Cloud platform-based medical image information management system

Abstract

The application relates to the field of medical information, and specifically discloses a medical image information management system based on a cloud platform, which, when performing compatibility evaluation on a preselected medical institution, introduces software version, imaging protocol and image processing technology information in addition to obtaining medical image acquisition equipment information, and adopts semantic embedding coding technology based on deep learning to respectively perform semantic coding on target image information acquisition equipment information and each preselected medical institution image acquisition information to capture the implied relationship and context information between the data, and further performs semantic query matching coding on the target image acquisition equipment information and the preselected medical institution image acquisition information to mine the significant correlation features between the two, so as to realize compatibility evaluation on the preselected medical institution. In this way, the accuracy of the compatibility degree evaluation on the preselected medical institution can be improved, and the medical service quality and efficiency are further improved.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and more specifically, to a cloud-based medical image information management system. Background Technology

[0002] Medical imaging information management systems play a vital role in the modern medical system. In the traditional medical system, the diagnosis and treatment systems of various hospitals are relatively independent, and the medical imaging information of each hospital is mostly limited to the internal medical institution, making it difficult to achieve cross-institutional data sharing. As a result, when patients are transferred to other hospitals for treatment, medical imaging information needs to be collected again, which not only wastes patients' time and examination fees, but also reduces the quality of medical services.

[0003] Chinese patent CN117316392A proposes a medical image information management system and method based on smart healthcare. It obtains target image information that matches the patient's condition from a medical image cloud database based on the patient's basic information, and judges the compatibility of the target image information in the corresponding medical institution by judging the similarity between the acquisition device of the target image information and the acquisition device of the target medical institution, thereby preventing the duplicate acquisition of medical images when the information is shared.

[0004] However, when conducting compatibility assessments for medical institutions, the above-mentioned scheme only uses a simple similarity calculation formula to calculate the degree of similarity between medical image acquisition devices as the basis for compatibility assessment. This method mainly relies on hardware information such as device model, brand, and specifications, while ignoring software-level compatibility, such as the version and protocol support of image processing software, which may result in incomplete and inaccurate assessment results.

[0005] Therefore, there is a need for an optimized cloud-based medical image information management system. Summary of the Invention

[0006] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a cloud-based medical image information management system. When conducting compatibility assessments of pre-selected medical institutions, in addition to acquiring information about the medical image acquisition equipment, it also incorporates software version, imaging protocol, and image processing technology information. Furthermore, it employs deep learning-based semantic embedding coding technology to semantically encode the acquisition equipment information of the target image information and the acquisition information of images from each pre-selected medical institution to capture implicit relationships and contextual information between the data. Secondly, it further performs semantic query matching coding on the acquisition equipment information of the target image information and the acquisition information of images from the pre-selected medical institutions to uncover significant correlation features between them, thereby achieving compatibility assessment of the pre-selected medical institutions. This approach improves the accuracy of compatibility assessments of pre-selected medical institutions, provides a scientific basis for the optimal allocation and sharing of medical resources, and further improves the quality and efficiency of medical services.

[0007] According to one aspect of this application, a cloud-based medical image information management system is provided, comprising:

[0008] The medical image information storage module is used to collect medical image information and store the medical image information in the cloud database of the cloud platform;

[0009] The target image information acquisition module is used to acquire acquisition device information of target image information from the medical image cloud database based on the patient's basic information.

[0010] The pre-selected medical institution image information acquisition module is used to acquire the acquisition information of images from pre-selected medical institutions, wherein the acquisition information includes acquisition device information, software version, imaging protocol and image processing technology;

[0011] The semantic feature extraction module is used to perform semantic embedding encoding on each data item in the acquisition device information of the target image information and the acquisition information of the preselected medical institution images to obtain a sequence of semantic embedding encoding vectors for the target image acquisition device information and the preselected medical institution image information.

