image correspond, thereby ensuring the accuracy of the obtained historical registration parameters.

[0073] Select a first feature from a specific position of the preprocessed historical CT image, and select a second feature at the corresponding position of the standard CT image and the historical CT image to obtain the first feature and the second feature corresponding to the same specific position. In order to better determine the corresponding relationship between the historical CT image and the standard CT image, features at multiple corresponding positions can be taken from the historical CT image and the standard CT image; establish a three-dimensional coordinate system, determine the three-dimensional coordinates of the first feature of the historical CT image and the three-dimensional coordinates of the second feature of the standard CT image, determine the registration function of the preprocessed historical CT image and the standard CT image based on the two three-dimensional coordinates, and obtain historical registration parameters based on this registration function. Specifically, the historical registration parameters are the parameters of the registration function, and the historical CT image is resampled to verify the accuracy of the historical registration parameters.

[0074] S402: Perform image registration on the historical dose distribution map and the analysis dose distribution map based on the historical registration parameters to obtain the historical registration distribution map and the analysis registration distribution map.

[0075] Specifically, perform registration on the historical dose distribution map corresponding to the first historical user identifier and the analysis dose distribution map according to the historical registration parameters, that is, perform spatial transformation on the historical dose distribution map corresponding to the first historical user identifier and the analysis dose distribution map according to the historical registration parameters. Among them, the spatial transformation can be conversions such as rotation, reduction, and magnification to obtain the historical registration distribution map and the analysis registration distribution map. Since the training samples are all subjected to registration processing, the differences between the images of different users can be eliminated, ensuring the accuracy of the generated image search model.

[0076] The image search method based on artificial intelligence provided in this embodiment uses an image registration algorithm to register the historical CT image corresponding to the first historical user identifier with the standard CT image to obtain the historical registration parameters. Perform image registration on the historical dose distribution map and the analysis dose distribution map based on the historical registration parameters to obtain the historical registration distribution map and the analysis registration distribution map, which can eliminate the differences between the images of different users and ensure the accuracy of the generated image search model.

[0077] In one embodiment, as Figure 5 shown, step S304, that is, query the image database and determine the comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers, includes:

[0078] S501: Determine the target area based on the historical registration distribution map corresponding to the first historical user identifier.

[0079] Among them, the target area refers to the tumor area. The historical registration distribution map includes the target area and the organs at risk. The target area is outlined in the historical registration distribution map corresponding to the first historical user identifier by computer or manually, so as to subsequently find the historical dose distribution maps corresponding to other historical user identifiers with the same target area. For example, if the target area is a lung tumor, then the historical dose distribution maps of lung tumors are screened from the image database, thereby reducing the number of image searches and improving the efficiency of obtaining the comparison dose distribution map subsequently.

[0080] S502: Query the image database based on the target area to obtain a comparison dose distribution map whose similarity to the historical registration distribution map is less than the first preset threshold.

[0081] Among them, the first preset threshold is a threshold for determining whether the dose distribution maps corresponding to different historical user identifiers reach the similarity standard.

[0082] Specifically, the similarity between the historical registration distribution map corresponding to the first historical user identifier and the historical registration distribution maps corresponding to other historical user identifiers is calculated through an image matching algorithm. The historical registration distribution maps corresponding to other historical user identifiers with a similarity less than the first preset threshold are determined as the comparison dose distribution maps for the first historical user identifier, so as to obtain samples for training the image search model. It can be understood that since the historical registration distribution maps corresponding to each user identifier may be different, therefore, the historical registration distribution maps corresponding to the smallest X (X is a positive integer) other historical user identifiers with the smallest similarity can be obtained as training samples to ensure that the number of samples for training the image search model is sufficient. In this embodiment, the image matching algorithm includes, but is not limited to, a gray-scale based matching algorithm and a feature-based matching algorithm.

[0083] The image search method based on artificial intelligence provided in this embodiment determines the target area based on the historical registration distribution map corresponding to the first historical user identifier, thereby reducing the number of image searches and improving the efficiency of obtaining the comparison dose distribution map. Query the image database based on the target area to obtain the comparison dose distribution map with a similarity less than the first preset threshold to the historical registration distribution map, so as to obtain samples for training the image search model and provide technology for training the image search model.

