Method for generating phase recognition model and image recognition method

Abstract

The present application discloses a method for generating a phase recognition model and an image recognition method, and relates to the field of artificial intelligence technology, wherein the generation method includes: obtaining multiple sample CT image groups; obtaining the first position information of the first target key point corresponding to the sample CT image groups in the multiple sample CT image groups and the first embedded feature information corresponding to the sample CT image groups in the multiple sample CT image groups; training an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model. The present application solves the technical problem in the related art that the phase recognition of CT is performed manually, resulting in a relatively low efficiency in the phase recognition of CT.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular, to a method for generating a phase recognition model and a method for recognizing an image. Background Art

[0002] Deep learning algorithms for medical image analysis have achieved remarkable success in various clinical tasks, such as screening and diagnosis. Large-scale imaging data are usually required to train such algorithms to achieve robust performance. However, imaging data from different sources are obtained from various scanners, and scanning protocols vary in different hospitals and examination sections. To structure these heterogeneous data, data management is a critical step, which is time-consuming and labor-intensive.

[0003] Multi-phase Contrast-enhanced Computed Tomography (CT) is a typical imaging method for clinical diagnosis. After intravenous injection, the contrast agent enhances specific organs, blood vessels or tissues through the circulatory system. Among them, the phase distinction is more important in CT data cleaning. In the existing technology, medical staff often identify and mark the CT phase when writing reports. The disadvantage of this method is that it cannot automatically distinguish phases in large quantities, and there may be human errors that lead to marking errors.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a method for generating a phase recognition model and a method for recognizing an image, so as to at least solve the technical problem in the related art that the phases of CT are recognized manually, resulting in relatively low efficiency in the phase recognition of CT.

[0006] According to one aspect of an embodiment of the present application, a method for generating a phase recognition model is provided, comprising: obtaining a plurality of sample CT image groups, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the plurality of sample CT image groups are different; obtaining first position information of a first target key point corresponding to a sample CT image group in the plurality of sample CT image groups and first embedded feature information corresponding to a sample CT image group in the plurality of sample CT image groups, wherein the first target key point is a key point related to the phase; training an initial phase recognition model based on the plurality of sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the plurality of sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image.

[0007] Furthermore, obtaining first position information of a first target key point corresponding to a sample CT image group in the multiple sample CT image groups and first embedded feature information corresponding to a sample CT image group in the multiple sample CT image groups includes: obtaining template embedded feature information of multiple key points; performing feature extraction on pixel points of the sample CT image group in the multiple sample CT image groups through a feature extraction model to obtain the first embedded feature information; and determining the first position information based on the template embedded feature information and the first embedded feature information.

[0008] Furthermore, obtaining the template embedding feature information of multiple key points includes: marking the key points in the first target CT image group to obtain the second target CT image group; performing global embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain global embedding features; performing local embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain local embedding features; performing semantic embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain semantic embedding features; and obtaining the template embedding feature information based on the global embedding features, the local embedding features and the semantic embedding features.

[0009] Furthermore, determining the first position information based on the template embedded feature information and the first embedded feature information includes: performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain target similarity; screening pixel points of the sample CT image groups in the multiple sample CT image groups based on the target similarity to obtain first pixel points; screening the first pixel points based on a preset similarity threshold to obtain second pixel points; and determining the first position information based on the second pixel points.

[0010] Furthermore, performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain a target similarity includes: performing similarity calculation on the global embedded features in the template embedded feature information and the global embedded features in the first embedded feature information to obtain a first similarity; performing similarity calculation on the local embedded features in the template embedded feature information and the local embedded features in the first embedded feature information to obtain a second similarity; performing similarity calculation on the semantic embedded features in the template embedded feature information and the semantic embedded features in the first embedded feature information to obtain a third similarity; and obtaining the target similarity based on the first similarity, the second similarity and the third similarity.

[0011] Furthermore, an initial phase recognition model is trained based on the real phases corresponding to the multiple sample CT image groups, the first position information, the first embedded feature information and the sample CT image groups in the multiple sample CT image groups to obtain a target phase recognition model, including: extracting features of the sample CT image groups in the multiple sample CT image groups, the first position information and the first embedded feature information through the initial phase recognition model to obtain a first target feature vector; performing classification prediction based on the first target feature vector through the initial phase recognition model to obtain a predicted phase; determining a target loss function based on the predicted phase, the real phase and the first target feature vector; and training the initial phase recognition model based on the target loss function to obtain the target phase recognition model.

[0012] Further, feature extraction is performed on the sample CT image group, the first position information and the first embedded feature information in the multiple sample CT image groups through the initial phase recognition model to obtain the first target feature vector, including: for a target sample CT image group in the multiple sample CT image groups, feature extraction is performed on the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model to obtain the first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any group of sample CT image groups in the multiple sample CT image groups; first feature information corresponding to the sample CT image group in the multiple sample CT image groups is obtained; feature extraction is performed on the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group in the multiple sample CT image groups through the cross-attention layer in the initial phase recognition model to obtain the second feature information corresponding to the target sample CT image group; the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group are processed through the pooling layer in the initial phase recognition model to obtain the first target feature vector.

[0013] Furthermore, feature extraction is performed on the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model to obtain the first feature information corresponding to the target sample CT image group, including: random sampling and occlusion processing are performed on the first position information to obtain the second position information; for the target position information in the second position information, an average HU value corresponding to the target position information is obtained from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; feature extraction is performed on the average HU value and the first embedded feature information to obtain the first feature information corresponding to the target sample CT image group.

[0014] Furthermore, determining a target loss function based on the predicted phase, the true phase and the first target feature vector includes: performing calculations based on the predicted phase and the true phase to obtain a classification loss function; for a target sample CT image group among the multiple sample CT image groups, obtaining order information of the target sample CT image group in the multiple sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups; performing calculations based on the order information and the first target feature vector to obtain a contrast loss function; and determining the target loss function based on the classification loss function and the contrast loss function.

[0015] Furthermore, calculating based on the order information and the first target feature vector to obtain the contrast loss function includes: obtaining an initial order vector of a sample CT image group among the multiple sample CT image groups; calculating based on the order information and the initial order vector to obtain a target order vector of the target sample CT image group; encoding the target order vector to obtain an encoded order vector; and calculating based on the encoded order vector and the first target feature vector to obtain the contrast loss function.

[0016] According to another aspect of an embodiment of the present application, a method for image recognition is also provided, including: obtaining a CT image group to be recognized; obtaining third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; and using any one of the target phase recognition models described above to identify the target phase corresponding to the CT image group based on the CT image group, the third position information and the second embedded feature information, to obtain the target phase corresponding to the CT image group.

[0017] Furthermore, the target phase recognition model is used to identify the CT image group, the third position information and the second embedded feature information to obtain the target phase corresponding to the CT image group, including: extracting features from the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain a third feature vector; processing the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; and identifying based on the second target feature vector through the output layer in the target phase recognition model to obtain the target phase.

[0018] According to another aspect of an embodiment of the present application, a method for recognizing an image is also provided, the method comprising: performing feature extraction on the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain a third feature vector; processing the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; performing recognition based on the second target feature vector through the output layer in the target phase recognition model to obtain the target phase; obtaining the third position information of the second target key point corresponding to the CT image group and the second embedded feature information corresponding to the CT image group in a cloud server; performing recognition based on the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain the target phase corresponding to the CT image group; and returning the target phase corresponding to the CT image group to the client.

[0019] According to another aspect of an embodiment of the present invention, a CT image recognition system is also provided, including an image acquisition module for acquiring a CT image group to be recognized; a feature extraction module for acquiring third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; a target phase recognition model for performing recognition based on the CT image group, the third position information and the second embedded feature information to obtain a target phase corresponding to the CT image group; and a result output module for displaying the target phase corresponding to the CT image group.

[0020] According to another aspect of an embodiment of the present application, a device for generating a phase recognition model is also provided, comprising: a first acquisition unit, used to acquire multiple sample CT image groups, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the multiple sample CT image groups are different; a second acquisition unit, used to acquire first position information of a first target key point corresponding to a sample CT image group in the multiple sample CT image groups and first embedded feature information corresponding to a sample CT image group in the multiple sample CT image groups, wherein the first target key point is a key point related to the phase; a generation unit, used to train an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups, to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image.

[0021] Furthermore, the second acquisition unit includes: an acquisition subunit, used to acquire template embedded feature information of multiple key points; a first extraction subunit, used to perform feature extraction on pixel points of the sample CT image groups in the multiple sample CT image groups through a feature extraction model to obtain the first embedded feature information; a first determination subunit, used to determine the first position information based on the template embedded feature information and the first embedded feature information.

[0022] Furthermore, the acquisition subunit includes: a labeling module, which is used to label the key points in the first target CT image group to obtain the second target CT image group; a first extraction module, which is used to perform global embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain global embedding features; a second extraction module, which is used to perform local embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain local embedding features; a third extraction module, which is used to perform semantic embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain semantic embedding features; a first determination module, which is used to obtain the template embedding feature information based on the global embedding features, the local embedding features and the semantic embedding features.

[0023] Furthermore, the determination subunit includes: a first calculation module, used to perform similarity calculation on the template embedded feature information and the first embedded feature information to obtain a target similarity; a first screening module, used to screen pixel points of the sample CT image group in the multiple sample CT image groups according to the target similarity to obtain a first pixel point; a second screening module, used to screen the first pixel point according to a preset similarity threshold to obtain a second pixel point; and a second determination module, used to determine the first position information based on the second pixel point.