[0012] A query encoding module is used to perform query encoding on the sequence of semantic embedding encoding vectors of the target image acquisition device information and the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information to obtain target image query response enhancement encoding features. This module includes: a salient feature extraction unit, used to extract salient features from the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information to obtain a semantic guidance benchmark; and a heterogeneous feature enhancement query encoding unit, used to perform heterogeneous feature enhancement query encoding on the sequence of semantic embedding encoding vectors of the target image acquisition device information and the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information based on the semantic guidance benchmark to obtain a target image query response enhancement encoding vector as the target image query response enhancement encoding feature.

[0013] The compatibility assessment module is used to evaluate the compatibility of the pre-selected medical institutions based on the enhanced coding features of the target image query response.

[0014] Compared with existing technologies, this application provides a cloud-based medical image information management system. When conducting compatibility assessments of pre-selected medical institutions, in addition to acquiring information about the medical image acquisition equipment, it also incorporates software version, imaging protocol, and image processing technology information. Furthermore, it employs deep learning-based semantic embedding coding technology to semantically encode the acquisition equipment information of the target image information and the acquisition information of images from each pre-selected medical institution to capture implicit relationships and contextual information between the data. Secondly, it further performs semantic query matching coding on the acquisition equipment information of the target image information and the acquisition information of images from the pre-selected medical institutions to uncover significant correlation features between them, thereby achieving compatibility assessment of the pre-selected medical institutions. This approach improves the accuracy of compatibility assessments of pre-selected medical institutions, provides a scientific basis for the optimal allocation and sharing of medical resources, and further improves the quality and efficiency of medical services. Attached Figure Description

[0015] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 This is a block diagram of a cloud-based medical image information management system according to an embodiment of this application;

[0017] Figure 2 This is a schematic diagram of data flow in a cloud-based medical image information management system according to an embodiment of this application;

[0018] Figure 3 This is a block diagram of the query encoding module in a cloud-based medical image information management system according to an embodiment of this application. Detailed Implementation

[0019] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0020] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0021] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0022] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0023] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0024] As mentioned in the background section above, patent CN117316392A proposes a medical image information management system and method based on smart healthcare, which includes: collecting and storing medical image information; establishing a medical image cloud database; the image information includes patient basic information, image acquisition equipment information, image content information, and image acquisition medical institution information; obtaining target image information corresponding to the patient's basic information based on the medical image cloud database; determining a licensed medical institution based on the target image information; the patient determining the target medical institution based on the licensed medical institution; reading the specific content of the target image information at the target medical institution based on the patient's basic information; medical personnel at the target medical institution diagnosing the patient's condition based on the specific content of the target image information, obtaining the patient's diagnosis result; and the patient understanding their condition based on the medical image information and the patient's diagnosis result.

[0025] The above scheme only uses a simple similarity calculation formula to calculate the similarity between medical image acquisition devices when conducting compatibility assessments of medical institutions. This method mainly relies on hardware information such as device model, brand, and specifications, while ignoring software-level compatibility, such as the version and protocol support of image processing software. As a result, the assessment results may not be comprehensive or accurate.

[0026] Based on this, the technical solution of this application proposes a medical image information management system based on a cloud platform. Figure 1 This is a block diagram of a cloud-based medical image information management system according to an embodiment of this application. Figure 2 This is a schematic diagram illustrating the data flow of a cloud-based medical image information management system according to an embodiment of this application. Figure 1 and Figure 2As shown, the cloud-based medical image information management system 300 according to an embodiment of this application includes: a medical image information storage module 310, used to collect medical image information and store the medical image information in a cloud database of the cloud platform; a target image information acquisition module 320, used to acquire acquisition device information of target image information from the medical image cloud database based on patient basic information; a pre-selected medical institution image information acquisition module 330, used to acquire acquisition information of images from pre-selected medical institutions, wherein the acquisition information includes acquisition device information, software version, imaging protocol and image processing technology; and a semantic feature extraction module 340. The system is configured to perform semantic embedding encoding on each data item in the acquisition device information of the target image information and the acquisition information of the pre-selected medical institution images to obtain a sequence of semantic embedding encoding vectors for the target image acquisition device information and the pre-selected medical institution image information; the query encoding module 350 is configured to perform query encoding on the sequence of semantic embedding encoding vectors for the target image acquisition device information and the pre-selected medical institution image information to obtain enhanced encoding features for the target image query response; and the compatibility evaluation module 360 ​​is configured to evaluate the compatibility of the pre-selected medical institutions based on the enhanced encoding features for the target image query response.

