Medical terminal image online reading method

Abstract

The invention discloses a medical terminal image online reading method, and belongs to the technical field of data storage, communication connection between a prepositive server and a hospital local area network and communication connection between the prepositive server and a cloud PACS are established respectively, and medical terminal images obtained from the hospital local area network are cached to the prepositive server, so that the image retrieval speed can be increased, and the image retrieval efficiency can be improved. According to the medical terminal image reading method and system, the network traffic characteristics are identified through the pre-deployed server security identification model, the service security operation identification result is determined, and the reading process is controlled according to the service security operation identification result, so that the security of the medical terminal image can be effectively ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of data storage, and particularly relates to a method for online reading of medical terminal images. Background Art

[0002] In the modern medical system, medical image data is an important basis for doctors to make diagnoses and treatments. With the development of digital medical technology, the digital storage and transmission of medical image data have become industry standards. With the rapid development of medical informatization, medical images play an increasingly important role in clinical diagnosis. However, in the prior art, there are the following problems in the transmission and reading of medical terminal image data: the transmission speed is slow, which affects the diagnosis efficiency of doctors; and the security is low, and the risk of patient privacy leakage is relatively high. Summary of the Invention

[0003] The present invention provides a method for online reading of medical terminal images to solve the technical problems existing in the prior art.

[0004] A method for online reading of medical terminal images includes:

[0005] Establish communication connections between a front-end server and a hospital local area network as well as a cloud PACS respectively, and cache the medical terminal images obtained from the hospital local area network in the front-end server to achieve preliminary information synchronization;

[0006] Globally upload each medical terminal image in the front-end server to the cloud PACS for storage, and perform timeliness update on the medical terminal images cached in the front-end server;

[0007] Obtain an image data query request transmitted by a doctor-held device, and query in the front-end server according to the image data query request to obtain a medical terminal image query result and the network traffic characteristics during the query process; wherein, the medical terminal image query result includes querying the target medical terminal image or not querying the target medical terminal image;

[0008] Identify the network traffic characteristics through a pre-deployed server security identification model to determine the service security operation identification result; wherein, the service security operation identification result includes running safely or running unsafely;

[0009] When the service security operation identification result is running unsafely, perform security guarantee processing on the front-end server and end the current online reading process;

[0010] When the service security operation identification result is running safely, pull the target medical terminal image from the front-end server and the cloud PACS according to the medical terminal image query result, and return the target medical terminal image to the doctor-held device for online display.

[0011] In a possible implementation, each piece of medical terminal image in the front-end server is globally uploaded to the cloud PACS for storage, and the medical terminal images cached in the front-end server are updated in a timely manner, including:

[0012] Globally upload each piece of medical terminal image in the front-end server to the cloud PACS for storage;

[0013] For any medical terminal image cached in the front-end server, determine whether the timeliness of the medical terminal image has expired. If so, perform a backup determination on the medical terminal image; otherwise, do not process the medical terminal image;

[0014] Determine whether the medical terminal image has been backed up to the cloud PACS. If so, delete the medical terminal image from the front-end server; otherwise, back up the medical terminal image to the cloud PACS and delete the medical terminal image from the front-end server.

[0015] In a possible implementation, it further includes: after the storage capacity of the front-end server exceeds the preset capacity threshold, if a new medical terminal image is cached from the hospital local area network to the front-end server, delete the medical terminal image with the longest storage time in the front-end server.

[0016] In a possible implementation, obtain an image data query request transmitted by a doctor-held device, and query in the front-end server according to the image data query request to obtain a medical terminal image query result and network traffic characteristics during the query process, including:

[0017] Obtain an image data query request transmitted by a doctor-held device, and at the same time obtain network traffic characteristics during the query process;

[0018] Use a keyword matching method to query the patient information associated with the medical terminal image, and query in the front-end server according to the image data query request to determine the medical terminal image query result;

[0019] Among them, if there is a medical terminal image corresponding to the image data query request in the front-end server, determine that the medical terminal image query result is that the target medical terminal image is found; otherwise, determine that the medical terminal image query result is that the target medical terminal image is not found.