[0024] Furthermore, the calculation module includes: a first calculation sub-module, used to calculate the similarity between the global embedded features in the template embedded feature information and the global embedded features in the first embedded feature information to obtain a first similarity; a second calculation sub-module, used to calculate the similarity between the local embedded features in the template embedded feature information and the local embedded features in the first embedded feature information to obtain a second similarity; a third calculation sub-module, used to calculate the similarity between the semantic embedded features in the template embedded feature information and the semantic embedded features in the first embedded feature information to obtain a third similarity; a first determination sub-module, used to obtain the target similarity based on the first similarity, the second similarity and the third similarity.

[0025] Furthermore, the generation unit includes: a second extraction subunit, used to perform feature extraction on the sample CT image group, the first position information and the first embedded feature information in the multiple sample CT image groups through the initial phase recognition model to obtain a first target feature vector; a prediction subunit, used to perform classification prediction based on the first target feature vector through the initial phase recognition model to obtain a predicted phase; a second determination subunit, used to determine a target loss function based on the predicted phase, the true phase and the first target feature vector; a training subunit, used to train the initial phase recognition model based on the target loss function to obtain the target phase recognition model.

[0026] Further, the second extraction subunit includes: a fourth extraction module, used for performing feature extraction on the target sample CT image group, the first position information and the first embedded feature information for a target sample CT image group among the multiple sample CT image groups through the self-attention mechanism in the initial phase recognition model, so as to obtain first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups; a first acquisition module, used for obtaining first feature information corresponding to a sample CT image group among the multiple sample CT image groups; a fifth extraction module, used for performing feature extraction on the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group among the multiple sample CT image groups through the cross-attention layer in the initial phase recognition model, so as to obtain second feature information corresponding to the target sample CT image group; a processing module, used for processing the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group through the pooling layer in the initial phase recognition model, so as to obtain the first target feature vector.

[0027] Furthermore, the fourth extraction module includes: a processing submodule, used to perform random sampling and occlusion processing on the first position information to obtain second position information; a second determination submodule, used to obtain, for the target position information in the second position information, an average HU value corresponding to the target position information from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; an extraction submodule, used to perform feature extraction on the average HU value and the first embedded feature information to obtain first feature information corresponding to the target sample CT image group.

[0028] Furthermore, the second determination subunit includes: a second calculation module, used to perform calculations based on the predicted phase and the true phase to obtain a classification loss function; a second acquisition module, used to obtain the order information of the target sample CT image group in the multiple sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups; a third calculation module, used to perform calculations based on the order information and the first target feature vector to obtain a contrast loss function; a third determination module, used to determine the target loss function based on the classification loss function and the contrast loss function.

[0029] Furthermore, the third calculation module includes: an acquisition submodule, used to obtain an initial order vector of a sample CT image group among the multiple sample CT image groups; a fourth calculation submodule, used to perform calculations based on the order information and the initial order vector to obtain a target order vector of the target sample CT image group; an encoding submodule, used to perform encoding processing on the target order vector to obtain an encoding order vector; and a fifth calculation submodule, used to perform calculations based on the encoding order vector and the first target feature vector to obtain the contrast loss function.

[0030] According to another aspect of an embodiment of the present application, an image recognition device is also provided, including: a third acquisition unit, used to acquire a CT image group to be recognized; a fourth acquisition unit, used to acquire third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; and an identification unit, used to identify the target phase corresponding to the CT image group based on the CT image group, the third position information and the second embedded feature information through a target phase recognition model described in any one of the above items.

[0031] Furthermore, the recognition unit includes: a third extraction subunit, used to perform feature extraction on the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain a third feature vector; a processing subunit, used to process the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; and an identification subunit, used to perform identification based on the second target feature vector through the output layer in the target phase recognition model to obtain the target phase.

[0032] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising a memory storing an executable program; and a processor for running the program, wherein when the program is run, the method for generating a phase recognition model or the method for recognizing an image described in any one of the above is executed.

[0033] According to another aspect of an embodiment of the present invention, a computer program product is also provided, which includes a stored computer program, and when the computer program is executed by a processor, it implements the method for generating a phase recognition model or the method for recognizing an image as described in any one of the above items.

[0034] In an embodiment of the present application, the following steps are adopted: obtaining multiple sample CT image groups, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the multiple sample CT image groups are different; obtaining the first position information of the first target key point corresponding to the sample CT image groups in the multiple sample CT image groups and the first embedded feature information corresponding to the sample CT image groups in the multiple sample CT image groups, wherein the first target key point is a key point related to the phase; training an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image, which solves the technical problem in the related art of manually identifying the phase of CT, resulting in relatively low efficiency in identifying the phase of CT.

[0035] In this scheme, a sample CT image group consisting of sample CT images of multiple phases is obtained, and the first position information and the first embedded feature information of the first target key point corresponding to the sample CT image are obtained. Then, the initial phase recognition model is trained with the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model. The position information of the first target key point enables the target phase recognition model to quickly pay attention to and learn information related to the phase, so that the target phase recognition model can accurately identify the phase of the CT image, avoid using artificial methods to achieve phase recognition, and thereby achieve the effect of improving the accuracy and efficiency of phase recognition of CT images. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0037] Figure 1 is a hardware structure block diagram of a computer terminal provided according to Embodiment 1 of the present application;

[0038] Figure 2 The process of the method for generating the phase recognition model according to the first embodiment of the present application is as follows Figure 1 ;

[0039] Figure 3 is a schematic diagram of a method for generating a phase recognition model according to the first embodiment of the present application;

[0040] Figure 4 is a schematic diagram of a phase recognition model provided according to the first embodiment of the present application;

[0041] Figure 5 The process of the method for generating the phase recognition model according to the first embodiment of the present application is as follows Figure 2 ;

[0042] Figure 6 is a flow chart of an image recognition method provided according to Embodiment 2 of the present application;

[0043] Figure 7 is a flow chart of an image recognition method provided according to Embodiment 3 of the present application;

[0044] Figure 8 is a schematic diagram of a CT image recognition system provided according to Embodiment 4 of the present application;

[0045] Fig. 9 is a schematic diagram of a generation device of a phase recognition model provided in Embodiment 5 of the present application;

[0046] Fig.10 is a schematic diagram of an image recognition device provided according to Embodiment 6 of the present application;

[0047] Fig.11 This is a structural block diagram of an electronic device provided according to Embodiment 7 of the present application. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0049] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0050] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0051] multi-phase contrast-enhanced computed tomography (CT): multi-phase contrast-enhanced computed tomography (CT): multi-phase contrast-enhanced computed tomography (CT).

[0052] Phases of CT images: Generally, there are four phases: plain scan phase, arterial phase, venous phase, and delayed phase. Refers to the different scanning phases in CT scanning. Plain scan is a scan without the use of any contrast agent. Arterial, venous, and delayed phases are scans performed approximately 20 to 35 seconds, 60 to 70 seconds, and 2 to 5 minutes after contrast agent injection, respectively. CT images of different phases help doctors better diagnose and evaluate the nature of the disease by comparing changes in organs and tissues at different time points.

[0053] Hounsfield Unit (HU): Hounsfield unit is a unit of measurement used by CT to measure the density of a certain local tissue or organ in the human body.

[0054] Self-supervised Anatomical eMbedding (SAM): A self-supervised anatomical location feature extraction method.

[0055] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards in the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0056] Example 1

[0057] According to an embodiment of the present application, a method for generating a phase recognition model is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0058] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for generating a phase recognition model is shown. Figure 1As shown, the computer terminal (or mobile device) 10 may include a processor set 102 (the processor set 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA, and the processor set 102 may include a processor set, Figure 1 102a, 102b, ..., 102n are used to illustrate), a memory 104 for storing data, and a transmission module 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0059] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0060] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for generating the phase recognition model in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned method for generating the phase recognition model. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0061] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0062] The display may be a touch screen type liquid crystal display, which may enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0063] Under the above operating environment, this application provides Figure 2 The method for generating the phase recognition model shown. Figure 2 The process of the method for generating the phase recognition model according to Example 1 of the present application is as follows Figure 1 .like Figure 2 As shown, the method also includes:

[0064] Step S201 : acquiring a plurality of sample CT image groups, wherein the sample CT images in the sample CT image groups correspond to the same phase, and the plurality of sample CT image groups correspond to different phases.

[0065] Optionally, dynamic enhanced CT scan images, i.e., the sample CT images mentioned above, are collected, including different inspection parts from the chest to the abdomen. It should be noted that multi-layer CT images will be generated at the same phase when using dynamic enhanced CT scanning. The sample CT image group mentioned above includes CT images of four common phases, i.e., CT images of the plain phase, arterial phase, venous phase, and delayed phase. Among them, the sample CT images in the sample CT image group correspond to the same phase, and the corresponding phases between multiple sample CT image groups are different.

[0066] Step S202, obtaining first position information of a first target key point corresponding to a sample CT image group among multiple sample CT image groups and first embedded feature information corresponding to a sample CT image group among multiple sample CT image groups, wherein the first target key point is a key point related to the period.

[0067] Optionally, when training the initial phase recognition model, in order to allow the initial phase recognition model to learn the phase information more accurately, the anatomical interest points (i.e., the first target key points mentioned above) of the sample CT image groups among the multiple sample CT image groups are determined and the first embedded feature information corresponding to the sample CT image groups is obtained.

[0068] For example, according to the propagation and distribution of contrast agents in the human body, 16 human points of interest (POIs) were manually selected to cover the scanning area from the chest to the lower abdomen, including 1 heart, 2 portal veins, 3 aortas, 6 kidneys, 2 ureters, and 2 iliac arteries. The first position information of the first target key point corresponding to the sample CT image was determined based on the above-set 16 human points of interest.

[0069] It should be noted that the first embedded feature information corresponding to the sample CT image group can be obtained through the feature extraction model.

[0070] Step S203, training an initial phase recognition model based on multiple sample CT image groups, the first position information, the first embedded feature information, and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image.