[0027] Specifically, the medical image information storage module 310 is used to acquire medical image information and store it in a cloud-based medical image database on a cloud platform. The medical image information includes the patient's basic information. The medical image cloud database refers to a database system that stores medical image data on a cloud platform. Storing the medical image information in the cloud-based medical image database ensures that all medical image information is completely and accurately preserved, such as the information of the acquisition device used to capture the medical image. In particular, the cloud platform supports data sharing among multiple medical institutions, promoting the optimized allocation and collaborative work of medical resources.

[0028] Specifically, the target image information acquisition module 320 is used to obtain the acquisition device information of the target image information from the medical image cloud database based on the patient's basic information. That is, in the cloud-based medical image information management system of this application, based on the patient's basic information (such as name, ID number, medical record number, etc.), the system can accurately identify and locate the medical image information of a specific patient. In one example, the patient extracts the acquisition device information of their medical image by inputting personal information on the platform, which serves as the acquisition device information of the target image information, providing support for subsequent data processing and analysis. It should be understood that obtaining the acquisition device information of the target image information from the medical image cloud database based on the patient's basic information is existing technology. For example, it can use the principle disclosed in Chinese Patent CN117316392A to obtain the acquisition device information of the target image information; of course, it can also use other principles to obtain the information, and this is not limited to this application.

[0029] Specifically, the pre-selected medical institution image information acquisition module 330 is used to acquire acquisition information of images from the pre-selected medical institution. This acquisition information includes acquisition equipment information, software version, imaging protocol, and image processing technology. In the technical solution of this application, the pre-selected medical institution is a medical institution connected to the medical image cloud database. Here, the acquisition equipment information, software version, imaging protocol, and image processing technology reflect the equipment and technical status of the pre-selected medical institution. Through cross-domain comprehensive analysis of the aforementioned pre-selected medical institution image information, the comprehensiveness of the analysis of the pre-selected medical institution image data can be ensured, providing a comprehensive reference for subsequent compatibility assessment.

[0030] Specifically, the semantic feature extraction module 340 is used to perform semantic embedding encoding on each data item in the acquisition device information of the target image information and the acquisition information of the pre-selected medical institution images to obtain a sequence of semantic embedding encoding vectors for the target image acquisition device information and the pre-selected medical institution image information. Considering that in medical image information management, different medical institutions may have different data formats, equipment models, software versions, etc., these differences make it difficult to directly compare and share data. Traditional data processing methods are usually limited to a single data domain, making it difficult to achieve cross-institutional data sharing and interoperability. In particular, semantic embedding encoding can cross different data domains and achieve a unified representation of data from different sources, providing a foundation for cross-domain queries. Therefore, in the technical solution of this application, the acquisition device information of the target image information and each data item in the acquisition information of the pre-selected medical institution images are input into a semantic encoder based on the BERT model to obtain a sequence of semantic embedding encoding vectors for the target image acquisition device information and the pre-selected medical institution image information. Specifically, the BERT model is a deep learning model based on the Transformer architecture, which excels at capturing implicit relationships and contextual information in text. By inputting the information from the data collection devices and the data collection information from the pre-selected medical institutions into the BERT model, the complex implicit relationships and contextual information between these information can be effectively captured, providing rich semantic representations. It can also achieve data standardization and unified representation between different data domains, providing strong support for cross-domain queries and compatibility assessments.

