A medical auxiliary review method and system of image recognition

Abstract

The application discloses a kind of medical auxiliary review methods and systems of image recognition, it is related to image processing technical field, the method includes: interactive and hospital information system through interface connection medical imaging equipment, export target user in preset window image frame sequence;Cross-frame feature association memory identification update is carried out to image frame sequence, and target association memory unit is obtained;Target user queuing level is used as target auxiliary review information;Interactive queuing management module of hospital information system, obtain target matching department;In combination with the target auxiliary review information, the queuing application of target user is added in the queue to be queued for optimization, and target queuing queue is obtained.The application solves the technical problems that the image data of user is deeply mined in the prior art, resulting in low intelligent triage efficiency of hospital information system, poor auxiliary review reliability, achieves to improve the accuracy of auxiliary review material, improves the technical effect of triage queuing efficiency and quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a medical auxiliary review method and system for image recognition. BACKGROUND

[0002] The current hospital information system (HIS) usually has functions of patient information management, registration, queuing, diagnosis and cost settlement, and part of the system has supported data docking with image acquisition equipment to realize the archiving and viewing of patient image information. However, most of the existing systems only stay at the level of image management or image browsing, and have not realized the auxiliary judgment and intelligent triage based on image content. Especially in the process of registration and queuing, it still mainly depends on the user's self-selection of department, and lacks systematic analysis support for the severity of the disease or structural abnormalities contained in the image, resulting in that some patients with serious conditions cannot be accurately identified and cannot obtain timely and preferential treatment. SUMMARY

[0003] The present application provides a medical auxiliary review method and system for image recognition, which is used to solve the technical problems that the existing technology lacks in-depth mining of user image data, resulting in low intelligent triage efficiency of the hospital information system and poor reliability of auxiliary review.

[0004] In view of the above problems, the present application provides a medical auxiliary review method and system for image recognition.

[0005] In a first aspect of the present application, a medical auxiliary review method for image recognition is provided, which comprises:

[0006] interacting with a medical imaging device connected to the hospital information system through an interface, exporting an image frame sequence of a target user within a preset window; traversing the image frame sequence to perform cross-frame feature association memory recognition update, obtaining a target association memory unit; calling a target user queuing level recognizer to perform information extraction on the target association memory unit, determining a target user queuing level, taking the target user queuing level as target auxiliary review information; interacting with a queuing management module of the hospital information system, performing department matching based on the target auxiliary review information, obtaining a target matching department; performing a queuing queue search based on the target matching department, combining the target auxiliary review information, adding a queuing application of the target user into a queuing queue for optimization, and obtaining a target queuing queue.

[0007] Preferably, based on the target matching department, the queue searching is performed, the target user's queuing application is added into the queue for optimization in combination with the target auxiliary review information, the target queuing queue is obtained, including: traversing the queue for auxiliary review information extraction, obtaining the queue for auxiliary review information; in the order from high to low of the queuing level, the target auxiliary review information and the queue for auxiliary review information are optimized to obtain the target queuing queue.

[0008] Preferably, the cross-frame feature association memory recognition update is performed on the image frame sequence to obtain a target association memory unit, including:

[0009] The image frame sequence is respectively subjected to feature extraction according to a preset image feature set to obtain an image frame feature vector sequence, wherein the preset image feature set includes structural texture features, edge information, shape features and gray density features;

[0010] The cross-frame feature association memory update is performed based on the image frame feature vector sequence to obtain a target association memory unit.

[0011] Preferably, the image frame sequence is respectively subjected to feature extraction according to a preset image feature set to obtain an image frame feature vector sequence, including: obtaining a sample image frame set, extracting the structural texture features, edge information, shape features and gray density features of each sample image frame in the sample image frame set based on the preset image feature set, and filling into an initially empty vector to obtain a sample image frame feature vector set; obtaining a feature extractor, wherein the feature extractor is constructed according to the mapping relationship between the sample image frame set and the sample image frame feature vector set; and using the feature extractor to perform feature extraction on the image frame sequence to obtain the image frame feature vector sequence.