[0084] In one embodiment, as Figure 6 shown, step S304, that is, querying the image database and determining the historical dose distribution maps of other historical user identifiers as the comparison dose distribution maps corresponding to the first historical user identifier, includes:

[0085] S601: Obtain the historical registration distribution map corresponding to the second historical user identifier from the image database.

[0086] Herein, the second historical user identifier refers to the identifier of any other historical user except the first historical user identifier. Specifically, the historical registration distribution map corresponding to the second historical user identifier is obtained from the image database for subsequent obtaining of the comparison dose distribution map for training.

[0087] S602: Obtain the first DVH map corresponding to the first historical user identifier based on the historical registration distribution map and the historical critical organ delineation map of the first historical user identifier, and generate the corresponding second DVH map based on the historical critical organ delineation map of the first historical user identifier and the historical registration distribution map corresponding to the second historical user identifier.

[0088] Among them, DVH is the abbreviation of Dose-Volume Histogram, which refers to the dose-volume histogram. In the DVH diagram, the vertical axis represents the volume of the diseased part, and the horizontal axis represents the radiotherapy dose. The dose-volume histogram specifically includes two curves. One curve reflects the dose-volume relationship of the target area in the radiotherapy plan, and the other curve reflects the dose-volume relationship of the organs at risk in the radiotherapy plan.

[0089] Specifically, the target area and the organs at risk are outlined in advance on the historical CT image corresponding to the first historical user identifier, and the area where the target area is located is converted into a vector 1 for representation, and the area where the organs at risk are located is converted into a vector 0 for representation to generate the first vector matrix; similarly, the numerical value of the radiotherapy dose in the historical registration distribution map corresponding to the first historical user identifier is converted into a corresponding second vector matrix; then the first vector matrix is multiplied by the second vector matrix to obtain the dose-volume curve of the target area corresponding to the first historical user identifier. It can be understood that since the vector of the area where the target area is located is 1, the dose-volume relationship of the target area is retained. By multiplying the first vector matrix by the second vector matrix, the dose-volume curve of the target area corresponding to the first historical user identifier can be obtained. Relatively, the area where the target area is located is converted into a vector 0 for representation, and the area where the organs at risk are located is converted into a vector 1 for representation to generate the third vector matrix. The second vector matrix is multiplied by the third vector matrix to obtain the dose-volume relationship in the organs at risk, that is, the dose-volume curve of the organs at risk corresponding to the first historical user identifier, so as to generate the first DVH diagram.

[0090] Similarly, the numerical value of the radiotherapy dose in the historical registration distribution map corresponding to the second historical user identifier is converted into a fourth vector matrix, and the first vector matrix is multiplied by the fourth vector matrix to obtain the dose-volume relationship of the target area corresponding to the second historical user identifier; relatively, the third vector matrix is multiplied by the fourth vector matrix to obtain the dose-volume relationship of the organs at risk of the second historical user identifier to generate the second DVH diagram.

[0091] S603: Use a similarity algorithm to calculate the similarity between the first DVH diagram and the second DVH diagram to obtain the target similarity.

[0092] Among them, the target similarity is a value used to represent the degree of similarity between the first DVH diagram and the second DVH diagram.

[0093] Specifically, take N points at equal intervals on the dose-volume curve of the target area in the first DVH diagram, and take N points at equal intervals on the dose-volume curve of the target area in the second DVH diagram. Calculate the distance differences of the N points on the dose-volume curves of the target areas in the first and second DVH diagrams to form a first coordinate difference. Similarly, take N points at equal intervals on the dose-volume curve of the organ at risk in the first DVH diagram, and take N points at equal intervals on the dose-volume curve of the organ at risk in the second DVH diagram. Calculate the distance differences of the N points on the dose-volume curves of the organs at risk in the first and second DVH diagrams to form a second coordinate difference. Calculate the average value of the first coordinate difference and the second coordinate difference as the target similarity. Subsequently, based on the target similarity, the comparative dose distribution diagrams can be accurately determined. It can be understood that the first DVH diagram is obtained from the historical registration distribution diagram and the historical organ-at-risk contour diagram corresponding to the first historical user identifier, and the second DVH diagram is obtained from the historical organ-at-risk contour diagram corresponding to the first historical user identifier and the historical registration distribution diagram corresponding to the second historical user identifier. If the historical registration distribution diagram corresponding to the first historical user identifier is similar to the historical registration distribution diagram corresponding to the second historical user identifier, then the first DVH diagram and the second DVH diagram should also be similar. On the contrary, if the historical registration distribution diagram corresponding to the first historical user identifier is not similar to the historical registration distribution diagram corresponding to the second historical user identifier, then the first DVH diagram and the second DVH diagram should also be dissimilar.