[0071] Optionally, a training set is constructed using multiple sample CT image groups, first position information, first embedded feature information, and real phases corresponding to sample CT image groups in the multiple sample CT image groups, and then the initial phase recognition model is trained using the training set to generate a target phase recognition model.

[0072] The target phase recognition model can be used for large-scale CT data cleaning, and can also be used as an automated tool for selecting required phases when deploying downstream clinical AI models.

[0073] In summary, a sample CT image group consisting of sample CT images of multiple phases is obtained, and the first position information and the first embedded feature information of the first target key point corresponding to the sample CT image are obtained. Then, the initial phase recognition model is trained through the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model. The position information of the first target key point can enable the target phase recognition model to quickly pay attention to and learn information related to the phase, so that the target phase recognition model can accurately identify the phase of the CT image, avoid using artificial methods to achieve phase recognition, and thereby achieve the effect of improving the accuracy and efficiency of phase recognition of CT images.

[0074] In order to improve the accuracy of determining the first position information, in the method for generating a phase recognition model provided in the first embodiment of the present application, obtaining the first position information of the first target key point corresponding to the sample CT image group in multiple sample CT image groups and the first embedded feature information corresponding to the sample CT image group in multiple sample CT image groups includes: obtaining the template embedded feature information of multiple key points; performing feature extraction on the pixel points of the sample CT image group in the multiple sample CT image groups through a feature extraction model to obtain the first embedded feature information; and determining the first position information based on the template embedded feature information and the first embedded feature information.

[0075] Optionally, 16 human points of interest (POIs) are manually selected to cover the scanning area from the chest to the lower abdomen, including 1 heart, 2 portal veins, 3 aortas, 6 kidneys, 2 ureters, and 2 iliac arteries. These points are automatically located in a one-time manner by an anatomical embedding extractor (i.e., feature extraction model), and the feature extraction model learns unique global and local embeddings for each anatomical point through self-supervised coarse-to-fine contrast learning, thereby obtaining the template embedding feature information of the above-mentioned multiple key points.

[0076] For example, four phase CT scan samples of a subject (e.g., plain scan, arterial, venous, and delayed phases) are randomly selected as a template image group and they are manually annotated with POI coordinates. By extracting the anatomical embeddings of these points, the above-mentioned template embedding feature information is obtained).

[0077] Then, feature extraction is performed on the pixel points of the sample CT image group through a feature extraction model (e.g., SAM) to obtain first embedded feature information. It should be noted that for POI positioning, the images in the sample CT image group are resampled to 2mm×2mm×2mm, and the HU is normalized to the range of 0 to 255 before POI positioning is performed.

[0078] Finally, the first position information is determined through the template embedding feature information and the first embedding feature information.

[0079] By comparing and analyzing the template embedded feature information of the key point and the first embedded feature information, the target key point in the sample CT image can be accurately identified, thereby improving the accuracy of determining the first position information.

[0080] In order to improve the comprehensiveness and accuracy of template embedded feature information, in the generation method of the phase recognition model provided in Example 1 of the present application, obtaining the template embedded feature information of multiple key points includes: marking the key points in the first target CT image group to obtain the second target CT image group; performing global embedded feature extraction on the key points in the second target CT image group through the feature extraction model to obtain global embedded features; performing local embedded feature extraction on the key points in the second target CT image group through the feature extraction model to obtain local embedded features; extracting semantic embedded features from the key points in the second target CT image group to obtain semantic embedded features; and obtaining template embedded feature information based on the global embedded features, local embedded features and semantic embedded features.

[0081] Optionally, the first target CT image group is obtained as the template image group. It should be noted that the template image group can be a four-phase CT scan sample image of the same object (for example, plain scan, arterial, venous and delayed phases). 16 human points of interest (POIs) were manually selected to cover the scanning area from the chest to the lower abdomen, including 1 heart, 2 portal veins, 3 aortas, 6 kidneys, 2 ureters and 2 iliac arteries. Then, these human points of interest were manually annotated with POI coordinates to obtain the above-mentioned second target CT image group.

[0082] The anatomical embedding features of these points are extracted through the feature extraction model, including local embedding features, semantic embedding features, and global embedding features. Global Embeddings: These embedding vectors capture the overall contextual information of the image. They may include the general layout, color distribution, texture, etc. of the image, but do not focus on specific areas or objects in the image. Local Embeddings: In contrast to global embeddings, local embeddings focus on specific areas or key points in the image. These embedding vectors can capture local features such as edges, corners, or parts of specific objects. Semantic Embeddings: These embedding vectors contain semantic information about specific objects or scenes in the image. They are usually related to the object category, attributes, or behaviors in the image, and can provide a deeper understanding of the image content. It should be noted that the feature extraction model can be a SAM (Self-supervised Anatomical eMbedding) model, a self-supervised anatomical location feature extraction model, or other feature extraction models.

[0083] For example, 64 templates were globally, locally, and semantically embedded using four CT scan samples. (i.e. the above template embedding feature information). is the global embedding feature, is the local embedding feature, and is the semantic embedding feature, and i represents the i-th key point.

[0084] Learning unique template embedding feature information for each anatomical point helps to improve the accuracy of subsequent key point recognition on sample CT images.

[0085] In order to further improve the accuracy of determining the first position information, in the generation method of the phase recognition model provided in the first embodiment of the present application, determining the first position information based on the template embedded feature information and the first embedded feature information includes: performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain the target similarity; based on the target similarity, screening the pixel points of the sample CT image groups in the multiple sample CT image groups to obtain the first pixel point; screening the first pixel point based on the preset similarity threshold to obtain the second pixel point; and determining the first position information based on the second pixel point.

[0086] Optionally, the first embedded feature information and the template embedded feature information are obtained using the same feature extraction model. Therefore, the similarity between the template embedded feature information and the first embedded feature information is calculated to obtain the target similarity. The first embedded feature information is the feature information corresponding to the pixel points of the sample CT image. Therefore, the similarity between the template embedded feature information and the first embedded feature information is calculated.

[0087] Then, the pixel points of the sample CT image groups in the plurality of sample CT image groups are screened by similarity, for example, the pixel points with high similarity in the sample CT images are screened out to obtain the above-mentioned first pixel point.

[0088] It should be noted that some target POIs may not appear in the query image due to different fields of view of the entire CT examination, which may result in poor matching quality. Therefore, it is necessary to filter out incorrect matches with similarity scores below the threshold, that is, the first pixel point is screened based on the preset similarity threshold to obtain the second pixel point.

[0089] Finally, the first position information of the first target key point is determined according to the position information of the second pixel point.

[0090] By screening the pixel points of the sample CT image through the similarity between the first embedded feature information and the template embedded feature information and the similarity threshold, the accuracy of the first position information of the first target key point can be effectively improved.

[0091] In order to improve the accuracy of target similarity, in the generation method of the phase recognition model provided in the first embodiment of the present application, the template embedded feature information and the first embedded feature information are similarly calculated to obtain the target similarity, including: calculating the similarity between the global embedded features in the template embedded feature information and the global embedded features in the first embedded feature information to obtain the first similarity; calculating the similarity between the local embedded features in the template embedded feature information and the local embedded features in the first embedded feature information to obtain the second similarity; calculating the similarity between the semantic embedded features in the template embedded feature information and the semantic embedded features in the first embedded feature information to obtain the third similarity; and obtaining the target similarity based on the first similarity, the second similarity and the third similarity.

[0092] Optionally, the first embedding feature information and the template embedding feature information are obtained using the same feature extraction model. Therefore, the first embedding feature information also includes global embedding features, local embedding features and semantic embedding features. For example, the first embedding feature information is ,in, is the global embedding feature in the first embedding feature information, is the local embedding feature in the first embedding feature information, is the semantic embedding feature in the first embedding feature information.

[0093] Calculate the similarity between the global embedding feature in the template embedding feature information and the global embedding feature in the first embedding feature information, calculate the similarity between the local embedding feature in the template embedding feature information and the local embedding feature in the first embedding feature information, and calculate the similarity between the semantic embedding feature in the template embedding feature information and the semantic embedding feature in the first embedding feature information to obtain a first similarity, a second similarity and a third similarity. For example, the first similarity, the second similarity and the third similarity are respectively .

[0094] Finally, the target similarity is calculated by the first similarity, the second similarity and the third similarity. For example, the formula The above target similarity is calculated.

[0095] By calculating the similarities between global embedding, local embedding and semantic embedding respectively, the accuracy of calculating target similarity can be effectively improved.

[0096] How to train the initial phase recognition model is crucial. Therefore, in the phase recognition model generation method provided in Example 1 of the present application, the initial phase recognition model is trained based on multiple sample CT image groups, first position information, first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to obtain the target phase recognition model, including: extracting features of the sample CT image groups, the first position information and the first embedded feature information in the multiple sample CT image groups through the initial phase recognition model to obtain the first target feature vector; performing classification prediction based on the first target feature vector through the initial phase recognition model to obtain the predicted phase; determining the target loss function based on the predicted phase, the real phase and the first target feature vector; training the initial phase recognition model based on the target loss function to obtain the target phase recognition model.

[0097] Optionally, multiple sample CT image groups, first position information, first embedded feature information, and real phases corresponding to sample CT image groups in the multiple sample CT image groups are input into an initial phase recognition model, and the initial phase recognition model performs feature extraction on the sample CT image groups in the multiple sample CT image groups, the first position information, and the first embedded feature information to obtain a first target feature vector, and the initial phase recognition model then performs classification prediction based on the first target feature vector to obtain a predicted phase.

[0098] Then, the above-mentioned target loss function is calculated based on the predicted phase, the true phase and the first target feature vector. Finally, the initial recognition model is iterated through the target loss function to obtain the target initial recognition model.