[0031] Specifically, the query encoding module 350 is used to perform query encoding on the sequence of the semantic embedding encoding vector of the target image acquisition device information and the semantic embedding encoding vector of the pre-selected medical institution image information to obtain enhanced encoding features for the target image query response. It is worth mentioning that, to further improve the accuracy of the evaluation, this application proposes a heterogeneous feature enhancement query encoding method based on salient feature guidance. This method uses salient features in the sequence of the semantic embedding encoding vector of the pre-selected medical institution image information as guidance to strengthen the feature parts in the semantic embedding encoding vector of the target image acquisition device information that are significantly related to the image information of the pre-selected medical institution, thereby optimizing the query process and improving the efficiency and accuracy of the query encoding. In a specific example of this application, such as... Figure 3As shown, the query encoding module 350 includes: a salient feature extraction unit 351, used to extract salient features from the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information to obtain a semantic guidance benchmark; and a heterogeneous feature enhancement query encoding unit 352, used to perform heterogeneous feature enhancement query encoding on the sequence of semantic embedding encoding vectors of the target image acquisition device information and the pre-selected medical institution image information based on the semantic guidance benchmark to obtain a target image query response enhancement encoding vector as the target image query response enhancement encoding feature.

[0032] Specifically, the salient feature extraction unit 351 is used to extract salient features from the sequence of semantic embedding encoding vectors of pre-selected medical institution image information to obtain a semantic guidance benchmark. In a specific example of this application, the salient feature extraction unit includes: a key matrix construction subunit, used to construct a key matrix based on the sequence of semantic embedding encoding vectors of pre-selected medical institution image information; and a maximum value extraction subunit, used to extract the maximum value of each key vector in the key matrix to obtain a salient feature vector of the key matrix as a semantic guidance benchmark. Specifically, firstly, in order to capture the relationship between each semantic embedding encoding vector of pre-selected medical institution image information in the sequence of semantic embedding encoding vectors of pre-selected medical institution image information, so as to better represent the complex patterns in the data, the key embedding matrix is ​​used to perform a linear transformation on each semantic embedding encoding vector of pre-selected medical institution image information in the sequence of semantic embedding encoding vectors of pre-selected medical institution image information, so as to map it to a new latent feature space, making it more suitable for subsequent query encoding process. Next, the sequence of semantic embedding encoding vectors of pre-selected medical institution image information after linear transformation is arranged in order as a key matrix to facilitate parallel processing.

[0033] More specifically, the key matrix construction subunit includes: a linear embedding encoding second-level subunit, used to perform linear embedding encoding on each of the pre-selected medical institution image information semantic embedding encoding vectors in the sequence of pre-selected medical institution image information semantic embedding encoding vectors using the key embedding matrix to obtain a sequence of linearly transformed pre-selected medical institution image information semantic embedding encoding vectors; and a matrix arrangement second-level subunit, used to arrange the sequence of linearly transformed pre-selected medical institution image information semantic embedding encoding vectors into a matrix using the linearly transformed pre-selected medical institution image information semantic embedding encoding vectors as key vectors to obtain the key matrix. Specifically, this process can be expressed by the following formula:

[0034] K = {k1,k2,...,k} n}

[0035] M o ={k1′;k2′;...;k n ′}

[0036] k i ′=f(k i W k )=k i W k +b k

[0037] Wherein, K represents the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information, k1, k2, k... i and k n These are the first, second, i-th, and n-th semantic embedding encoding vectors of the pre-selected medical institution image information, respectively, where n is the number of the pre-selected medical institution image information semantic embedding encoding vectors. k Let b represent the key embedding matrix, respectively. k M represents the key embedding bias vector, respectively. o Denotes the key matrix, k1′, k2′, k i ′ and k n ′ are the first, second, i-th, and n-th key vectors in the key matrix, respectively, which are the semantic embedding encoding vectors of the pre-selected medical institution image information after linear transformation.

[0038] The linear embedding coding second-level subunit is used to: multiply each of the preselected medical institution image information semantic embedding coding vectors in the sequence of preselected medical institution image information semantic embedding coding vectors by the key embedding matrix, and then add it to the key embedding bias vector positionally to obtain the sequence of preselected medical institution image information semantic embedding coding vectors after linear transformation.