[0012] Preferably, the cross-frame feature association memory update is performed based on the image frame feature vector sequence to obtain a target association memory unit, including: extracting a first image frame feature vector and a second image frame feature vector from the image frame feature vector sequence; performing cross-frame feature association analysis on the first image frame feature vector and the second image frame feature vector to obtain a second associated image frame feature vector, and adding the first image frame feature vector and the second associated image frame feature vector into a first association memory unit; again extracting a third image frame feature vector from the image frame feature vector sequence, and using the first association memory unit to perform cross-frame feature association memory update on the third image frame feature vector to obtain a second association memory unit; based on the second association memory unit, the remaining image frame feature vectors of the image frame feature vector sequence are sequentially subjected to cross-frame feature association analysis, and the second association memory unit is updated according to the analysis result to obtain a target association memory unit.

[0013] Preferably, the first image frame feature vector and the second image frame feature vector are subjected to cross-frame feature correlation analysis to obtain a second correlation image frame feature vector, including: calculating the similarity of corresponding elements of the first image frame feature vector and the second image frame feature vector to obtain an element similarity set; normalizing the element similarity set and matrixing the processed result to obtain a correlation analysis matrix; and using the correlation analysis matrix to enhance the second image frame feature vector to obtain the second correlation image frame feature vector.

[0014] Preferably, the method further includes: using the correlation analysis matrix to perform weighted fusion on the second image frame feature vector to obtain a second enhanced image frame feature vector; and using a weighted average method to fuse the second enhanced image frame feature vector and the second image frame feature vector to obtain the second correlation image frame feature vector.

[0015] Preferably, a third image frame feature vector is extracted again from the sequence of image frame feature vectors, and the first correlation memory unit is used to update the third image frame feature vector to obtain a second correlation memory unit, including: using the first image frame feature vector in the first correlation memory unit to perform cross-frame feature correlation analysis on the third image frame feature vector to obtain a first initial third correlation image frame feature vector; using the second correlation image frame feature vector in the first correlation memory unit to perform cross-frame feature correlation analysis on the third image frame feature vector to obtain a second initial third correlation image frame feature vector; performing mean value processing on the first initial third correlation image frame feature vector and the second initial third correlation image frame feature vector to obtain a third correlation image frame feature vector, and adding the third correlation image frame feature vector to the first correlation memory unit to obtain the second correlation memory unit.

[0016] Preferably, the target auxiliary review information is verified, and the network parameter of the target user queuing level identifier is optimized according to the verification result.

[0017] In a second aspect, the present application provides a medical auxiliary review system for image recognition, including:

[0018] The image frame sequence derivation module is used for deriving an image frame sequence of a target user within a preset window through interaction with a medical imaging device connected to a hospital information system through an interface; the target association memory unit obtaining module is used for performing cross-frame feature association memory identification updating on the image frame sequence to obtain a target association memory unit; the target auxiliary review information obtaining module is used for calling a target user queuing level identifier to perform information extraction on the target association memory unit, determining a target user queuing level, and taking the target user queuing level as target auxiliary review information; the target matching department obtaining module is used for interacting with a queuing management module of the hospital information system, performing department matching based on the target auxiliary review information, and obtaining a target matching department; and the target queuing queue obtaining module is used for performing a queuing queue search based on the target matching department, combining the target auxiliary review information, adding a queuing application of the target user into a queuing queue for optimization, and obtaining a target queuing queue.

[0019] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0020] The present application derives an image frame sequence of a target user within a preset window through interaction with a medical imaging device connected to a hospital information system through an interface, then performs cross-frame feature association memory identification updating on the image frame sequence to obtain a target association memory unit, further calls a target user queuing level identifier to perform information extraction on the target association memory unit, determines a target user queuing level, takes the target user queuing level as target auxiliary review information, then interacts with a queuing management module of the hospital information system, performs department matching based on the target auxiliary review information, obtains a target matching department, further performs a queuing queue search based on the target matching department, combines the target auxiliary review information, adds a queuing application of the target user into a queuing queue for optimization, and obtains a target queuing queue. The technical effect of improving the dynamic trend analysis accuracy of image recognition is achieved, and further reliable auxiliary review information is provided to provide reliable data support for queuing queue optimization. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0022] Figure 1 A medical auxiliary review method flowchart of image recognition is provided for the embodiments of the present application.

[0023] Figure 2A structure schematic diagram of a medical auxiliary review system of image recognition provided by an embodiment of the present application.

[0024] Reference signs: image frame sequence derivation module 11, target association memory unit obtaining module 12, target auxiliary review information obtaining module 13, target matching department obtaining module 14, and target queuing queue obtaining module 15. DETAILED DESCRIPTION

[0025] The present application provides a medical auxiliary review method and system of image recognition, which is used to solve the technical problems of low intelligent triage efficiency and poor auxiliary review reliability of a hospital information system caused by deep mining of image data of a user in the prior art.