[0020] In a possible implementation, identify the network traffic characteristics through a pre-deployed server security identification model to determine the service security operation identification result, including:

[0021] Construct the network traffic characteristics as the input of a pre-deployed server security recognition model, and obtain the actual output of the pre-deployed server security recognition model to obtain the service security operation recognition result.

[0022] In a possible implementation manner, the method for pre-deploying the server security recognition model includes:

[0023] Construct a server security recognition model using a deep learning model, and initialize the hyperparameters of the server security recognition model to obtain multiple different hyperparameter encodings;

[0024] For any hyperparameter encoding, use the training dataset to obtain the fitness value corresponding to each hyperparameter encoding, and determine the optimal hyperparameter encoding according to the fitness value corresponding to each hyperparameter encoding;

[0025] For any hyperparameter encoding, according to the optimal hyperparameter encoding, use the full information fusion method to determine the search space corresponding to each hyperparameter encoding, and obtain the hyperparameter encoding after determining the search space;

[0026] For the hyperparameter encoding after determining the search space, obtain the current central hyperparameter encoding, and use the sine and cosine functions to determine the spiral search factor. According to the current central hyperparameter encoding and the spiral search factor, use the spiral search method to perform spiral search on the hyperparameter encoding to obtain the hyperparameter encoding after spiral search;

[0027] For the hyperparameter encoding after spiral search, randomly match the first cooperative search encoding and the second cooperative search encoding for the hyperparameter encoding. According to the first cooperative search encoding and the second cooperative search encoding, and use the adaptive cooperative search method based on the encoding position to perform interval search on the hyperparameter encoding to obtain the hyperparameter encoding after interval search;

[0028] For the hyperparameter encoding after interval search, use the Levy reverse flight strategy to perform global search on the hyperparameter encoding to obtain the hyperparameter encoding after global search;

[0029] Repeat the search space, spiral search, interval search, and global search until the training end condition is met. According to the hyperparameter encoding after global search in the last training process, deploy the server security recognition model.

[0030] In a possible implementation manner, for any hyperparameter encoding, according to the optimal hyperparameter encoding, using the full information fusion method to determine the search space corresponding to each hyperparameter encoding, and obtaining the hyperparameter encoding after determining the search space includes:

[0031] For any hyperparameter encoding, the information fusion encoding between the hyperparameter encoding and other hyperparameter encodings is obtained as follows: Among them, represents the j-th hyperparameter encoding in the t-th training process, j = 1, 2,..., P, where P represents the total number of hyperparameter encodings, represents the i-th hyperparameter encoding in the t-th training process, represents the hyperparameter encoding corresponding information fusion encoding;

[0032] The information fusion encoding is perturbed using an exponential function to obtain the perturbed information fusion encoding as follows: Among them, r 1 represents the first random number between [-2, 2], and r 2 represents the second random number between [-1, 1];

[0033] According to the perturbed information fusion encoding and the optimal hyperparameter encoding, the search space corresponding to the hyperparameter encoding is determined, and the hyperparameter encoding after determining the search space is obtained as follows: Among them, represents the optimal hyperparameter encoding in the t-th training process, represents the hyperparameter encoding after determining the search space

[0034] In a possible implementation manner, for the hyperparameter encoding after determining the search space, the current central hyperparameter encoding is obtained, and the spiral search factor is determined using the sine and cosine functions. According to the current central hyperparameter encoding and the spiral search factor, the spiral search method is used to perform a spiral search on the hyperparameter encoding to obtain the hyperparameter encoding after the spiral search, including:

[0035] For the hyperparameter encoding after determining the search space, the current central hyperparameter encoding is obtained as follows: Among them, represents the current central hyperparameter encoding, represents the k-th hyperparameter encoding after determining the search space in the t-th training process, represents the hyperparameter encoding output weight parameter, and f k represents the fitness value of the hyperparameter encoding , and f sum represents the sum of the fitness values of all hyperparameter encodings after determining the search space;

[0036] The spiral search angle is generated as: θ k = απr 3 ; Among them, θ kDenote the spiral search angle corresponding to the hyperparameter encoding after the k-th determined search space in the t-th training process. Let α denote the first spiral search control coefficient, and α is a constant between (5, 10); π represents the pi, and r 3 Denote the third random number between (0, 1);