[0039] Then, the maximum value among the key vectors in the key matrix is ​​selected to obtain the salient feature vector of the key matrix. This salient feature vector is then used as a semantic guidance benchmark, representing the most prominent or representative part of the information in the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information after linear transformation. Introducing this salient feature vector as an additional guidance signal helps reduce noise interference, strengthens the model's learning of important patterns, and also helps the attention mechanism focus more on the key parts of the data, thereby improving the model's attention to important details.

[0040] Specifically, the calculation process for extracting the maximum value subunit can be expressed by the following formula:

[0041] v tip ={max(k1′);max(k2′);...;max(k n ′)}

[0042] Where max(·) represents the function that takes the maximum value, v tip This refers to the semantic guidance benchmark.

[0043] Specifically, the heterogeneous feature enhancement query encoding unit 352 is used to perform heterogeneous feature enhancement query encoding on the sequence of the semantic embedding encoding vector of the target image acquisition device information and the semantic embedding encoding vector of the pre-selected medical institution image information based on the semantic guidance benchmark, so as to obtain the target image query response enhancement encoding vector as the target image query response enhancement encoding feature. In a specific example of this application, the heterogeneous feature enhancement query encoding unit includes: a target image acquisition device information linear embedding encoding subunit, used to perform linear embedding encoding on the target image acquisition device information semantic embedding encoding vector using a query embedding matrix and a value embedding matrix to obtain a value vector and a query vector; a heterogeneous feature enhancement query encoding subunit, used to input the query vector, the value vector, each linearly transformed pre-selected medical institution image information semantic embedding encoding vector in the key matrix and the semantic guidance benchmark into a feature-guided cross-domain attention transformation layer to obtain a sequence of heterogeneous feature enhancement query encoding vectors; and a query response enhancement encoding subunit, used to calculate the positional mean vector of the sequence of heterogeneous feature enhancement query encoding vectors to obtain the query response enhancement encoding vector as the target image query response encoding feature. Here, query embedding matrices and value embedding matrices are applied to the semantic embedding encoding vector of the target image acquisition device information to generate corresponding query vectors and value vectors. A specially designed heterogeneous converter structure is used to interactively encode the query vector, the value vector, each key vector in the key matrix, and the semantic guidance benchmark. By introducing an additional semantic guidance benchmark into the traditional attention mechanism, the weight allocation strategy is automatically adjusted, effectively optimizing the semantic embedding encoding vector of the target image acquisition device information. Finally, the sequence of heterogeneous feature-enhanced query encoding vectors is averaged according to position to obtain a single query response-enhanced encoding vector, thereby effectively mining the feature correlation distribution information between the target image information and the image information of each pre-selected medical institution.

[0044] Specifically, the linear embedding coding subunit for target image acquisition device information is used to: multiply the semantic embedding coding vector of the target image acquisition device information by the query embedding matrix and then add it positionally to the query embedding bias vector to obtain the query vector; multiply the semantic embedding coding vector of the target image acquisition device information by the value embedding matrix and then add it positionally to the value embedding bias vector to obtain the value vector. Specifically, this process can be expressed by the following formula:

[0045] v q =x1W q +bq

[0046] v v =x1W v +b v

[0047] Where x1 represents the semantic embedding encoding vector of the target image acquisition device information, W q and W v These represent the query embedding matrix and the value embedding matrix, respectively. q and b v These represent the query embedding bias vector and the value embedding bias vector, respectively. q and v v These represent the query vector and the value vector, respectively.