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0027] It should be noted that the terms “include” and “have” are intended to cover the non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units need not be limited to those clearly listed steps or units, but can include other steps or modules not clearly listed or inherent to the process, method, product or device.

[0028] Embodiment one, as shown in the present application provides a medical auxiliary review method of image recognition, wherein the method comprises: Figure 1

[0029] Step S100: interact with a medical imaging device connected to a hospital information system through an interface, and derive an image frame sequence of a target user in a preset window;

[0030] In one possible embodiment, the medical imaging device is an X-ray machine, a CT (computed tomography), an MRI (magnetic resonance imaging), an ultrasound instrument, or other devices capable of obtaining structural images. The preset window is a time period set by a person skilled in the art in advance, which can be half a month, a month, etc. By deriving the images collected by the medical imaging device in the preset window, the image frame sequence is obtained. The image frame sequence reflects the fluctuation of the target user's condition in the preset window.

[0031] ​By connecting with the PACS system or the image acquisition server and the medical imaging device, image frame sequences in a preset window are extracted and transmitted to the hospital information system for subsequent analysis and processing, thereby providing a basis for analysis data for subsequent cross-frame feature correlation analysis. This step plays a role in data input source control in the entire method and directly affects the context accuracy of feature extraction and the cross-frame correlation quality of memory update.

[0032] Step S200: traversing the image frame sequence to update the cross-frame feature correlation memory recognition, and obtaining a target correlation memory unit;

[0033] Further, traversing the image frame sequence to update the cross-frame feature correlation memory recognition, and obtaining a target correlation memory unit, comprises:

[0034] According to the preset image feature set, the image frame sequence is respectively subjected to feature extraction, and an image frame feature vector sequence is obtained, wherein the preset image feature set comprises structural texture features, edge information, shape features, and gray density features.

[0035] Based on the image frame feature vector sequence, the cross-frame feature correlation memory is updated, and a target correlation memory unit is obtained.

[0036] Further, according to the preset image feature set, the image frame sequence is respectively subjected to feature extraction, and an image frame feature vector sequence is obtained, and the embodiment of the application step S200 further comprises:

[0037] A sample image frame set is obtained, and structural texture features, edge information, shape features, and gray density features of each sample image frame in the sample image frame set are respectively extracted based on the preset image feature set and filled into an initially empty vector to obtain a sample image frame feature vector set.

[0038] A feature extractor is obtained, wherein the feature extractor is constructed according to the mapping relationship between the sample image frame set and the sample image frame feature vector set.

[0039] The image frame sequence is subjected to feature extraction by using the feature extractor, and the image frame feature vector sequence is obtained.

[0040] In one possible embodiment, the preset image feature set refers to a set of multidimensional feature items defined by a person skilled in the art in advance for identifying key contents of medical images, including structural texture features, edge information, shape features, and gray density features. The sample image frame set is a set of labeled images for training the feature extractor. The feature vector is a structure for encoding various image features into a vector representation. The feature extractor is a trained model for receiving images as input and outputting multidimensional feature vectors.

[0041] A set of sample image frames is obtained. Then, a structural texture feature, edge information, shape feature and gray-density feature of each sample image frame in the set of sample image frames are extracted by a person skilled in the art according to the set of preset image features, and the extraction results are filled into an initially empty vector, thereby obtaining a corresponding set of sample image frame feature vectors, wherein the vector representation is as shown in Table 1.

[0042] Table 1: Sample image frame feature vector table

[0043]

[0044] Preferably, the set of sample image frames and the set of sample image frame feature vectors are used as training data sets, and a framework based on a convolutional neural network is supervised trained by using the training data sets. In the training, a one-to-one mapping relationship between the sample image frames and the sample image frame feature vectors is learned until the training converges, and the trained feature extractor is obtained.

[0045] Then, the feature extractor is used to extract features from the image frame sequence one by one, and the extracted image frame feature vector sequence is obtained. Preferably, the image frame feature vector sequence reflects the representation of the image frame sequence.

[0046] By batch processing of the image frame sequence for feature extraction, the technical effect of providing data support for subsequent processing is achieved.