[0099] By training the initial recognition model through the sample CT image group, the first position information and the first embedded feature information, the initial recognition model can learn the feature information related to the phase more accurately, thereby improving the recognition performance of the target initial recognition model.

[0100] In order to more comprehensively extract the feature information of the sample CT image, in the method for generating the phase recognition model provided in the first embodiment of the present application, the sample CT image group, the first position information and the first embedded feature information in the multiple sample CT image groups are feature extracted by the initial phase recognition model to obtain the first target feature vector, including: for the target sample CT image group in the multiple sample CT image groups, the target sample CT image group, the first position information and the first embedded feature information are feature extracted by the self-attention mechanism in the initial phase recognition model to obtain the first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any group of sample CT image groups in the multiple sample CT image groups; the first feature information corresponding to the sample CT image group in the multiple sample CT image groups is obtained; the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group in the multiple sample CT image groups are feature extracted by the cross-attention layer in the initial phase recognition model to obtain the second feature information corresponding to the target sample CT image group; the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group are processed by the pooling layer in the initial phase recognition model to obtain the first target feature vector.

[0101] Optionally, after multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups are input into the initial phase recognition model, for any group of sample CT image groups in the multiple sample CT image groups (i.e., the above-mentioned target sample CT image group), the self-attention mechanism in the initial phase recognition model performs feature extraction on the target sample CT image group, the first position information and the first embedded feature information, i.e., the feature information of the target sample CT image group itself is extracted through the self-attention mechanism to obtain the above-mentioned first feature information.

[0102] Then, for any one of the plurality of sample CT image groups, first feature information corresponding to the sample CT image group in the plurality of sample CT image groups is obtained. For example, the plurality of sample CT image groups are , where I j represents the jth sample CT image group, and N represents the total number of multiple sample CT image groups. The first feature information corresponding to the sample CT image group in the multiple sample CT image groups is .

[0103] After obtaining the first feature information corresponding to the sample CT image group in the multiple sample CT image groups, the cross-attention layer in the initial phase recognition model performs feature extraction on the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group in the multiple sample CT image groups, that is, the cross-attention layer extracts the feature information between the multiple sample CT image groups to obtain the above-mentioned second feature information.

[0104] For example, the second feature information is ,in, is the first feature information corresponding to the target sample CT image group, This is the second characteristic information mentioned above.

[0105] Finally, the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group are processed by the pooling layer to obtain a first target feature vector. For example, the POI features are aggregated using global average pooling (GAP) to form the final first target feature vector.

[0106] The characteristic information of the sample CT images can be extracted more comprehensively through the characteristic information of the target sample CT image group itself and the characteristic information between multiple sample CT image groups.

[0107] In order to improve the accuracy of the first feature information, in the generation method of the phase recognition model provided in Example 1 of the present application, the self-attention mechanism in the initial phase recognition model is used to extract features of the target sample CT image group, the first position information and the first embedded feature information to obtain the first feature information corresponding to the target sample CT image group, including: randomly sampling and occlusion processing of the first position information to obtain the second position information; for the target position information in the second position information, the average HU value corresponding to the target position information is obtained from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; feature extraction is performed on the average HU value and the first embedded feature information to obtain the first feature information corresponding to the target sample CT image group.

[0108] Optionally, changes in the CT scan area result in differences in the presence of POIs. For example, a lung CT may contain only the first four POIs (i.e., one in the heart, one in the thoracic aorta, and two in the portal vein), while the others are out of view. In order to train an initial phase recognition model that is robust to different valid POI combinations, a random POI masking strategy is introduced during the training phase, which randomly samples a subset of POIs and removes the remaining POIs. That is, the first position information is randomly sampled and occluded as described above to obtain the second position information.

[0109] Furthermore, as the contrast agent circulates in the human body after injection, the HU value in the organ or blood vessel of interest shows a certain enhancement pattern, thereby being able to distinguish different CT stages. Therefore, for the target position information in the second position information, the average HU value corresponding to the target position information is obtained from the target sample CT image group with the target position information as the center, for example, by using a 4mm HU value centered on the target position information. 2 The HU value in the image patch is used to obtain the average HU value corresponding to the target position information.

[0110] For example, the average HU value corresponding to the second position information in the target sample CT image group is , where K is the total number of second location information, is the i-th second position information. Then, the embedded feature information corresponding to the second position information is determined from the first embedded feature information. .

[0111] Finally, by performing feature extraction on the embedded feature information corresponding to the average HU value and the second position information, the first feature information corresponding to the target sample CT image group is obtained.

[0112] For example, first pass Position encoding is performed, where represents the connection operation, W P is a linear projection, and summing up all the second position information to get , and finally, through The first feature information corresponding to the above-mentioned target sample CT image group is obtained.

[0113] By randomly sampling and occlusion processing the position information, an initial phase recognition model that is robust to different valid POI combinations can be trained to improve the accuracy of subsequent phase recognition.

[0114] In order to improve the accuracy of the loss function, in the generation method of the phase recognition model provided in Example 1 of the present application, the target loss function is determined based on the predicted phase, the true phase and the first target feature vector, including: calculating based on the predicted phase and the true phase to obtain a classification loss function; for a target sample CT image group in multiple sample CT image groups, obtaining order information of the target sample CT image group in the multiple sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups of the multiple sample CT image groups; calculating based on the order information and the first target feature vector to obtain a contrast loss function; and determining the target loss function based on the classification loss function and the contrast loss function.

[0115] Optionally, the target loss function is composed of a classification loss function and a contrast loss function, and the classification loss function is the loss between the predicted phase and the true phase. It should be noted that the classification loss function can be an ordinal loss function. For example, the initial phase recognition model maps the first target feature vector to a scalar prediction score, that is:

[0116]

[0117] where ψ(•) and σ(•) represent the prediction head and the sigmoid function, and θ1, θ2, …, θC-1 are the ordered class thresholds that partition the prediction space R into C class intervals, z j is the first target feature vector mentioned above. For example, the classification loss function is , where y c , 1≤c≤C represents the label (immediate phase).

[0118] In order to further utilize the temporal order relationship between different sequences, contrast loss is introduced on the image order pair, that is, for the target sample CT image group in the multiple sample CT image groups, the order information of the target sample CT image group in the multiple sample CT image groups is obtained, and then the contrast loss function is obtained by calculation through the order information and the first target feature vector.

[0119] Through the classification loss function and the contrast loss function, the similarity between the positive sample image-sequence pairs can be made greater, while the similarity between the negative sample pairs can be made smaller, thereby improving the effectiveness of the phase recognition model training.

[0120] In an optional embodiment, the following steps are used to calculate and obtain the contrast loss function, including: obtaining an initial sequence vector of a sample CT image group among multiple sample CT image groups; calculating based on the order information and the initial sequence vector to obtain a target sequence vector of the target sample CT image group; encoding the target sequence vector to obtain a coded sequence vector; and calculating based on the coded sequence vector and the first target feature vector to obtain a contrast loss function.

[0121] Optionally, an initial sequence vector of a sample CT image group among the plurality of sample CT image groups is obtained, and then a calculation is performed based on the sequence information and the initial sequence vector to obtain a sequence vector of the target sample CT image group.

[0122] For example, multiple sample CT image groups are sorted by acquisition time, and the corresponding initial order vector is [1, 2, ..., N]. By subtracting [1, 2, ..., N] from the current order j, [j1, j2, ..., j n], which is the target order vector mentioned above. For example, if the initial order vector is [1, 2, ..., 5] and the current order is 2, then the corresponding target order vector is [-1, 0, ..., 3].

[0123] Then, the target sequence vector is encoded to obtain the encoded sequence vector, for example, ,in, That is, the encoding order vector corresponding to the target sample CT image group, For the encoder.

[0124] Finally, the contrast loss function is calculated by encoding the order vector and the first target feature vector, for example, , where sim(•) represents cosine similarity, τ is the temperature parameter, This is the contrast loss function mentioned above.

[0125] In an optional embodiment, the following may be used: Figure 3 The schematic diagram shown in the figure is used to train the target phase recognition model mentioned above, specifically including: using a manually annotated image set of a single patient as a template, automatically locating a set of pre-defined POIs related to the phase, and obtaining a sample CT series. The sample CT series is located by the feature extraction model to obtain POIs, and a subset of POIs is randomly sampled by random POI masking, and the remaining POIs are removed.

[0126] Then, a subset of POIs and HU values ​​are extracted through the self-attention layer, and feature extraction is performed again through the cross-attention layer. The extracted features are aggregated through GAP to obtain the final feature information Z1, Z2, Z N Based on Z1, Z2, Z N Output the classification result of the initial phase recognition model, iterate the initial phase recognition model based on the classification loss function (for example, sequence regression loss) through MLP (multi-layer perceptron), and then obtain the above-mentioned target phase recognition model. It should be noted that the loss function also includes the contrast loss function, such as Figure 3 As shown, the order of the sample CT series is obtained by subtracting [1, 2, ..., N] from the current order j to obtain the order vector [j1, j2, ..., j n ], the encoder encodes the sequential vector to obtain o1, o2, o N , and finally through o1, o2, o N and Z1, Z2, Z N Get the contrast loss function.

[0127] In an optional embodiment, the schematic diagram of the target phase recognition model is as follows: Figure 4 As shown, the target phase recognition model can be composed of a self-attention layer, a cross-attention layer, a GAP and an output layer.

[0128] It should be noted that if there is only one CT series of the target object to be identified in actual applications, the cross-attention layer in the target phase recognition model can be removed.

[0129] In an optional embodiment, the following may be used: Figure 5 The flowchart shown in the figure obtains the target phase recognition model mentioned above, which specifically includes: POI localization based on feature extraction model: using manually annotated images of a single patient as a template, automatically locate a set of pre-defined POIs related to the phase. Multi-series POI encoder (i.e., the initial phase recognition model mentioned above): capturing the CT intensity relationship between POIs in the attention-based encoder, and introducing a random POI masking strategy as an enhancement for scan range changes. Series order perception target: using ordinal regression loss and contrast loss to train the model.