[0048] More specifically, the heterogeneous feature-enhanced query encoding subunit is used to: multiply the query vector by the transpose of the key vector and divide it by the L2 norm of the semantic guidance benchmark to obtain an attention score matrix; multiply the attention score matrix by the semantic guidance benchmark after applying a softmax function to obtain a template hint optimization attention weight vector; and calculate the positional multiplication between the template hint optimization attention weight vector and the value vector to obtain the heterogeneous feature-enhanced query encoding vector. Specifically, this process can be expressed by the following formula:

[0049]

[0050] in,(·) T Let ||·||² represent the transpose of a vector, ||·||² represent the L2 norm of the vector, and softmax is the normalization exponential function. Represents matrix multiplication, ⊙ represents positional multiplication, v pi This represents the i-th heterogeneous feature enhancement query encoding vector in the sequence of heterogeneous feature enhancement query encoding vectors.

[0051] Specifically, the operation process of the query response enhancement coding subunit can be expressed by the following formula:

[0052]

[0053] Among them, v p This represents the heterogeneous feature-enhanced query encoding vector.

[0054] Specifically, the compatibility evaluation module 360 ​​is used to evaluate the compatibility level of the pre-selected medical institutions based on the enhanced coding features of the target image query response. In a specific example of this application, the enhanced coding vector of the target image query response is input into a decoder-based compatibility evaluation module to obtain a compatibility evaluation value. That is, by fully learning the target image query response information contained in the enhanced coding vector of the target image query response, the enhanced coding vector of the target image query response is decoded to map it to the compatibility space of the pre-selected medical institutions, thereby generating a compatibility evaluation value.

[0055] In the technical solution of this application, the semantic embedding encoding vector of the target image acquisition device information represents the semantic embedding encoding feature of the target image acquisition device information. The semantic embedding encoding vector of the acquisition device information, the semantic embedding encoding vector of the software version, the semantic embedding encoding vector of the imaging protocol, and the semantic embedding encoding vector of the image processing technology respectively represent the semantic embedding encoding features of the acquisition device information of the pre-selected medical institution, the semantic embedding encoding features of the software version, the semantic embedding encoding features of the imaging protocol, and the semantic embedding encoding features of the image processing technology. Thus, when the sequence of the semantic embedding encoding vector of the target image acquisition device information and the feature vector composed of the semantic embedding encoding vector of the acquisition device information, the semantic embedding encoding vector of the software version, the semantic embedding encoding vector of the imaging protocol, and the semantic embedding encoding vector of the image processing technology are input into the query encoding optimization module with the salient features of the key matrix as the prompt template, the sparsity of the prompt template of the feature vector sequence, coupled with the correlation differences between the semantic embedding encoding vector of the target image acquisition device information and the various semantic embedding encoding vectors in the sequence of feature vectors, will cause the feature manifold of the target image query response optimization encoding vector to have sparse expression and uneven distribution in the high-dimensional feature space, thereby affecting the accuracy of the compatibility evaluation value obtained by the compatibility evaluation module based on the decoder.

[0056] Based on this, in a preferred embodiment, the optimized encoding vector of the target image query response is input into a decoder-based compatibility evaluation module to obtain a compatibility evaluation value, including:

[0057] The target image query response optimization coding vector is subjected to feature modulation relative to the target decoding domain to obtain an adapted target image query response optimization coding vector.

[0058] The adapted target image query response optimized encoding vector is input into the decoder-based compatibility evaluation module to obtain the compatibility evaluation value.

[0059] Specifically, in this preferred embodiment, the process of performing feature modulation relative to the target decoding domain on the target image query response optimization coding vector to obtain a suitable target image query response optimization coding vector includes the following steps:

[0060] Calculate the global topological relation mapping of the target image query response optimization encoding vector to obtain the target image query response optimization global topological relation matrix, which is expressed as:

[0061]

[0062] Among them, v i and v j Let represent the i-th and j-th elements in the target image query response optimized encoding vector, respectively, and D1 represent the global topological relation matrix of the first target image. Let D1 represent the (i,j)th element in the global topological relation matrix of the first target image, and D2 represent the global topological relation matrix of the second target image. This represents the (i,j)th element in the global topological relation matrix of the second target image. represents matrix multiplication, and M represents the global topology relation matrix for target image query response optimization.