[0047] Further, based on the image frame feature vector sequence, cross-frame feature association memory updating is performed to obtain a target association memory unit. The step S200 of the embodiment of the present application further includes:

[0048] A first image frame feature vector and a second image frame feature vector are extracted from the image frame feature vector sequence;

[0049] The first image frame feature vector and the second image frame feature vector are subjected to cross-frame feature association analysis to obtain a second associated image frame feature vector, and the first image frame feature vector and the second associated image frame feature vector are added to a first association memory unit;

[0050] A third image frame feature vector is extracted from the image frame feature vector sequence again, and the first association memory unit is used to perform cross-frame feature association memory updating on the third image frame feature vector to obtain a second association memory unit;

[0051] Based on the second association memory unit, the remaining image frame feature vectors of the image frame feature vector sequence are subjected to cross-frame feature association analysis in turn, and the second association memory unit is updated according to the analysis results to obtain a target association memory unit.

[0052] In a possible embodiment, the sequence of image frame feature vectors refers to an ordered set of features extracted from consecutive image frames by a feature extractor, embodying the information of structure, texture, shape, gray scale, etc. of the image changing over time. Cross-frame feature association analysis refers to the process of comparing, fusing or strengthening feature vectors from different image frames to identify common features or evolving features between frames. The associated image frame feature vector is an enhanced feature representation obtained by analysis between two frames, with stronger temporal semantics and context consistency. The associated memory unit is essentially a memory structure that stores and updates the cross-frame feature structure, used to progressively integrate significant features between multiple frames.

[0053] In a possible embodiment, the sequence of image frame feature vectors is analyzed frame by frame by an iterative operation, and the memory structure is constructed. First, a first image frame feature vector and a second image frame feature vector are extracted from the sequence of image frame feature vectors, and a second associated image frame feature vector is obtained by performing cross-frame feature association analysis. The second associated image frame feature vector contains the implicit trend of element change from the first image frame feature vector to the second image frame feature vector. Then, the first image frame feature vector and the second associated image frame feature vector are added to the first associated memory unit. The first associated memory unit is used for subsequent cross-frame feature association analysis.

[0054] Next, a third associated image frame feature vector is extracted from the sequence of image frame feature vectors, and the third associated image frame feature vector is further updated based on the existing content (the first image frame feature vector and the second associated image frame feature vector) in the first associated memory unit, to output a new enhanced feature and extend to form a second associated memory unit. This process continues in an iterative manner, and the remaining image frames are analyzed and fused frame by frame, and the memory unit is updated, to finally obtain a target associated memory unit representing the associated expression of the entire image sequence.

[0055] Through inter-frame association semantic mining and step-by-step knowledge accumulation, dynamic change trends can be identified, thereby enhancing the ability of the target user queuing level identifier in time sequence perception and context analysis, and further laying a solid foundation for generating high-confidence auxiliary review information.

[0056] Further, the first image frame feature vector and the second image frame feature vector are subjected to cross-frame feature association analysis to obtain a second associated image frame feature vector, and the step S200 of the embodiment of the application further includes:

[0057] The similarity of the corresponding elements of the first image frame feature vector and the second image frame feature vector is calculated to obtain a set of element similarities;

[0058] normalize the element similarity set, and matrix the processed result to obtain a correlation analysis matrix;

[0059] The second image frame feature vector is enhanced by using the correlation analysis matrix to obtain a second correlation image frame feature vector.

[0060] Further, the step S200 of the embodiment of the present application further comprises:

[0061] The second image frame feature vector is enhanced by using the correlation analysis matrix to obtain a second correlation image frame feature vector.

[0062] The second image frame feature vector is enhanced by using the correlation analysis matrix to obtain a second correlation image frame feature vector.

[0063] In an embodiment of the present application, the similarity degree of the corresponding elements of the first image frame feature vector and the second image frame feature vector is calculated by using a cosine similarity calculation formula to obtain an element similarity set. Further, the element similarity set is normalized by using a min-max linear normalization processing method to obtain an element normalized similarity set. Further, the element normalized similarity set is added to an initially empty matrix to obtain a correlation analysis matrix. The correlation analysis matrix reflects the correlation degree of different elements between the first image frame feature vector and the second image frame feature vector.