[0037] Generate the spiral search polar radius as: R k = θ k + βr 4 ; where, R k Denote the spiral search polar radius corresponding to the hyperparameter encoding after the k-th determined search space in the t-th training process. Let β denote the second spiral search control coefficient, and β is a constant between (0.5, 2); r 4 Denote the fourth random number between (0, 1);

[0038] According to the spiral search polar radius and the spiral search angle, generate the first spiral search factor as: where, θ m Denote the spiral search angle corresponding to the hyperparameter encoding after the m-th determined search space in the t-th training process. R m Denote the spiral search polar radius corresponding to the hyperparameter encoding after the k-th determined search space in the t-th training process. γ 1m Denote the first spiral search factor corresponding to the hyperparameter encoding after the m-th determined search space in the t-th training process;

[0039] According to the spiral search polar radius and the spiral search angle, generate the second spiral search factor as: where, γ 2m Denote the second spiral search factor;

[0040] According to the current central hyperparameter encoding, the first spiral search factor, and the second spiral search factor, use the spiral search method to perform spiral search on the hyperparameter encoding, and obtain the hyperparameter encoding after spiral search as: where, Denote the hyperparameter encoding after the (m + 1)-th determined search space, and when m = P, Set it as the hyperparameter encoding after the 1st determined search space; Denote the hyperparameter encoding after spiral search Denote the hyperparameter encoding after the m-th determined search space in the t-th training process.

[0041] In a possible implementation, for the hyperparameter encoding after spiral search, a first collaborative search encoding and a second collaborative search encoding are randomly matched for the hyperparameter encoding. According to the first collaborative search encoding and the second collaborative search encoding, and by using an adaptive collaborative search method based on the encoding position, an interval search is performed on the hyperparameter encoding to obtain the hyperparameter encoding after interval search, including:

[0042] For the hyperparameter encoding after spiral search, different first collaborative search encodings are randomly matched for the hyperparameter encoding and a second collaborative search encoding

[0043] Obtain the fitness value f corresponding to the first collaborative search encoding R1 and the second collaborative search encoding corresponding fitness value f R2 ;

[0044] According to the fitness value f corresponding to the first collaborative search encoding R1 and the second collaborative search encoding corresponding fitness value f R2 , an interval search is performed on the hyperparameter encoding, and the hyperparameter encoding after interval search is:

[0045]

[0046] wherein, represents the d-th dimensional parameter of the first collaborative search encoding , represents the d-th dimensional parameter of the second collaborative search encoding, represents the d-th dimensional parameter of the n-th hyperparameter encoding after spiral search, n = 1, 2,..., P, P represents the total number of hyperparameter encodings, d = 1, 2,..., D, D represents the total dimension of hyperparameters, represents the fitness value corresponding to the n-th hyperparameter encoding after spiral search, represents the d-th dimensional hyperparameter of the n-th hyperparameter encoding after interval search.

[0047] In a possible implementation, for the hyperparameter encoding after interval search, a Levy reverse flight strategy is used to perform a global search on the hyperparameter encoding to obtain the hyperparameter encoding after global search, including:

[0048] For the hyperparameter encoding after interval search, obtain the reverse search encoding corresponding to the h-th hyperparameter encoding as: where h = 1, 2,..., P, P represents the total number of hyperparameter encodings;

[0049] ​​For the hyperparameter encoding after interval search, the Levy search information corresponding to the h-th hyperparameter encoding is obtained as follows: where Levy represents the random step size generated by Levy flight, and σ represents the flight step size control factor, represents the optimal hyperparameter encoding, represents the hyperparameter encoding after the h-th interval search;

[0050] According to the Levy search information and the reverse search encoding, a global search is performed on the hyperparameter encoding, and the hyperparameter encoding after global search is obtained as follows: where, represents the hyperparameter encoding after global search

[0051] Determine whether the fitness value corresponding to the hyperparameter encoding after global search increases. If so, accept this global search; otherwise, reject this global search.