[0094] S604: If the target similarity is less than the second preset threshold, then use the historical registration distribution diagram of the second historical user identifier as the comparative dose distribution diagram corresponding to the historical registration distribution diagram of the first historical user identifier.

[0095] Among them, the second preset threshold is a value used to determine whether the first DVH diagram and the second DVH diagram meet the similarity standard.

[0096] Specifically, when the target similarity is less than the preset threshold, it indicates that the first DVH diagram and the second DVH diagram are not similar. Then use the historical dose distribution diagram corresponding to the second historical user identifier as the comparative dose distribution diagram corresponding to the historical registration distribution diagram of the first historical user identifier, and use the dissimilar historical dose distribution diagram as the comparative dose distribution diagram to ensure that the generated image search model can accurately identify the distances between similar images and dissimilar images, improve the accuracy of the generated image search model, and ensure the feature vectors generated by inputting the dose distribution diagram into the image search model subsequently.

[0097] In the image search method based on artificial intelligence provided in this embodiment, the historical registration distribution map corresponding to the second historical user identifier is obtained from the image database, so as to subsequently obtain a comparative dose distribution map for training. Based on the historical registration distribution map and the historical contour drawing of the organs at risk of the first historical user identifier, the first DVH map corresponding to the first historical user identifier is obtained. Based on the historical contour drawing of the organs at risk of the first historical user identifier and the historical registration distribution map corresponding to any historical user identifier, the corresponding second DVH map is generated. The similarity algorithm is used to calculate the similarity between the first DVH map and the second DVH map. Subsequently, the comparative dose distribution map can be accurately determined according to the target similarity. If the target similarity is less than the preset threshold, the historical dose distribution map of other user identifiers is used as the comparative dose distribution map corresponding to the historical registration distribution map of the first historical user identifier.

[0098] In one embodiment, as Figure 7 shown, step S205, query the radiotherapy plan database based on the target feature vector, and obtain the target dose distribution map that matches the target feature vector, including:

[0099] S701: Query the radiotherapy plan database to obtain the historical feature vector corresponding to any historical user identifier.

[0100] Specifically, after the image search model is trained, the historical dose distribution maps corresponding to all historical user identifiers are registered, and the registered historical registration distribution maps are input into the image search model to generate the historical feature vector corresponding to each historical user identifier, and stored in the radiotherapy plan database. Therefore, when the server queries the radiotherapy plan database, it can quickly obtain the historical feature vectors corresponding to all historical user identifiers.

[0101] S702: Calculate the target similarity value between the target feature vector and the historical feature vector.

[0102] Among them, the target similarity value is a value representing the similarity degree between the target feature vector and the historical feature vector.

[0103] Specifically, the server can quickly calculate the target similarity value between the target feature vector and the historical feature vector through the similarity algorithm. In this embodiment, the similarity algorithm includes but is not limited to the cosine similarity algorithm, the Euclidean distance algorithm, and the Manhattan algorithm, etc.

[0104] S703: If the target similarity value is greater than the third preset threshold, the historical registration distribution map corresponding to the historical feature vector is determined as the target dose distribution map.

[0105] Among them, the third preset threshold is a value used to determine whether the historical feature vector and the target feature vector reach the similarity standard.

[0106] Specifically, calculate the target similarity value between the target feature vector and the historical feature vectors, sort the target similarity values from largest to smallest, and according to the sorting result, select the first M (M is a positive integer) historical registration distribution maps corresponding to the historical feature vectors whose target similarity values are greater than the third preset threshold. Use the selected historical dose distribution maps as the target dose distribution maps, so as to subsequently search for the associated stored historical radiotherapy plans based on the historical dose distribution maps, and provide a reference for clinicians to formulate the historical radiotherapy plan of the target user.