[0064] Further, each dimension feature is weighted and calculated according to the weight in the correlation analysis matrix by matrix operation to obtain the second enhanced image frame feature vector. The second enhanced image frame feature vector reflects the feature enhancement of the second image frame feature vector under the influence of the first image frame feature vector. Further, the second enhanced image frame feature vector and the second image frame feature vector are weighted and averaged to fuse the independent feature case and the affected feature case to obtain the second correlation image frame feature vector. By obtaining the second correlation image frame feature vector, the technical effect of improving the inter-frame semantic continuity and context consistency is achieved, and higher quality information input is provided for subsequent construction of a memory unit.

[0065] Further, a third image frame feature vector is extracted again from the sequence of image frame feature vectors, and the first correlation memory unit is used for cross-frame feature correlation memory updating of the third image frame feature vector to obtain a second correlation memory unit. The step S200 of the embodiment of the present application further comprises:

[0066] The third image frame feature vector is subjected to cross-frame feature correlation analysis by using the first image frame feature vector in the first correlation memory unit to obtain a first initial third correlation image frame feature vector.

[0067] perform cross-frame feature correlation analysis on the third image frame feature vector based on the second associated image frame feature vector in the first associated memory unit to obtain a second initial third associated image frame feature vector;

[0068] perform mean value processing on the first initial third associated image frame feature vector and the second initial third associated image frame feature vector to obtain a third associated image frame feature vector, and add the third associated image frame feature vector into the first associated memory unit to obtain a second associated memory unit.

[0069] In the embodiments of the present application, based on the same principle as obtaining the second associated image frame feature vector, cross-frame feature correlation analysis is performed on the third image frame feature vector based on the first image frame feature vector in the first associated memory unit to obtain a first initial third associated image frame feature vector. The first initial third associated image frame feature vector implies the correlation between the third image frame feature vector and the first image frame feature vector. Similarly, cross-frame feature correlation analysis is performed on the third image frame feature vector based on the second associated image frame feature vector in the first associated memory unit to obtain a second initial third associated image frame feature vector. The second initial third associated image frame feature vector implies the correlation between the third image frame feature vector and the second associated image frame feature vector.

[0070] Further, the third associated image frame feature vector is obtained by performing mean value processing on the first initial third associated image frame feature vector and the second initial third associated image frame feature vector. The third associated image frame feature vector reflects the implied features of the third image frame, and the third associated image frame feature vector is added into the first associated memory unit to update it to obtain a second associated memory unit.

[0071] Step S400: calling the target user queuing level identifier to perform information extraction on the target associated memory unit, determining the target user queuing level, and taking the target user queuing level as the target auxiliary review information.

[0072] Further, the target auxiliary review information is verified, and the network parameter of the target user queuing level identifier is optimized according to the verification result.

[0073] In a possible embodiment, a plurality of sample target association memory units and a plurality of sample target user queuing grades are acquired as training corpus. The framework constructed based on the feedforward neural network is supervised trained by using the training corpus, and the result output in the training is verified, and when the verification is passed, the target user queuing grade identifier trained is obtained. The target user queuing grade identifier is a deep neural network with learning ability, which is used for discriminant analysis on semantic features of the target association memory unit to extract the target user queuing grade.

[0074] Preferably, the training corpus is input into the framework to obtain a plurality of output target user queuing grades, the plurality of output target user queuing grades are compared with the plurality of sample target user queuing grades, and a proportion of successful comparison is obtained. When the proportion is greater than or equal to a proportion preset by a person skilled in the art, the training is completed.

[0075] In a possible embodiment, when the proportion is less than or equal to the proportion preset by the person skilled in the art, the network parameters of the target user queuing grade identifier are updated and adjusted according to the difference between the proportions until the requirement is met. The technical effect of improving the efficiency and reliability of the auxiliary review is achieved.

[0076] Step S400: The queuing management module of the interactive hospital information system performs department matching based on the target auxiliary review information to obtain a target matching department.

[0077] Step S500: The target matching department is searched based on the target matching department, and the queuing application of the target user is added into the queuing queue for optimization based on the target auxiliary review information to obtain a target queuing queue.

[0078] Further, the target matching department is searched based on the target matching department, and the queuing application of the target user is added into the queuing queue for optimization based on the target auxiliary review information to obtain a target queuing queue. The step S500 of the embodiment of the application further includes:

[0079] The queuing auxiliary review information queue is obtained by traversing the queuing queue to extract the queuing auxiliary review information.

[0080] The target auxiliary review information and the queuing auxiliary review information queue are optimized in a descending order of queuing grades to obtain the target queuing queue.