[0130] In the method for generating a phase recognition model provided in Example 1 of the present application, a plurality of sample CT image groups are obtained, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the plurality of sample CT image groups are different; first position information of a first target key point corresponding to the sample CT image groups in the plurality of sample CT image groups and first embedded feature information corresponding to the sample CT image groups in the plurality of sample CT image groups are obtained, wherein the first target key point is a key point related to the phase; an initial phase recognition model is trained based on the plurality of sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the plurality of sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image, thereby solving the technical problem in the related art of manually identifying the phase of CT, resulting in relatively low efficiency in identifying the phase of CT.

[0131] In this scheme, a sample CT image group consisting of sample CT images of multiple phases is obtained, and the first position information and the first embedded feature information of the first target key point corresponding to the sample CT image are obtained. Then, the initial phase recognition model is trained with the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model. The position information of the first target key point can enable the target phase recognition model to quickly pay attention to and learn information related to the phase, so that the target phase recognition model can accurately identify the phase of the CT image, avoid using artificial methods to achieve phase recognition, and thereby achieve the effect of improving the accuracy and efficiency of phase recognition of CT images.

[0132] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0134] Example 2

[0135] According to an embodiment of the present application, a method for recognizing an image is also provided, such as Figure 6 As shown, the identification method includes:

[0136] Step S601, obtaining a group of CT images to be identified;

[0137] Step S602, obtaining third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group;

[0138] Optionally, a CT image group to be identified is determined, and third position information of a second target key point in the CT image group and second embedded feature information corresponding to the CT image group are determined through template embedding feature information of multiple key points and a feature extraction model.

[0139] Step S602, using the target phase recognition model provided in the first embodiment of the present application to perform recognition based on the CT image group, the third position information and the second embedded feature information, to obtain the target phase corresponding to the CT image group.

[0140] Optionally, the third position information and the second embedded feature information corresponding to the CT image group and the CT image group are input into a target phase recognition model. The target phase recognition model performs recognition based on the CT image group, the third position information and the second embedded feature information to obtain the target phase corresponding to the CT image group.

[0141] In order to improve the accuracy of recognition, in the image recognition method provided in Example 2 of the present application, recognition is performed based on the CT image group, the third position information and the second embedded feature information through a target phase recognition model to obtain the target phase corresponding to the CT image group, including: extracting features of the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain a third feature vector; processing the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; and obtaining the target phase through the output layer in the target phase recognition model based on the second target feature vector.

[0142] Optionally, if there are other CT image groups corresponding to the CT image group, the CT image group, the third position information and the second embedded feature information are feature extracted through the self-attention mechanism in the target phase recognition model to obtain the first initial feature information corresponding to the target CT image, that is, the average HU value corresponding to the third position information is first obtained, and then the average HU value and the second embedded feature information are feature extracted to obtain the above-mentioned first initial feature information. It should be noted that when performing phase recognition through the target phase recognition model, it is not necessary to sample and occlude the identified third position information.

[0143] Then, the first initial feature information corresponding to the CT image group and the first initial feature information of other CT image groups are feature extracted through the cross-attention layer in the target phase recognition model to obtain the second initial feature information, and the above-mentioned third feature information is determined through the first initial feature information and the second initial feature information.

[0144] If there is no other CT image group corresponding to the CT image group, the first initial feature information is used as the third feature information. Finally, the third feature information is processed by the pooling layer to obtain the second target feature vector, and finally, the output layer in the target phase recognition model is used to perform recognition based on the second target feature vector to obtain the target phase.

[0145] In the image recognition method provided in the second embodiment of the present application, by obtaining a CT image group to be recognized; obtaining the third position information of the second target key point corresponding to the CT image group and the second embedded feature information corresponding to the CT image group; and using the target phase recognition model to recognize the target phase corresponding to the CT image group based on the CT image group, the third position information and the second embedded feature information, the technical problem of low efficiency of CT phase recognition caused by artificially recognizing the phase of CT in the related technology is solved. The position information of the first target key point enables the target phase recognition model to quickly pay attention to and learn information related to the phase, so that the target phase recognition model can accurately recognize the phase of the CT image, avoiding the use of artificial methods to achieve phase recognition, thereby achieving the effect of improving the accuracy and efficiency of phase recognition of CT images.

[0146] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0147] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0148] Example 3

[0149] According to an embodiment of the present application, a method for recognizing an image is also provided, such as Figure 7 As shown, the identification method includes:

[0150] Step S701, obtaining a group of CT images to be identified uploaded by a client;

[0151] Step S702, obtaining, in the cloud server, third position information of the second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; performing recognition based on the CT image group, the third position information and the second embedded feature information by the target phase recognition model to obtain the target phase corresponding to the CT image group;

[0152] Step S703: Return the target phase corresponding to the CT image group to the client.

[0153] It should be noted that the specific method for processing the problem information in the cloud server is the same as that in implementation 1 and will not be repeated here.

[0154] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0155] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0156] Example 4

[0157] According to an embodiment of the present application, a CT image recognition system for implementing the above-mentioned image recognition method is also provided. Figure 8 As shown, the CT image recognition system includes: an image acquisition module 801, a feature extraction module 802, a target phase recognition model 803 and a result output module 804.

[0158] An image acquisition module 801 is used to acquire a CT image group to be identified;

[0159] A feature extraction module 802 is used to obtain third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group;

[0160] A target phase recognition model 803, configured to perform recognition based on the CT image group, the third position information and the second embedded feature information to obtain a target phase corresponding to the CT image group;

[0161] The result output module 804 is used to display the target phase corresponding to the CT image group.

[0162] Optionally, the CT image group to be identified is acquired through the image acquisition module 801, and the feature extraction module 802 uses the template embedding feature information of multiple key points and the feature extraction model to determine the third position information of the second target key point in the CT image group and the second embedded feature information corresponding to the CT image group.

[0163] Then, the third position information and the second embedded feature information corresponding to the CT image group and the CT image group are input into the target phase recognition model. The target phase recognition model performs recognition based on the CT image group, the third position information and the second embedded feature information to obtain the target phase corresponding to the CT image group.

[0164] It should be noted that if there are other CT image groups corresponding to the CT image group (i.e., CT image groups of other phases of the target object corresponding to the CT image group), the CT image group, the third position information, and the second embedded feature information are feature extracted through the self-attention mechanism in the target phase recognition model to obtain the first initial feature information corresponding to the target CT image, that is, the average HU value corresponding to the third position information is first obtained, and then the average HU value and the second embedded feature information are feature extracted to obtain the above-mentioned first initial feature information. It should be noted that when performing phase recognition through the target phase recognition model, it is not necessary to sample and occlude the identified third position information.

[0165] Then, the first initial feature information corresponding to the CT image group and the first initial feature information of other CT image groups are feature extracted through the cross-attention layer in the target phase recognition model to obtain the second initial feature information, and the above-mentioned third feature information is determined through the first initial feature information and the second initial feature information.

[0166] If there is no other CT image group corresponding to the CT image group, the first initial feature information is used as the third feature information. Finally, the third feature information is processed by the pooling layer to obtain the second target feature vector, and finally, the output layer in the target phase recognition model is used to perform recognition based on the second target feature vector to obtain the target phase.

[0167] The CT image recognition system can enable the target phase recognition model to quickly pay attention to and learn information related to the phase, so that the target phase recognition model can accurately identify the phase of the CT image, avoiding the use of artificial methods to achieve phase recognition, thereby achieving the effect of improving the accuracy and efficiency of phase recognition of CT images.

[0168] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 2 as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 2.

[0169] Example 5

[0170] According to an embodiment of the present application, a generation device of a phase recognition model for implementing the generation method of the phase recognition model is also provided, such as Fig. 9 As shown, the generating device includes: a first acquiring unit 901, a second acquiring unit 902 and a generating unit 903.

[0171] A first acquisition unit 901 is used to acquire a plurality of sample CT image groups, wherein the sample CT images in the sample CT image groups have the same phase, and the phases corresponding to the plurality of sample CT image groups are different;

[0172] A second acquisition unit 902 is used to acquire first position information of a first target key point corresponding to a sample CT image group in a plurality of sample CT image groups and first embedded feature information corresponding to a sample CT image group in a plurality of sample CT image groups, wherein the first target key point is a key point related to the period;

[0173] A generation unit 903 is used to train an initial phase recognition model based on multiple sample CT image groups, first position information, first embedded feature information, and real phases corresponding to sample CT image groups in multiple sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image.

[0174] In the generation device of the phase recognition model provided in the fourth embodiment of the present application, a plurality of sample CT image groups are acquired by a first acquisition unit 901, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the plurality of sample CT image groups are different; a second acquisition unit 902 acquires the first position information of the first target key point corresponding to the sample CT image groups in the plurality of sample CT image groups and the first embedded feature information corresponding to the sample CT image groups in the plurality of sample CT image groups, wherein the first target key point is a key point related to the phase; a generation unit 903 trains an initial phase recognition model based on the plurality of sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the plurality of sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image, thereby solving the technical problem in the related art of manually identifying the phase of CT, resulting in relatively low efficiency in identifying the phase of CT.

[0175] In this scheme, a sample CT image group consisting of sample CT images of multiple phases is obtained, and the first position information and the first embedded feature information of the first target key point corresponding to the sample CT image are obtained. Then, the initial phase recognition model is trained with the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model. The position information of the first target key point can enable the target phase recognition model to quickly pay attention to and learn information related to the phase, so that the target phase recognition model can accurately identify the phase of the CT image, avoid using artificial methods to achieve phase recognition, and thereby achieve the effect of improving the accuracy and efficiency of phase recognition of CT images.