[0052] A method for online reading of medical terminal images provided by the present invention can accelerate the image retrieval speed, improve the doctor's experience and efficiency in using the system by respectively establishing communication connections between the front-end server and the hospital local area network and the cloud PACS, and caching the medical terminal images obtained from the hospital local area network in the front-end server. Then, the network traffic characteristics are identified by a pre-deployed server security identification model to determine the service security operation identification result, and the reading process is controlled according to the service security operation identification result, which can effectively ensure the security of medical terminal images. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0054] Figure 1 It is a flowchart of a method for online reading of medical terminal images provided by an embodiment of the present invention.

[0055] Through the above accompanying drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0057] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0058] As Figure 1 shown, an embodiment of the present invention provides a method for online reading of medical terminal images, including:

[0059] S1. Establish communication connections between the front-end server and the hospital local area network and the cloud PACS (Picture Archiving and Communication System) respectively, and cache the medical terminal images obtained from the hospital local area network in the front-end server to achieve preliminary information synchronization;

[0060] Optionally, when storing the medical terminal images, the medical terminal images can be associated with patient information and timestamps, so as to facilitate information retrieval and maintenance of the front-end server.

[0061] S2. Globally upload each medical terminal image in the front-end server to the cloud PACS for storage, and perform timeliness update on the medical terminal images cached in the front-end server;

[0062] By globally uploading each medical terminal image in the front-end server to the cloud PACS for storage, a large number of medical terminal images can be effectively stored, and by performing timeliness update on the medical terminal images cached in the front-end server, the normal operation of the front-end server can be effectively ensured, and the problem of the front-end server being filled up can be avoided.

[0063] S3. Obtain an image data query request transmitted by the doctor's holding device, and query in the front-end server according to the image data query request to obtain a medical terminal image query result and the network traffic characteristics during the query process; wherein, the medical terminal image query result includes querying the target medical terminal image or not querying the target medical terminal image;

[0064] The image data query request can be patient information and date information, so that the corresponding medical terminal images can be screened according to the patient information and date information, realizing the image data query function. At the same time, the network traffic characteristics during the query process can be queried, and by identifying the network traffic characteristics, it can be determined whether the access process is secure, ensuring data security while realizing fast online reading.

[0065] S4. Identify the network traffic characteristics through a pre-deployed server security identification model to determine the service security operation identification result; wherein, the service security operation identification result includes secure operation or insecure operation.

[0066] When the service security operation identification result is insecure operation, it proves that there is an illegal attack during the access process, which may cause data loss in the front-end server. Therefore, temporary control can be carried out to improve the data security performance.

[0067] S5. When the service security operation identification result is insecure operation, perform security guarantee processing on the front-end server and end the current online reading process.

[0068] The security guarantee processing can be: blocking the corresponding illegal account or prohibiting illegal devices with the same IP from accessing data for a certain period of time. In more serious cases, the access port of the front-end server can be blocked.

[0069] S6. When the service security operation identification result is secure operation, pull the target medical terminal image from the front-end server and the cloud PACS according to the medical terminal image query result, and send the target medical terminal image back to the doctor's holding device for online display.

[0070] When the service security operation identification result is secure operation, if the target medical terminal image is found in the front-end server, the target medical terminal image can be directly transmitted from the front-end server to the doctor's holding device for display; if the target medical terminal image is not found in the front-end server, the target medical terminal image can be pulled from the cloud PACS first and then displayed through the front-end server, thus realizing online reading.

[0071] Optionally, a data encryption algorithm can also be used for encrypted transmission during the data transmission process to avoid data interception and improve data security.

[0072] In a possible implementation, each medical terminal image in the front-end server is globally uploaded to the cloud PACS for storage, and the medical terminal images cached in the front-end server are updated in a timely manner, including:

[0073] Globally upload each medical terminal image in the front-end server to the cloud PACS for storage.

[0074] For any medical terminal image cached in the front-end server, determine whether the timeliness of the medical terminal image has expired. If so, perform a backup judgment on the medical terminal image; otherwise, do not process the medical terminal image.

[0075] Determine whether the medical terminal image has been backed up to the cloud PACS. If so, delete the medical terminal image from the front-end server; otherwise, back up the medical terminal image to the cloud PACS and delete the medical terminal image from the front-end server.