[0081] In one possible embodiment, the target auxiliary review information obtained after image recognition is mapped and interacted with department information in the hospital queuing management module. For example, when the target auxiliary review information is a cerebral hemorrhage sign detected in the image, the semantic similarity of the cerebral hemorrhage sign is matched with a set of department keywords preset by a person skilled in the art, the department keyword corresponding to the maximum semantic similarity is taken as the matching department keyword, and then the department corresponding to the matching department keyword, such as neurosurgery, is taken as the target matching department. After obtaining the target matching department, a clear direction is provided for the next queuing optimization.

[0082] In the list of departments to be queued in the target matching department, a search is performed to form a queue to be queued. Further, the auxiliary review information of all users is extracted from the queue to be queued, and the information is sorted into a queue of auxiliary review information to be queued, wherein each auxiliary review information to be queued reflects the queuing level of each user. Next, the auxiliary review information of the target user is compared and sorted in priority with the queue of auxiliary review information to be queued. The sorting manner is from high to low in level. For example, the user with the level of “emergency” is queued before the user with the level of “general”, so as to realize the priority guarantee of medical resources. The final output of the new sequence is the target queuing queue. By using the auxiliary review result for queuing logic optimization, the medical record image of the user is deeply mined, and the target of improving the triage efficiency is achieved.

[0083] In summary, the embodiments of the present application have at least the following technical effects:

[0084] The present application obtains the image frame sequence of the target user in the preset window through the interface connection between the medical imaging device and the hospital information system, then traverses the image frame sequence to perform cross-frame feature association memory recognition update, obtains a target association memory unit, and then calls a target user queuing level identifier to perform information extraction on the target association memory unit, determines the queuing level of the target user, takes the queuing level of the target user as target auxiliary review information, then interacts with the queuing management module of the hospital information system, performs department matching based on the target auxiliary review information, obtains a target matching department, and then performs a queue to be queued based on the target matching department, combines the target auxiliary review information, adds the queuing application of the target user into the queue to be queued for optimization, and obtains a target queuing queue. The technical effects of improving the dynamic trend analysis accuracy of image recognition and providing reliable auxiliary review information for queuing queue optimization are achieved.

[0085] Embodiment two, based on the same inventive concept as the medical auxiliary review method of one image recognition in the foregoing embodiments, such as Figure 2As shown, the present application provides a medical auxiliary review system for image recognition, and the system and method embodiments in the present application are based on the same inventive concept. The system comprises:

[0086] An image frame sequence derivation module 11 is configured to derive an image frame sequence of a target user within a preset window from a medical imaging device connected to a hospital information system through an interface;

[0087] A target association memory unit obtaining module 12 is configured to perform cross-frame feature association memory identification update on the image frame sequence to obtain a target association memory unit;

[0088] A target auxiliary review information obtaining module 13 is configured to call a target user queuing level identifier to perform information extraction on the target association memory unit, determine a target user queuing level, and take the target user queuing level as target auxiliary review information;

[0089] A target matching department obtaining module 14 is configured to interact with a queuing management module of the hospital information system, perform department matching based on the target auxiliary review information, and obtain a target matching department;

[0090] A target queuing queue obtaining module 15 is configured to perform a queuing queue search based on the target matching department, add a queuing application of the target user into a queuing queue for optimization in combination with the target auxiliary review information, and obtain a target queuing queue.

[0091] Further, the target queuing queue obtaining module 15 is configured to perform the following steps:

[0092] Perform queuing auxiliary review information extraction on the queuing queue to obtain a queuing auxiliary review information queue;

[0093] Optimize the target auxiliary review information and the queuing auxiliary review information queue in order of queuing level from high to low to obtain the target queuing queue.

[0094] Further, the target association memory unit obtaining module 12 is configured to perform the following steps:

[0095] Perform feature extraction on the image frame sequence according to a preset image feature set to obtain an image frame feature vector sequence, wherein the preset image feature set comprises structural texture features, edge information, shape features, and gray density features;

[0096] Perform cross-frame feature association memory update based on the image frame feature vector sequence to obtain a target association memory unit.

[0097] Further, the target association memory unit obtaining module 12 is configured to perform the following steps:

[0098] obtain a sample image frame set, extract structural texture features, edge information, shape features and gray-density features of each sample image frame in the sample image frame set based on the preset image feature set respectively, fill into the initially empty vector to obtain a sample image frame feature vector set;

[0099] obtain a feature extractor, wherein the feature extractor is constructed according to the mapping relationship between the sample image frame set and the sample image frame feature vector set;

[0100] extract features of the image frame sequence by using the feature extractor to obtain the image frame feature vector sequence.