[0176] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the second acquisition unit includes: an acquisition subunit, used to obtain template embedded feature information of multiple key points; a first extraction subunit, used to perform feature extraction on pixel points of a sample CT image group in a plurality of sample CT image groups through a feature extraction model to obtain first embedded feature information; and a first determination subunit, used to determine the first position information based on the template embedded feature information and the first embedded feature information.

[0177] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the acquisition subunit includes: a labeling module, which is used to label the key points in the first target CT image group to obtain the second target CT image group; a first extraction module, which is used to perform global embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain global embedding features; a second extraction module, which is used to perform local embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain local embedding features; a third extraction module, which is used to perform semantic embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain semantic embedding features; a first determination module, which is used to obtain template embedding feature information based on global embedding features, local embedding features and semantic embedding features.

[0178] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the determination subunit includes: a first calculation module, used to calculate the similarity between the template embedded feature information and the first embedded feature information to obtain the target similarity; a first screening module, used to screen the pixel points of the sample CT image group in the multiple sample CT image groups according to the target similarity to obtain the first pixel point; a second screening module, used to screen the first pixel point according to a preset similarity threshold to obtain the second pixel point; and a second determination module, used to determine the first position information based on the second pixel point.

[0179] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the calculation module includes: a first calculation sub-module, used to calculate the similarity between the global embedded features in the template embedded feature information and the global embedded features in the first embedded feature information, so as to obtain a first similarity; a second calculation sub-module, used to calculate the similarity between the local embedded features in the template embedded feature information and the local embedded features in the first embedded feature information, so as to obtain a second similarity; a third calculation sub-module, used to calculate the similarity between the semantic embedded features in the template embedded feature information and the semantic embedded features in the first embedded feature information, so as to obtain a third similarity; and a first determination sub-module, used to obtain a target similarity based on the first similarity, the second similarity and the third similarity.

[0180] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the generation unit includes: a second extraction subunit, used to perform feature extraction on a sample CT image group, first position information and first embedded feature information in a plurality of sample CT image groups through an initial phase recognition model to obtain a first target feature vector; a prediction subunit, used to perform classification prediction based on the first target feature vector through the initial phase recognition model to obtain a predicted phase; a second determination subunit, used to determine a target loss function based on the predicted phase, the true phase and the first target feature vector; and a training subunit, used to train the initial phase recognition model based on the target loss function to obtain a target phase recognition model.

[0181] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the second extraction subunit includes: a fourth extraction module, which is used to perform feature extraction on the target sample CT image group, the first position information and the first embedded feature information for a target sample CT image group among multiple sample CT image groups through the self-attention mechanism in the initial phase recognition model, so as to obtain the first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups; a first acquisition module, which is used to obtain the first feature information corresponding to the sample CT image group among the multiple sample CT image groups; a fifth extraction module, which is used to perform feature extraction on the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group among the multiple sample CT image groups through the cross-attention layer in the initial phase recognition model, so as to obtain the second feature information corresponding to the target sample CT image group; a processing module, which is used to process the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group through the pooling layer in the initial phase recognition model, so as to obtain the first target feature vector.

[0182] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the fourth extraction module includes: a processing submodule, used to perform random sampling and occlusion processing on the first position information to obtain the second position information; a second determination submodule, used to obtain the average HU value corresponding to the target position information from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; an extraction submodule, used to perform feature extraction on the average HU value and the first embedded feature information to obtain the first feature information corresponding to the target sample CT image group.

[0183] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the second determination subunit includes: a second calculation module, used to perform calculations based on the predicted phase and the actual phase to obtain a classification loss function; a second acquisition module, used to obtain the order information of the target sample CT image group in the multiple sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups; a third calculation module, used to perform calculations based on the order information and the first target feature vector to obtain a contrast loss function; a third determination module, used to determine the target loss function based on the classification loss function and the contrast loss function.

[0184] Optionally, in the generation device of the phase recognition model provided in Example 4 of the present application, the third calculation module includes: an acquisition submodule, used to obtain the initial sequence vector of a sample CT image group among multiple sample CT image groups; a fourth calculation submodule, used to perform calculations based on the order information and the initial sequence vector to obtain a target sequence vector of the target sample CT image group; an encoding submodule, used to encode the target sequence vector to obtain a coded sequence vector; and a fifth calculation submodule, used to perform calculations based on the coded sequence vector and the first target feature vector to obtain a contrast loss function.

[0185] It should be noted that the first acquisition unit 901, the second acquisition unit 902 and the generation unit 903 described above correspond to steps S201 to S203 in the first embodiment, and the three units and the corresponding steps implement the same examples and application scenarios, but are not limited to the contents disclosed in the first embodiment. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in the first embodiment.

[0186] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0187] Example 6

[0188] According to an embodiment of the present application, an image recognition device for implementing the above-mentioned image recognition method is also provided. Fig.10 As shown, the device includes: a third acquisition unit 1001, a fourth acquisition unit 1002 and an identification unit 1003.

[0189] The third acquisition unit 1001 is used to acquire a CT image group to be identified;

[0190] A fourth acquisition unit 1002 is used to acquire third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group;

[0191] The identification unit 1003 is used to perform identification based on the CT image group, the third position information and the second embedded feature information through the target phase identification model to obtain the target phase corresponding to the CT image group.

[0192] Optionally, in the image recognition device provided in Example 5 of the present application, the recognition unit includes: a third extraction subunit, used to perform feature extraction on the CT image group, the third position information and the second embedded feature information through a target phase recognition model to obtain a third feature vector; a processing subunit, used to process the third feature information corresponding to the CT image group through a pooling layer in the target phase recognition model to obtain a second target feature vector; and a recognition subunit, used to perform recognition based on the second target feature vector through the output layer in the target phase recognition model to obtain a target phase.

[0193] It should be noted that the third acquisition unit 1001, the fourth acquisition unit 1002 and the identification unit 1003 described above correspond to steps S601 to S603 in the second embodiment, and the three units and the corresponding steps implement the same examples and application scenarios, but are not limited to the contents disclosed in the first embodiment. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in the first embodiment.

[0194] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 2 as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 2.

[0195] Example 7

[0196] The embodiment of the present application may provide an electronic device, which may be any electronic device in an electronic device terminal group. Optionally, in this embodiment, the electronic device may also be replaced by a terminal device such as a mobile terminal.

[0197] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0198] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the method provided in Embodiment 1 and Embodiment 2: obtaining multiple sample CT image groups, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the multiple sample CT image groups are different; obtaining the first position information of the first target key point corresponding to the sample CT image groups in the multiple sample CT image groups and the first embedded feature information corresponding to the sample CT image groups in the multiple sample CT image groups, wherein the first target key point is a key point related to the phase; training an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image.

[0199] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: Obtaining the first position information of the first target key point corresponding to the sample CT image group in multiple sample CT image groups and the first embedded feature information corresponding to the sample CT image group in multiple sample CT image groups includes: obtaining the template embedded feature information of multiple key points; performing feature extraction on the pixel points of the sample CT image group in the multiple sample CT image groups through the feature extraction model to obtain the first embedded feature information; determining the first position information based on the template embedded feature information and the first embedded feature information.

[0200] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: Acquiring the template embedding feature information of multiple key points includes: marking the key points in the first target CT image group to obtain the second target CT image group; performing global embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain global embedding features; performing local embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain local embedding features; performing semantic embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain semantic embedding features; obtaining template embedding feature information based on the global embedding features, local embedding features and semantic embedding features.

[0201] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: Determining the first position information based on the template embedded feature information and the first embedded feature information includes: performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain target similarity; screening pixel points of a sample CT image group in a plurality of sample CT image groups based on the target similarity to obtain a first pixel point; screening the first pixel point based on a preset similarity threshold to obtain a second pixel point; and determining the first position information based on the second pixel point.

[0202] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain the target similarity includes: performing similarity calculation on the global embedded features in the template embedded feature information and the global embedded features in the first embedded feature information to obtain the first similarity; performing similarity calculation on the local embedded features in the template embedded feature information and the local embedded features in the first embedded feature information to obtain the second similarity; performing similarity calculation on the semantic embedded features in the template embedded feature information and the semantic embedded features in the first embedded feature information to obtain the third similarity; and obtaining the target similarity based on the first similarity, the second similarity and the third similarity.

[0203] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: training the initial phase recognition model based on multiple sample CT image groups, the first position information, the first embedded feature information and the real phase corresponding to the sample CT image groups in the multiple sample CT image groups to obtain the target phase recognition model, including: extracting features of the sample CT image groups, the first position information and the first embedded feature information in the multiple sample CT image groups through the initial phase recognition model to obtain the first target feature vector; performing classification prediction based on the first target feature vector through the initial phase recognition model to obtain the predicted phase; determining the target loss function based on the predicted phase, the real phase and the first target feature vector; training the initial phase recognition model based on the target loss function to obtain the target phase recognition model.

[0204] The computer terminal can execute the program code of the following steps in the method provided in the first and second embodiments: extracting features from a sample CT image group, first position information and first embedded feature information in a plurality of sample CT image groups through an initial phase recognition model to obtain a first target feature vector, including: for a target sample CT image group in a plurality of sample CT image groups, extracting features from the target sample CT image group, first position information and first embedded feature information through a self-attention mechanism in the initial phase recognition model to obtain first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any one group of sample CT image groups in the plurality of sample CT image groups; obtaining first feature information corresponding to a sample CT image group in the plurality of sample CT image groups; extracting features from the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group in the plurality of sample CT image groups through a cross-attention layer in the initial phase recognition model to obtain second feature information corresponding to the target sample CT image group; processing the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group through a pooling layer in the initial phase recognition model to obtain a first target feature vector.