[0107] The image search model provided in this embodiment queries the radiotherapy plan database to obtain the historical feature vectors corresponding to any historical user identifier. Calculate the target similarity value between the target feature vector and the historical feature vectors. If the target similarity value is greater than the third preset threshold, determine the historical registration distribution map corresponding to the historical feature vector as the target dose distribution map, so as to subsequently search for the associated stored historical radiotherapy plans based on the historical dose distribution map, and provide a reference for clinicians to formulate the historical radiotherapy plan of the target user.

[0108] In one embodiment, as Figure 8 shown, before step S205, before querying the radiotherapy plan database based on the feature vector corresponding to the target user identifier, the artificial intelligence-based image search method further includes:

[0109] S801: Obtain N historical dose distribution maps from the image database.

[0110] Specifically, store the image data of all historical users who have undergone radiotherapy in the image database, and associate and store the image data of each historical user with the corresponding historical user identifier, and store it in the server. The server can quickly obtain the historical dose distribution maps of all historical users through query algorithms such as keyword matching. For example, the server can find all historical dose distribution maps through the keyword "dose distribution map".

[0111] S802: Perform registration processing on the historical dose distribution maps using an image registration algorithm to obtain historical registration distribution maps.

[0112] Specifically, the process of obtaining the historical registration distribution maps by performing registration processing on the historical dose distribution maps using an image registration algorithm is the same as that in step S401. To avoid repetition, it will not be elaborated here.

[0113] S803: Input the historical registration distribution maps into the image search model to generate corresponding historical feature vectors.

[0114] Specifically, the process of inputting the historical registration distribution maps into the image search model to generate corresponding historical feature vectors is the same as the process of generating the target feature vector in step S204. To avoid repetition, it will not be elaborated here.

[0115] S804: Associatively store the historical feature vector of each historical user identifier and the corresponding historical registration distribution map in the radiotherapy plan database.

[0116] Specifically, associatively store the historical feature vector, historical registration distribution map, historical dose distribution map, and historical radiotherapy plan of the same historical user identifier in the radiotherapy plan database, so as to subsequently find the historical feature vector similar to the target feature vector, thereby obtaining the historical radiotherapy plan of the historical user with a similar diseased part to the target user, and providing a reference for clinicians.

[0117] The image search model provided in this embodiment obtains N historical dose distribution maps from the image database, performs registration processing on the historical dose distribution maps using an image registration algorithm to obtain historical registration distribution maps, inputs the historical registration distribution maps into the image search model to generate corresponding historical feature vectors, so as to subsequently calculate the target similarity value between the historical feature vector and the target feature vector. Associatively store the historical feature vector of each historical user identifier and the corresponding historical registration distribution map in the radiotherapy plan database, so as to subsequently find the historical feature vector similar to the target feature vector, thereby obtaining the historical radiotherapy plan of the historical user with a similar diseased part to the target user, and providing a reference for clinicians.

[0118] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0119] In one embodiment, an artificial intelligence-based image search device is provided, and the artificial intelligence-based image search device corresponds one-to-one with the artificial intelligence-based image search method in the above embodiment. As Figure 9 shown, the artificial intelligence-based image search device includes an image search request acquisition module 901, an original dose distribution map acquisition module 902, a standard dose distribution map acquisition module 903, a target feature vector acquisition module 904, and a target dose distribution map acquisition module 905.

[0120] The detailed description of each functional module is as follows:

[0121] The image search request acquisition module 901 is used to acquire an image search request, and the image search request includes a target user identifier, an original CT image corresponding to the target user identifier, and an original critical organ contour drawing.

[0122] The original dose distribution map acquisition module 902 is used to input the original CT image and the original critical organ contour drawing into a dose analysis model to generate an original dose distribution map corresponding to the target user identifier.

[0123] The standard dose distribution map acquisition module 903 is used to perform registration processing on the original dose distribution map to obtain the standard dose distribution map.

[0124] The target feature vector acquisition module 904 is used to input the standard dose distribution map into the image search model to obtain the target feature vector corresponding to the target user identifier.