[0101] Further, the target association memory unit obtaining module 12 is used to perform the following steps:

[0102] extract a first image frame feature vector and a second image frame feature vector from the image frame feature vector sequence;

[0103] perform cross-frame feature association analysis on the first image frame feature vector and the second image frame feature vector to obtain a second associated image frame feature vector, and add the first image frame feature vector and the second associated image frame feature vector into a first association memory unit;

[0104] extract a third image frame feature vector from the image frame feature vector sequence again, perform cross-frame feature association memory update on the third image frame feature vector by using the first association memory unit to obtain a second association memory unit;

[0105] perform cross-frame feature association analysis on the remaining image frame feature vectors of the image frame feature vector sequence based on the second association memory unit in turn, and update the second association memory unit according to the analysis result to obtain a target association memory unit.

[0106] Further, the target association memory unit obtaining module 12 is used to perform the following steps:

[0107] calculate the similarity of the corresponding elements of the first image frame feature vector and the second image frame feature vector to obtain an element similarity set;

[0108] perform normalization processing on the element similarity set, and matrix the processed result to obtain an association analysis matrix;

[0109] enhance the second image frame feature vector by using the association analysis matrix to obtain a second associated image frame feature vector.

[0110] Further, the target association memory unit obtaining module 12 is used to perform the following steps:

[0111] The second image frame feature vector is weighted and fused by using the correlation analysis matrix to obtain a second enhanced image frame feature vector;

[0112] The second enhanced image frame feature vector and the second image frame feature vector are fused by using the weighted average method to obtain the second correlation image frame feature vector.

[0113] Further, the target correlation memory unit obtaining module 12 is configured to perform the following steps:

[0114] The third image frame feature vector is cross-frame feature correlation analyzed by using the first image frame feature vector in the first correlation memory unit to obtain a first initial third correlation image frame feature vector;

[0115] The third image frame feature vector is cross-frame feature correlation analyzed by using the second correlation image frame feature vector in the first correlation memory unit to obtain a second initial third correlation image frame feature vector;

[0116] The first initial third correlation image frame feature vector and the second initial third correlation image frame feature vector are processed by using the mean value to obtain a third correlation image frame feature vector, and the third correlation image frame feature vector is added into the first correlation memory unit to obtain a second correlation memory unit.

[0117] Further, the target auxiliary review information is verified, and the network parameter of the target user queuing level identifier is optimized according to the verification result.

[0118] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of the present application are described. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0119] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0120] The present application and the drawings are only exemplary descriptions of the present application, and are considered to cover any and all modifications, changes, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art can make various modifications and changes to the present application without departing from the scope of the present application. Thus, if these modifications and changes of the present application belong to the scope of the present application and its equivalent technology, the present application intends to include these modifications and changes.