[0205] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: feature extraction is performed on the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model to obtain the first feature information corresponding to the target sample CT image group, including: random sampling and occlusion processing are performed on the first position information to obtain the second position information; for the target position information in the second position information, the average HU value corresponding to the target position information is obtained from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; feature extraction is performed on the average HU value and the first embedded feature information to obtain the first feature information corresponding to the target sample CT image group.

[0206] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: Determining the target loss function based on the predicted period, the true period and the first target feature vector includes: calculating based on the predicted period and the true period to obtain a classification loss function; for a target sample CT image group in multiple sample CT image groups, obtaining the order information of the target sample CT image group in the multiple sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups in the multiple sample CT image groups; calculating based on the order information and the first target feature vector to obtain a contrast loss function; determining the target loss function based on the classification loss function and the contrast loss function.

[0207] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: Calculating based on the order information and the first target feature vector to obtain the contrast loss function includes: obtaining an initial order vector of a sample CT image group in multiple sample CT image groups; calculating based on the order information and the initial order vector to obtain a target order vector of the target sample CT image group; encoding the target order vector to obtain an encoding order vector; calculating based on the encoding order vector and the first target feature vector to obtain a contrast loss function.

[0208] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: obtaining the CT image group to be identified; obtaining the third position information of the second target key point corresponding to the CT image group and the second embedded feature information corresponding to the CT image group; and identifying based on the CT image group, the third position information and the second embedded feature information through any of the above-mentioned target phase recognition models to obtain the target phase corresponding to the CT image group.

[0209] The above-mentioned computer terminal can execute the program code of the following steps in the method provided in Example 1 and Example 2: identifying based on the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain the target phase corresponding to the CT image group, including: extracting features of the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain a third feature vector; processing the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; identifying based on the second target feature vector through the output layer in the target phase recognition model to obtain the target phase.

[0210] Optionally, Fig.11 is a structural block diagram of an electronic device according to an embodiment of the present application. Fig.11 As shown, the electronic device 1100 may include: one or more ( Fig.11 Only one is shown in the figure) processor 1102, memory 1104. The electronic device 1100 may also include a storage controller, through which the memory 1104 is controlled and managed; the electronic device 1100 may also include a peripheral interface, through which the radio frequency module, audio module and display screen are connected.

[0211] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and devices provided in Embodiments 1 and 2 of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the methods provided in Embodiments 1 and 2 are realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the electronic device 100 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0212] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain multiple sample CT image groups, wherein the phases corresponding to the sample CT images in the sample CT image groups are the same, and the phases corresponding to the multiple sample CT image groups are different; obtain the first position information of the first target key point corresponding to the sample CT image groups in the multiple sample CT image groups and the first embedded feature information corresponding to the sample CT image groups in the multiple sample CT image groups, wherein the first target key point is a key point related to the phase; train an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to identify the phase corresponding to the CT image.

[0213] Optionally, the processor may also execute program code of the following steps: obtaining first position information of a first target key point corresponding to a sample CT image group in a plurality of sample CT image groups and first embedded feature information corresponding to a sample CT image group in a plurality of sample CT image groups includes: obtaining template embedded feature information of a plurality of key points; performing feature extraction on pixel points of a sample CT image group in a plurality of sample CT image groups through a feature extraction model to obtain first embedded feature information; and determining first position information based on the template embedded feature information and the first embedded feature information.

[0214] Optionally, the processor may also execute the program code of the following steps: obtaining the template embedding feature information of multiple key points includes: marking the key points in the first target CT image group to obtain the second target CT image group; performing global embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain the global embedding feature; performing local embedding feature extraction on the key points in the second target CT image group through the feature extraction model to obtain the local embedding feature; performing semantic embedding feature extraction on the key points in the second target CT image group to obtain the semantic embedding feature; obtaining the template embedding feature information based on the global embedding feature, the local embedding feature and the semantic embedding feature.

[0215] Optionally, the processor may also execute program code of the following steps: determining the first position information based on the template embedded feature information and the first embedded feature information includes: performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain target similarity; screening pixel points of a sample CT image group in a plurality of sample CT image groups based on the target similarity to obtain a first pixel point; screening the first pixel point based on a preset similarity threshold to obtain a second pixel point; and determining the first position information based on the second pixel point.

[0216] Optionally, the processor may also execute program code of the following steps: performing similarity calculation on the template embedded feature information and the first embedded feature information to obtain a target similarity, including: performing similarity calculation on the global embedded features in the template embedded feature information and the global embedded features in the first embedded feature information to obtain a first similarity; performing similarity calculation on the local embedded features in the template embedded feature information and the local embedded features in the first embedded feature information to obtain a second similarity; performing similarity calculation on the semantic embedded features in the template embedded feature information and the semantic embedded features in the first embedded feature information to obtain a third similarity; and obtaining the target similarity based on the first similarity, the second similarity and the third similarity.

[0217] Optionally, the processor may also execute the program code of the following steps: training an initial phase recognition model based on multiple sample CT image groups, first position information, first embedded feature information and the true phase corresponding to the sample CT image groups in the multiple sample CT image groups to obtain a target phase recognition model, including: extracting features of the sample CT image groups, first position information and first embedded feature information in the multiple sample CT image groups through the initial phase recognition model to obtain a first target feature vector; performing classification prediction based on the first target feature vector through the initial phase recognition model to obtain a predicted phase; determining a target loss function based on the predicted phase, the true phase and the first target feature vector; and training the initial phase recognition model based on the target loss function to obtain a target phase recognition model.

[0218] Optionally, the processor may also execute the program code of the following steps: extracting features from a sample CT image group, first position information and first embedded feature information in a plurality of sample CT image groups through an initial phase recognition model to obtain a first target feature vector, including: for a target sample CT image group in a plurality of sample CT image groups, extracting features from the target sample CT image group, first position information and first embedded feature information through a self-attention mechanism in the initial phase recognition model to obtain first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any group of sample CT image groups in the plurality of sample CT image groups; obtaining first feature information corresponding to a sample CT image group in the plurality of sample CT image groups; extracting features from the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image group in the plurality of sample CT image groups through a cross-attention layer in the initial phase recognition model to obtain second feature information corresponding to the target sample CT image group; processing the first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group through a pooling layer in the initial phase recognition model to obtain a first target feature vector.

[0219] Optionally, the processor may also execute the program code of the following steps: performing feature extraction on the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model to obtain the first feature information corresponding to the target sample CT image group, including: performing random sampling and occlusion processing on the first position information to obtain the second position information; for the target position information in the second position information, obtaining the average HU value corresponding to the target position information from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; performing feature extraction on the average HU value and the first embedded feature information to obtain the first feature information corresponding to the target sample CT image group.

[0220] Optionally, the processor may also execute the program code of the following steps: determining the target loss function based on the predicted phase, the true phase and the first target feature vector, including: performing calculations based on the predicted phase and the true phase to obtain a classification loss function; for a target sample CT image group in a plurality of sample CT image groups, obtaining order information of the target sample CT image group in the plurality of sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups among the plurality of sample CT image groups; performing calculations based on the order information and the first target feature vector to obtain a contrast loss function; and determining the target loss function based on the classification loss function and the contrast loss function.

[0221] Optionally, the processor may also execute program code of the following steps: The processor may also execute program code of the following steps: Calculating based on the order information and the first target feature vector to obtain a contrast loss function includes: obtaining an initial order vector of a sample CT image group in multiple sample CT image groups; calculating based on the order information and the initial order vector to obtain a target order vector of the target sample CT image group; encoding the target order vector to obtain an encoded order vector; calculating based on the encoded order vector and the first target feature vector to obtain a contrast loss function.

[0222] Optionally, the processor may also execute the program code of the following steps: obtaining a CT image group to be identified; obtaining the third position information of a second target key point corresponding to the CT image group and the second embedded feature information corresponding to the CT image group; and performing identification based on the CT image group, the third position information and the second embedded feature information using any one of the target phase recognition models mentioned above to obtain the target phase corresponding to the CT image group.

[0223] Optionally, the processor may also execute the program code of the following steps: extracting features of the CT image group, the third position information and the second embedded feature information through the target phase recognition model to obtain a third feature vector; processing the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; and identifying based on the second target feature vector through the output layer in the target phase recognition model to obtain a target phase.

[0224] It can be understood by those skilled in the art that Fig.11 The structure shown is for illustration only, and the electronic device 1100 may also be a terminal device such as a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), or a PAD. Fig.11 The structure of the electronic device is not limited. Fig.11 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Fig.11 Different configurations shown.

[0225] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0226] Example 8

[0227] The embodiment of the present application also provides a computer program product. Optionally, in this embodiment, the computer program product can be used to store the program code executed by the method for generating the phase recognition model provided in the first embodiment, or the program code executed by the method for recognizing the image provided in the second embodiment.

[0228] Optionally, in this embodiment, the computer program product may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0229] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0230] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0231] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0232] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0233] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0234] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc., which can store program code.