[0125] The target dose distribution map acquisition module 905 is used to query the radiotherapy plan database based on the target feature vector to obtain the target dose distribution map that matches the target feature vector.

[0126] Furthermore, before the target feature vector acquisition module 904, the artificial intelligence-based image search device further includes: a historical user image data acquisition module, an analysis dose distribution map acquisition module, an image registration processing module, a comparison dose distribution map determination module, a training sample determination module, and an image search model acquisition module.

[0127] The historical user image data acquisition module is used to acquire the historical user image data of the first historical user identifier, and the historical user image data includes historical CT images, historical delineation maps of organs at risk, and historical dose distribution maps.

[0128] The analysis dose distribution map acquisition module is used to input the historical CT image and the historical delineation map of organs at risk corresponding to the first historical user identifier into the dose analysis model to obtain the analysis dose distribution map corresponding to the first historical user identifier.

[0129] The image registration processing module is used to obtain the historical registration distribution map and the analysis registration distribution map based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier.

[0130] The comparison dose distribution map determination module is used to query the image database and determine the comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers.

[0131] The training sample determination module is used to use the historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier as training samples.

[0132] The image search model acquisition module is used to input the training samples into a convolutional neural network based on the triplet loss function for model training to obtain the image search model.

[0133] Furthermore, the image registration processing module includes:

[0134] The historical registration parameter acquisition unit is used to register the historical CT image corresponding to the first historical user identifier with the standard CT image using an image registration algorithm to obtain the historical registration parameters.

[0135] The registration distribution map acquisition unit is configured to perform image registration on the historical dose distribution map and the analysis dose distribution map based on the historical registration parameters, and acquire the historical registration distribution map and the analysis registration distribution map.

[0136] Further, the comparison dose distribution map determination module includes: a target area location determination unit and a first judgment unit.

[0137] The target area location determination unit is configured to determine the target area location based on the historical registration distribution map corresponding to the first historical user identifier.

[0138] The first judgment unit is configured to query the image database based on the target area location, and acquire a comparison dose distribution map whose similarity to the historical registration distribution map is less than a first preset threshold.

[0139] Further, the comparison dose distribution map determination module includes: a historical registration distribution map acquisition unit, a DVH map acquisition unit, a target similarity acquisition unit, and a second judgment unit.

[0140] The historical registration distribution map acquisition unit is configured to acquire the historical registration distribution map corresponding to the second historical user identifier from the image database.

[0141] The DVH map acquisition unit is configured to acquire the first DVH map corresponding to the first historical user identifier based on the historical registration distribution map and the historical critical organ contour map of the first historical user identifier, and generate a corresponding second DVH map based on the historical critical organ contour map of the first historical user identifier and the historical registration distribution map corresponding to the second historical user identifier.

[0142] The target similarity acquisition unit is configured to calculate the similarity between the first DVH map and the second DVH map using a similarity algorithm, and acquire the target similarity.

[0143] The second judgment unit is configured to, if the target similarity is less than a second preset threshold, use the historical registration distribution map of the second historical user identifier as the comparison dose distribution map corresponding to the historical registration distribution map of the first historical user identifier.

[0144] Further, the target dose distribution map acquisition module 905 includes: a radiotherapy plan database query unit, a feature vector calculation unit, and a third judgment unit.

[0145] The radiotherapy plan database query unit is configured to query the radiotherapy plan database and acquire the historical feature vector corresponding to any historical user identifier.

[0146] The feature vector calculation unit is configured to calculate the target similarity value between the target feature vector and the historical feature vector.

[0147] A third judgment unit, configured to determine the historical registration distribution map corresponding to the historical feature vector as the target dose distribution map if the target similarity value is greater than a third preset threshold.

[0148] Further, before the target dose distribution map acquisition module 905, the artificial intelligence-based image search device further includes: a historical dose distribution map acquisition unit, a registration processing unit, a historical feature vector generation unit, and a radiotherapy plan database generation unit.

[0149] The historical dose distribution map acquisition unit is configured to acquire N historical dose distribution maps from the image database.

[0150] The registration processing unit is configured to perform registration processing on the historical dose distribution maps by using an image registration algorithm to obtain historical registration distribution maps.