Claims

1. A medical auxiliary review method of image recognition, characterized by, The method comprises: The medical imaging device connected with the hospital information system through the interface exports the image frame sequence of the target user within the preset window; Cross-frame feature association memory recognition update is performed on the image frame sequence to obtain a target association memory unit; A target user queuing level identifier is called to perform information extraction on the target association memory unit to determine a target user queuing level, and the target user queuing level is taken as target auxiliary review information. The target user queuing level identifier is a deep neural network with learning ability, which is used to perform discriminant analysis on the semantic features of the target association memory unit to extract the target user queuing level; The queuing management module of the hospital information system is interacted with, and department matching is performed based on the target auxiliary review information to obtain a target matching department; Based on the target matching department, a queuing queue search is performed, and the queuing application of the target user is added into the queuing queue for optimization in combination with the target auxiliary review information to obtain a target queuing queue; Based on the target matching department, a queuing queue search is performed, and the queuing application of the target user is added into the queuing queue for optimization in combination with the target auxiliary review information to obtain a target queuing queue, comprising: Queuing auxiliary review information extraction is performed on the queuing queue to obtain a queuing auxiliary review information queue; The target auxiliary review information and the queuing auxiliary review information queue are optimized in the order from high to low of queuing levels to obtain the target queuing queue; Cross-frame feature association memory recognition update is performed on the image frame sequence to obtain a target association memory unit, comprising: The image frame sequence is respectively subjected to feature extraction according to a preset image feature set to obtain an image frame feature vector sequence, wherein the preset image feature set comprises structural texture features, edge information, shape features and gray density features; Cross-frame feature association memory update is performed based on the image frame feature vector sequence to obtain a target association memory unit; The image frame sequence is respectively subjected to feature extraction according to a preset image feature set to obtain an image frame feature vector sequence, comprising: A sample image frame set is obtained, and structural texture features, edge information, shape features and gray density features of each sample image frame in the sample image frame set are respectively extracted based on the preset image feature set and filled into an initially empty vector to obtain a sample image frame feature vector set; A feature extractor is obtained, wherein the feature extractor is constructed according to the mapping relationship between the sample image frame set and the sample image frame feature vector set; The image frame sequence is subjected to feature extraction by using the feature extractor to obtain the image frame feature vector sequence; Cross-frame feature association memory update is performed based on the image frame feature vector sequence to obtain a target association memory unit, comprising: First and second image frame feature vectors are extracted from the image frame feature vector sequence; perform cross-frame feature correlation analysis on the first image frame feature vector and the second image frame feature vector, and obtain a second correlation image frame feature vector, and add the first image frame feature vector and the second correlation image frame feature vector into a first correlation memory unit; extract a third image frame feature vector from the sequence of image frame feature vectors again, and perform cross-frame feature correlation memory update on the third image frame feature vector by using the first correlation memory unit, and obtain a second correlation memory unit; based on the second correlation memory unit, perform cross-frame feature correlation analysis on the remaining image frame feature vectors of the sequence of image frame feature vectors in turn, and update the second correlation memory unit according to the analysis result, and obtain a target correlation memory unit; perform cross-frame feature correlation analysis on the first image frame feature vector and the second image frame feature vector, and obtain a second correlation image frame feature vector, including: calculate the similarity of the corresponding elements of the first image frame feature vector and the second image frame feature vector, and obtain an element similarity set; perform normalization processing on the element similarity set, and matrix the processed result, and obtain a correlation analysis matrix; use the correlation analysis matrix to enhance the second image frame feature vector, and obtain a second correlation image frame feature vector; use the correlation analysis matrix to perform weighted fusion on the second image frame feature vector, and obtain a second enhanced image frame feature vector; use the weighted average method to fuse the second enhanced image frame feature vector and the second image frame feature vector, and obtain the second correlation image frame feature vector; verify the target auxiliary review information, and optimize the network parameters of the target user queuing level identifier according to the verification result.

2. The medical auxiliary review method of image recognition of claim 1, wherein, extract a third image frame feature vector from the sequence of image frame feature vectors again, and perform cross-frame feature correlation memory update on the third image frame feature vector by using the first correlation memory unit, and obtain a second correlation memory unit, including: perform cross-frame feature correlation analysis on the third image frame feature vector by using the first image frame feature vector in the first correlation memory unit, and obtain a first initial third correlation image frame feature vector; perform cross-frame feature correlation analysis on the third image frame feature vector by using the second correlation image frame feature vector in the first correlation memory unit, and obtain a second initial third correlation image frame feature vector; perform mean value processing on the first initial third correlation image frame feature vector and the second initial third correlation image frame feature vector, and obtain a third correlation image frame feature vector, and add the third correlation image frame feature vector into the first correlation memory unit, and obtain a second correlation memory unit.

3. A medical auxiliary review system of image recognition, characterized in that, The system is used to execute the medical auxiliary review method of image recognition as claimed in any one of claims 1-2, and the system comprises: an image frame sequence derivation module, which is used to derive the image frame sequence of a target user within a preset window by connecting a medical imaging device to a hospital information system through an interface; a target correlation memory unit obtaining module, which is used to perform cross-frame feature correlation memory identification update by traversing the image frame sequence, and obtain a target correlation memory unit; The target auxiliary review information obtaining module is configured to call the target user queuing level identifier to perform information extraction on the target association memory unit, determine a target user queuing level, and take the target user queuing level as target auxiliary review information. The target matching department obtaining module is configured to interact with a queuing management module of a hospital information system, perform department matching based on the target auxiliary review information, and obtain a target matching department. The target queuing queue obtaining module is configured to perform a queuing queue search based on the target matching department, combine the target auxiliary review information, add the queuing application of the target user into a queuing queue for optimization, and obtain a target queuing queue.

Images