[0235] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for generating a phase recognition model, characterized in that: include: Acquire a plurality of sample CT image groups, wherein the sample CT images in the sample CT image groups correspond to the same phase, and the plurality of sample CT image groups correspond to different phases; Acquire first position information of a first target key point corresponding to a sample CT image group among the plurality of sample CT image groups and first embedded feature information corresponding to a sample CT image group among the plurality of sample CT image groups, wherein the first target key point is a key point related to the phase; Training an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information, and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to generate a target phase recognition model, wherein the target phase recognition model is used to recognize the phase corresponding to the CT image; Wherein, the first embedding feature information is composed of global embedding features, local embedding features and semantic embedding features; Wherein, training the initial phase recognition model according to the multiple sample CT image groups, the first position information, the first embedded feature information and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to obtain the target phase recognition model includes: extracting features from the sample CT image groups in the multiple sample CT image groups, the first position information and the first embedded feature information through the initial phase recognition model to obtain a first target feature vector, wherein the first target feature vector is used to train the initial phase recognition model to obtain the target phase recognition model; For a target sample CT image group among the multiple sample CT image groups, feature extraction is performed on the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model to obtain first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups, and the first feature information is used to obtain a first target feature vector; Extracting features from the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model to obtain first feature information corresponding to the target sample CT image group includes: performing random sampling and occlusion processing on the first position information to obtain second position information; For the target position information in the second position information, taking the target position information as the center, obtaining an average HU value corresponding to the target position information from the target sample CT image group, wherein the target position information is any one of the position information in the second position information; Feature extraction is performed on the average HU value and the first embedded feature information to obtain first feature information corresponding to the target sample CT image group.

2. The method according to claim 1, characterized in that Acquiring first position information of a first target key point corresponding to a sample CT image group among the multiple sample CT image groups and first embedded feature information corresponding to a sample CT image group among the multiple sample CT image groups includes: Obtain template embedding feature information of multiple key points; Extracting features from pixel points of a sample CT image group in the plurality of sample CT image groups by using a feature extraction model to obtain the first embedded feature information; The first position information is determined according to the template embedding feature information and the first embedding feature information.

3. The method according to claim 2, characterized in that The template embedding feature information of multiple key points is obtained including: Marking key points in the first target CT image group to obtain a second target CT image group; Performing global embedding feature extraction on key points in the second target CT image group by using the feature extraction model to obtain global embedding features; Extracting local embedded features from key points in the second target CT image group using the feature extraction model to obtain local embedded features; Extracting semantic embedding features of key points in the second target CT image group by using the feature extraction model to obtain semantic embedding features; The template embedding feature information is obtained according to the global embedding feature, the local embedding feature and the semantic embedding feature.

4. The method according to claim 2, characterized in that: Determining the first position information according to the template embedding feature information and the first embedding feature information includes: Calculating similarity between the template embedded feature information and the first embedded feature information to obtain a target similarity; Screening pixel points of a sample CT image group in the plurality of sample CT image groups according to the target similarity to obtain a first pixel point; The first pixel points are screened according to a preset similarity threshold to obtain a second pixel point; The first position information is determined according to the second pixel point.

5. The method according to claim 4, characterized in that Calculating the similarity between the template embedded feature information and the first embedded feature information to obtain a target similarity includes: Calculating similarity between the global embedding feature in the template embedding feature information and the global embedding feature in the first embedding feature information to obtain a first similarity; Calculating similarity between the local embedding features in the template embedding feature information and the local embedding features in the first embedding feature information to obtain a second similarity; Calculating similarity between the semantic embedding feature in the template embedding feature information and the semantic embedding feature in the first embedding feature information to obtain a third similarity; The target similarity is obtained according to the first similarity, the second similarity and the third similarity.

6. The method according to claim 1, characterized in that Training an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information, and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups to obtain a target phase recognition model includes: Performing classification prediction based on the first target feature vector by the initial phase recognition model to obtain a predicted phase; Determining a target loss function according to the predicted phase, the true phase and the first target feature vector; The initial phase recognition model is trained according to the target loss function to obtain the target phase recognition model.

7. The method according to claim 6, characterized in that Performing feature extraction on the sample CT image groups, the first position information and the first embedded feature information in the plurality of sample CT image groups by using the initial phase recognition model to obtain a first target feature vector includes: Acquire first feature information corresponding to a sample CT image group among the plurality of sample CT image groups; Extracting features of the first feature information corresponding to the target sample CT image group and the first feature information corresponding to the sample CT image groups in the multiple sample CT image groups through the cross-attention layer in the initial phase recognition model, obtaining second feature information corresponding to the target sample CT image group; The first feature information corresponding to the target sample CT image group and the second feature information corresponding to the target sample CT image group are processed through the pooling layer in the initial phase recognition model to obtain the first target feature vector.

8. The method according to claim 6, characterized in that Determining a target loss function according to the predicted phase, the true phase and the first target feature vector includes: Calculating according to the predicted phase and the true phase to obtain a classification loss function; For a target sample CT image group among the multiple sample CT image groups, obtaining order information of the target sample CT image group among the multiple sample CT image groups, wherein the target sample CT image group is any group of sample CT image groups among the multiple sample CT image groups; Calculating according to the order information and the first target feature vector to obtain a contrast loss function; The target loss function is determined according to the classification loss function and the contrast loss function.

9. The method according to claim 8, characterized in that The contrast loss function is obtained by calculating according to the order information and the first target feature vector: acquiring an initial sequence vector of a sample CT image group among the plurality of sample CT image groups; Calculating according to the order information and the initial order vector to obtain a target order vector of the target sample CT image group; Performing encoding processing on the target sequence vector to obtain an encoding sequence vector; The contrast loss function is obtained by performing calculation based on the encoding order vector and the first target feature vector.

10. A method for recognizing an image, characterized in that: include: Acquire a group of CT images to be identified; Acquire third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; The target phase recognition model is used to identify the target phase corresponding to the CT image group based on the CT image group, the third position information and the second embedded feature information, wherein the target phase recognition model is obtained by using the phase recognition model generation method described in any one of claims 1 to 9.

11. The method according to claim 10, characterized in that The target phase corresponding to the CT image group is obtained by identifying the target phase recognition model based on the CT image group, the third position information and the second embedded feature information, including: Performing feature extraction on the CT image group, the third position information and the second embedded feature information by using the target phase recognition model to obtain a third feature vector; Processing the third feature information corresponding to the CT image group through the pooling layer in the target phase recognition model to obtain a second target feature vector; The target phase is obtained by performing identification based on the second target feature vector through the output layer in the target phase identification model.

12. A method for recognizing an image, characterized in that: include: Obtaining the CT image group to be identified uploaded by the client; Acquire the third position information of the second target key point corresponding to the CT image group and the second embedded feature information corresponding to the CT image group in the cloud server, and identify the target phase corresponding to the CT image group through the target phase recognition model based on the CT image group, the third position information and the second embedded feature information to obtain the target phase corresponding to the CT image group, wherein the target phase recognition model is obtained by using the generation method of the phase recognition model described in any one of claims 1 to 9; The target phase corresponding to the CT image group is returned to the client.

13. A CT image recognition system, characterized in that: include: An image acquisition module, used for acquiring a group of CT images to be identified; a feature extraction module, used for acquiring third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; a target phase recognition model, used for performing recognition based on the CT image group, the third position information and the second embedded feature information to obtain a target phase corresponding to the CT image group, wherein the target phase recognition model is obtained by using the phase recognition model generation method according to any one of claims 1 to 9; The result output module is used to display the target phase corresponding to the CT image group.

14. A device for generating a phase recognition model, characterized in that: include: A first acquisition unit is used to acquire a plurality of sample CT image groups, wherein the sample CT images in the sample CT image groups have the same phase, and the phases corresponding to the plurality of sample CT image groups are different; a second acquisition unit, configured to acquire first position information of a first target key point corresponding to a sample CT image group among the plurality of sample CT image groups and first embedded feature information corresponding to a sample CT image group among the plurality of sample CT image groups, wherein the first target key point is a key point related to the phase; A generating unit, configured to train an initial phase recognition model based on the multiple sample CT image groups, the first position information, the first embedded feature information, and the real phases corresponding to the sample CT image groups in the multiple sample CT image groups, to generate a target phase recognition model, wherein the target phase recognition model is used to recognize the phase corresponding to the CT image; Wherein, the first embedding feature information is composed of global embedding features, local embedding features and semantic embedding features; The generating unit comprises: a second extracting subunit, configured to perform feature extraction on the sample CT image groups, the first position information and the first embedded feature information in the plurality of sample CT image groups through the initial phase recognition model to obtain a first target feature vector, wherein the first target feature vector is used to train the initial phase recognition model to obtain a target phase recognition model; The second extraction subunit includes: a fourth extraction module, used for performing feature extraction on a target sample CT image group among the multiple sample CT image groups, the target sample CT image group, the first position information and the first embedded feature information through the self-attention mechanism in the initial phase recognition model, to obtain first feature information corresponding to the target sample CT image group, wherein the target sample CT image group is any one of the multiple sample CT image groups, and the first feature information is used to obtain a first target feature vector; The fourth extraction module includes: a processing submodule, which is used to perform random sampling and occlusion processing on the first position information to obtain second position information; a second determination submodule, which is used to obtain, for the target position information in the second position information, an average HU value corresponding to the target position information from the target sample CT image group with the target position information as the center, wherein the target position information is any one of the position information in the second position information; and an extraction submodule, which is used to perform feature extraction on the average HU value and the first embedded feature information to obtain first feature information corresponding to the target sample CT image group.

15. An image recognition device, characterized in that: include: A third acquisition unit is used to acquire a CT image group to be identified; A fourth acquisition unit, used to acquire third position information of a second target key point corresponding to the CT image group and second embedded feature information corresponding to the CT image group; An identification unit is used to identify the target phase corresponding to the CT image group based on the CT image group, the third position information and the second embedded feature information through a target phase identification model, wherein the target phase identification model is obtained by using the phase identification model generation method described in any one of claims 1 to 9.

16. An electronic device, characterized in that: include: A memory storing an executable program; A processor for running the program, wherein the program, when running, executes the method for generating a phase recognition model as described in any one of claims 1 to 9, or the method for recognizing an image as described in any one of claims 10 to 11.

17. A computer program product, characterized in that The computer program product includes a stored computer program, and when the computer program is executed by a processor, it implements the method for generating a phase recognition model as described in any one of claims 1 to 9, or the method for recognizing an image as described in any one of claims 10 to 11.

Images