[0063] Extract the decoding weight matrix M from the decoder-based compatibility evaluation module. d ;

[0064] The target image query response global topology relation matrix is ​​optimized and the decoding weight matrix is ​​correlated with a fine-grained response to obtain the target image query response global topology-decoding joint matrix, which is expressed as:

[0065] M t =Sigmoid(W t (αM+βM d ))

[0066] Where α represents the first weight hyperparameter, β represents the second weight hyperparameter, and M d W represents the decoding weight matrix. t M represents the pre-trained weight matrix, Sigmoid represents the sigmoid activation function, and M... t This represents the global topology-decoding joint matrix of the target image query response.

[0067] Based on the global topology-decoding joint matrix of the target image query response, a recursive self-consistent decoding is performed on the optimized encoding vector of the target image query response to obtain the target image query recursive decoding optimized vector, which is expressed as:

[0068]

[0069] Where V represents the target image query response optimized encoding vector, V t This represents the target image query recursive decoding optimization vector.

[0070] Based on the target image query recursive decoding optimization vector, a target image query recursive decoding optimization weight vector is constructed, represented as follows:

[0071] V a =Softmax(V t )

[0072] Where Softmax represents the normalized exponential function, V a This represents the weight vector for recursive decoding optimization of the target image query.

[0073] The target image query response optimization encoding vector is obtained by feature modulation of the target image query recursive decoding optimization weight vector, as shown below:

[0074] V f =V a ⊙V

[0075] Where ⊙ represents dot product by position, V f This represents the optimized encoding vector for the target image query response.

[0076] Accordingly, in this preferred embodiment, the global topological relationship mapping of the target image query response optimization encoding vector is semantically fused with the decoding weight matrix of the decoder. Then, based on the information matrix of the fused target decoding information, the target image query response optimization encoding vector is optimized for multi-body coupling trajectory based on recursive self-consistent coding to improve the detail and structure of feature expression, thereby improving the accuracy of the compatibility evaluation value obtained by the compatibility evaluation module based on the decoder.

[0077] As described above, the cloud-based medical image information management system 300 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with cloud-based medical image information management algorithms. In one possible implementation, the cloud-based medical image information management system 300 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the cloud-based medical image information management system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the cloud-based medical image information management system 300 can also be one of many hardware modules of the wireless terminal.

[0078] Alternatively, in another example, the cloud-based medical image information management system 300 and the wireless terminal can also be separate devices, and the cloud-based medical image information management system 300 can connect to the wireless terminal via wired and / or wireless networks and transmit interactive information in accordance with an agreed data format.

[0079] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A cloud-based medical image information management system, characterized in that, include: The medical image information storage module is used to collect medical image information and store the medical image information in the cloud database of the cloud platform; The target image information acquisition module is used to acquire acquisition device information of target image information from the medical image cloud database based on the patient's basic information. The pre-selected medical institution image information acquisition module is used to acquire the acquisition information of images from pre-selected medical institutions, wherein the acquisition information includes acquisition device information, software version, imaging protocol and image processing technology; The semantic feature extraction module is used to perform semantic embedding encoding on each data item in the acquisition device information of the target image information and the acquisition information of the preselected medical institution images to obtain a sequence of semantic embedding encoding vectors for the target image acquisition device information and the preselected medical institution image information. A query encoding module is used to perform query encoding on the sequence of semantic embedding encoding vectors of the target image acquisition device information and the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information to obtain target image query response enhancement encoding features. This module includes: a salient feature extraction unit, used to extract salient features from the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information to obtain a semantic guidance benchmark; and a heterogeneous feature enhancement query encoding unit, used to perform heterogeneous feature enhancement query encoding on the sequence of semantic embedding encoding vectors of the target image acquisition device information and the sequence of semantic embedding encoding vectors of the pre-selected medical institution image information based on the semantic guidance benchmark to obtain a target image query response enhancement encoding vector as the target image query response enhancement encoding feature. The heterogeneous feature-enhanced query encoding unit includes: The target image acquisition device information linear embedding coding subunit is used to perform linear embedding coding on the semantic embedding coding vector of the target image acquisition device information using a query embedding matrix and a value embedding matrix to obtain a value vector and a query vector. The heterogeneous feature-enhanced query encoding subunit is used to input the query vector, the value vector, the linearly transformed semantic embedding encoding vectors of preselected medical institution image information in the key matrix, and the semantic guidance benchmark into a feature-guided cross-domain attention transformation layer to obtain a sequence of heterogeneous feature-enhanced query encoding vectors. The key matrix is ​​constructed based on the sequence of the semantic embedding encoding vectors of preselected medical institution image information. The query response enhancement coding subunit is used to calculate the positional mean vector of the sequence of the heterogeneous feature enhancement query coding vectors to obtain the query response enhancement coding vector as the query response coding feature of the target image; The compatibility assessment module is used to evaluate the compatibility of the pre-selected medical institutions based on the enhanced coding features of the target image query response.