[0151] The historical feature vector generation unit is configured to input the historical registration distribution maps into an image search model to generate corresponding historical feature vectors.

[0152] The radiotherapy plan database generation unit is configured to associate and store the historical feature vectors of each historical user identifier and the corresponding historical registration distribution maps in the radiotherapy plan database.

[0153] For the specific limitations of the artificial intelligence-based image search device, reference may be made to the limitations of the artificial intelligence-based image search method in the foregoing text, which will not be elaborated herein. Each module in the foregoing artificial intelligence-based image search device can be implemented in whole or in part by software, hardware, and their combination. The foregoing modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to the foregoing modules.

[0154] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 10 shown. The computer device includes a processor, a memory, a network interface, and a database connected by 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, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store historical registration distribution maps. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements an artificial intelligence-based image search method.

[0155] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the image search method based on artificial intelligence in the above embodiment are implemented, for example Figure 2 the steps S201 - S205 shown, or Figures 3 to 8 the steps shown in, and to avoid repetition, they will not be elaborated here. Or, when the processor executes the computer program, the functions of each module / unit in the embodiment of the image search device based on artificial intelligence are implemented, for example Figure 9 the functions of the image search request acquisition module 901, the original dose distribution map acquisition module 902, the standard dose distribution map acquisition module 903, the target feature vector acquisition module 904, and the target dose distribution map acquisition module 905 shown, and to avoid repetition, they will not be elaborated here.

[0156] In one embodiment, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the image search method based on artificial intelligence in the above embodiment are implemented, for example Figure 2 the steps S201 - S205 shown, or Figures 3 to 8 the steps shown in, and to avoid repetition, they will not be elaborated here. Or, when the processor executes the computer program, the functions of each module / unit in the embodiment of the image search device based on artificial intelligence are implemented, for example Figure 9 the functions of the image search request acquisition module 901, the original dose distribution map acquisition module 902, the standard dose distribution map acquisition module 903, the target feature vector acquisition module 904, and the target dose distribution map acquisition module 905 shown, and to avoid repetition, they will not be elaborated here.

[0157] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. 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 methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0158] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0159] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An image search method based on artificial intelligence, characterized in that, Including: Obtain an image search request, where the image search request includes a target user identifier, an original CT image corresponding to the target user identifier, and an original delineation map of organs at risk; the original delineation map of organs at risk refers to a map obtained by delineating non-diseased important tissues or organs within the radiation range of radiotherapy rays in the original CT image; Input the original CT image and the original delineation map of organs at risk into a dose analysis model to generate an original dose distribution map corresponding to the target user identifier; Perform registration processing on the original dose distribution map to obtain a standard dose distribution map; Input the standard dose distribution map into an image search model to obtain a target feature vector corresponding to the target user identifier; Query a radiotherapy plan database based on the target feature vector to obtain a target dose distribution map that matches the target feature vector; Before inputting the standard dose distribution map into the image search model to obtain a target feature vector corresponding to the target user identifier, the image search method based on artificial intelligence further includes: Obtain historical user image data of a first historical user identifier, where the historical user image data includes a historical CT image, a historical delineation map of organs at risk, and a historical dose distribution map; Input the historical CT image and the historical delineation map of organs at risk corresponding to the first historical user identifier into the dose analysis model to obtain an analysis dose distribution map corresponding to the first historical user identifier; Based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier, obtain a historical registration distribution map and an analysis registration distribution map; Query an image database and determine a comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers; Use the historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier as training samples; Input the training samples into a convolutional neural network based on a triplet loss function for model training to obtain an image search model.

2. The image search method based on artificial intelligence according to claim 1, wherein, The obtaining of the historical registration distribution map and the analysis registration distribution map based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier includes: Use an image registration algorithm to register the historical CT image corresponding to the first historical user identifier with a standard CT image to obtain historical registration parameters; Based on the historical registration parameters, perform image registration on the historical dose distribution map and the analysis dose distribution map to obtain the historical registration distribution map and the analysis registration distribution map.

3. The image search method based on artificial intelligence according to claim 1, characterized in that, The querying of the image database and determining the comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers includes: Determine the target area based on the historical registration distribution map corresponding to the first historical user identifier; Query the image database based on the target area to obtain a comparison dose distribution map whose similarity to the historical registration distribution map is less than a first preset threshold.

4. The image search method based on artificial intelligence according to claim 1, wherein The querying of the image database and determining the comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers includes: Obtain the historical registration distribution map corresponding to a second historical user identifier from the image database; Based on the historical registration distribution map and the historical contoured map of critical organs corresponding to the first historical user identifier, obtain the first DVH map corresponding to the first historical user identifier. Based on the historical contoured map of critical organs of the first historical user identifier and the historical registration distribution map corresponding to the second historical user identifier, generate the second DVH map corresponding to the second historical user identifier; Use a similarity algorithm to calculate the similarity between the first DVH map and the second DVH map to obtain the target similarity; If the target similarity is less than the second preset threshold, then use the historical registration distribution map of the second historical user identifier as the comparative dose distribution map corresponding to the historical registration distribution map of the first historical user identifier.

5. The method for image search based on artificial intelligence according to claim 1, characterized in that, The querying the radiotherapy plan database based on the target feature vector to obtain a target dose distribution map matching the target feature vector includes: Query the radiotherapy plan database to obtain the historical feature vector corresponding to any historical user identifier; Calculate the target similarity value between the target feature vector and the historical feature vector; If the target similarity value is greater than the third preset threshold, then determine the historical registration distribution map corresponding to the historical feature vector as the target dose distribution map.

6. The method for image search based on artificial intelligence according to claim 1, wherein, Before the querying the radiotherapy plan database based on the target feature vector, the artificial intelligence-based image search method further includes: Obtain N historical dose distribution maps from the image database; Use an image registration algorithm to perform registration processing on the historical dose distribution maps to obtain historical registration distribution maps; Input the historical registration distribution maps into the image search model to generate corresponding historical feature vectors; Associate and store the historical feature vectors of each historical user identifier and the corresponding historical registration distribution maps in the radiotherapy plan database.

7. An image search device based on artificial intelligence, characterized in that, It includes: An image search request acquisition module, configured to acquire an image search request, where the image search request includes a target user identifier, an original CT image, and an original contoured map of critical organs corresponding to the target user identifier; the original contoured map of critical organs refers to a map obtained by contouring non-diseased important tissues or organs within the radiation range of the radiotherapy rays in the original CT image; An original dose distribution map acquisition module, configured to input the original CT image and the original contoured map of critical organs into a dose analysis model to generate an original dose distribution map corresponding to the target user identifier; A standard dose distribution map acquisition module, configured to perform registration processing on the original dose distribution map to obtain a standard dose distribution map; A target feature vector acquisition module, configured to input the standard dose distribution map into the image search model to obtain a target feature vector corresponding to the target user identifier; A target dose distribution map acquisition module, configured to query the radiotherapy plan database based on the target feature vector to obtain a target dose distribution map matching the target feature vector; The artificial intelligence-based image search device further includes: A historical user image data acquisition module, configured to acquire historical user image data of a first historical user identifier, where the historical user image data includes a historical CT image, a historical contoured map of critical organs, and a historical dose distribution map; An analysis dose distribution map acquisition module, configured to input a historical CT image corresponding to the first historical user identifier and a historical critical organ delineation map into a dose analysis model, and acquire an analysis dose distribution map corresponding to the first historical user identifier; An image registration processing module, configured to acquire a historical registration distribution map and an analysis registration distribution map based on the historical dose distribution map and the analysis dose distribution map corresponding to the first historical user identifier; A comparison dose distribution map determination module, configured to query an image database and determine a comparison dose distribution map corresponding to the first historical user identifier based on the historical dose distribution maps of other historical user identifiers; A training sample determination module, configured to use the historical registration distribution map, the analysis registration distribution map, and the comparison dose distribution map corresponding to the first historical user identifier as training samples; An image search model acquisition module, configured to input the training samples into a convolutional neural network based on a triplet loss function for model training to acquire an image search model.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the artificial intelligence-based image search method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the artificial intelligence-based image search method according to any one of claims 1 to 6 are implemented.

Images