2. The cloud-based medical image information management system according to claim 1, characterized in that, The semantic feature extraction module is used for: Each data item in the acquisition device information of the target image information and the acquisition information of the pre-selected medical institution images is input into a semantic encoder based on the Bert model to obtain a sequence of semantic embedding encoding vectors for the target image acquisition device information and the pre-selected medical institution image information.

3. The cloud-based medical image information management system according to claim 2, characterized in that, The salient feature extraction unit includes: A key matrix construction subunit is used to construct a key matrix based on the sequence of semantic embedding encoding vectors of the image information of the preselected medical institutions; The maximum value extraction subunit is used to extract the maximum value of each key vector in the key matrix to obtain the significant feature vector of the key matrix as a semantic guidance benchmark.

4. The cloud-based medical image information management system according to claim 3, characterized in that, The key matrix construction subunit includes: A linear embedding coding second-level subunit is used to perform linear embedding coding on each of the pre-selected medical institution image information semantic embedding coding vectors in the sequence of pre-selected medical institution image information semantic embedding coding vectors using a key embedding matrix to obtain a sequence of pre-selected medical institution image information semantic embedding coding vectors after linear transformation. The matrix arrangement second-level subunit is used to arrange the sequence of the linearly transformed preselected medical institution image information semantic embedding encoding vectors into a matrix to obtain the key matrix, using the linearly transformed preselected medical institution image information semantic embedding encoding vectors as key vectors.

5. The cloud-based medical image information management system according to claim 4, characterized in that, The linear embedding coding second-level subunit is used for: The sequence of semantic embedding encoding vectors for pre-selected medical institution image information is obtained by multiplying each pre-selected medical institution image information semantic embedding encoding vector in the sequence by the key embedding matrix and then adding it to the key embedding bias vector positionally.

6. The cloud-based medical image information management system according to claim 5, characterized in that, The target image acquisition device information linear embedding encoding subunit is used for: The query vector is obtained by multiplying the semantic embedding encoding vector of the target image acquisition device information by the query embedding matrix and then adding it to the query embedding bias vector in terms of position. The value vector is obtained by multiplying the semantic embedding encoding vector of the target image acquisition device information by the value embedding matrix and then adding it to the value embedding bias vector positionally.

7. The cloud-based medical image information management system according to claim 6, characterized in that, The heterogeneous feature-enhanced query encoding subunit is used for: The attention score matrix is ​​obtained by multiplying the query vector by the transpose of the key vector and then dividing by the L2 norm of the semantic guidance benchmark. The attention score matrix is ​​passed through a softmax function and then multiplied by the semantic guidance benchmark to obtain the template cueing optimized attention weight vector; The heterogeneous feature-enhanced query encoding vector is obtained by calculating the positional dot product between the template hint optimization attention weight vector and the value vector.

8. The cloud-based medical image information management system according to claim 7, characterized in that, The compatibility evaluation module is used to input the target image query response enhancement encoding vector into the decoder-based compatibility evaluation module to obtain a compatibility evaluation value.

Images