[0076] By performing timeliness processing on the medical terminal images in the front-end server, it is possible to effectively prevent the front-end server from being filled up, and at the same time, the possibility of viewing data that is relatively old is low. Deleting these data has little impact on the need to quickly view images. The front-end server can be regarded as an edge server located near or inside the hospital, which can greatly shorten the communication time and communication cost, while the cloud PACS can be regarded as a large cloud server that can store data in large quantities, thus realizing data storage and traceability.

[0077] In a possible implementation manner, it further includes: after the storage capacity of the front-end server exceeds the preset capacity threshold, if a new medical terminal image is cached from the hospital local area network to the front-end server, delete the medical terminal image with the longest storage time in the front-end server, so as to ensure that medical terminal images can always be stored.

[0078] In a possible implementation manner, obtain an image data query request transmitted by a doctor-held device, and query in the front-end server according to the image data query request to obtain a medical terminal image query result and network traffic characteristics during the query process, including:

[0079] Obtain an image data query request transmitted by a doctor-held device, and at the same time obtain network traffic characteristics during the query process;

[0080] Use a keyword matching method to query the patient information associated with the medical terminal image, and query in the front-end server according to the image data query request to determine the medical terminal image query result;

[0081] Among them, if there is a medical terminal image corresponding to the image data query request in the front-end server, determine that the medical terminal image query result is that the target medical terminal image is found; otherwise, determine that the medical terminal image query result is that the target medical terminal image is not found.

[0082] In a possible implementation manner, identify the network traffic characteristics through a pre-deployed server security identification model to determine the service security operation identification result, including:

[0083] Construct the network traffic characteristics as the input of a pre-deployed server security recognition model, and obtain the actual output of the pre-deployed server security recognition model to obtain the service security operation recognition result.

[0084] In a possible implementation manner, the pre-deployment method of the server security recognition model includes:

[0085] Use a deep learning model to construct a server security recognition model, and initialize the hyperparameters of the server security recognition model to obtain multiple different hyperparameter encodings;

[0086] Optionally, the deep learning model can be set to a convolutional neural network or other neural networks that can classify data, so as to realize the recognition of network traffic characteristics.

[0087] For any hyperparameter encoding, use the training data set to obtain the fitness value corresponding to each hyperparameter encoding, and determine the optimal hyperparameter encoding according to the fitness value corresponding to each hyperparameter encoding;

[0088] The training data set can be set to the KDD Cup 99 data set, the UNB_ISCX data set, the ISCX_IDS data set, etc. After applying the hyperparameter encoding to the server security recognition model, the error function value (such as the cross-entropy error function value) corresponding to each hyperparameter encoding can be obtained through these data sets, and then the error function value is added to a preset constant term (such as 0.1, 0.01, or 0.001, etc.), and the reciprocal is taken to obtain the fitness value corresponding to the hyperparameter encoding.

[0089] For any hyperparameter encoding, according to the optimal hyperparameter encoding, use the full information fusion method to determine the search space corresponding to each hyperparameter encoding to obtain the hyperparameter encoding after determining the search space;

[0090] For the hyperparameter encoding after determining the search space, obtain the current central hyperparameter encoding, and use the sine and cosine functions to determine the spiral search factor. According to the current central hyperparameter encoding and the spiral search factor, use the spiral search method to perform spiral search on the hyperparameter encoding to obtain the hyperparameter encoding after spiral search;

[0091] For the hyperparameter encoding after spiral search, randomly match the first collaborative search encoding and the second collaborative search encoding for the hyperparameter encoding. According to the first collaborative search encoding and the second collaborative search encoding, use the adaptive collaborative search method based on the encoding position to perform interval search on the hyperparameter encoding to obtain the hyperparameter encoding after interval search;

[0092] For the hyperparameter coding after interval search, the Lévy flight reverse strategy is adopted to globally search the hyperparameter coding, and the hyperparameter coding after global search is obtained;

[0093] Repeat the execution of search space, spiral search, interval search, and global search until the training end condition is met (such as the number of training times reaches the preset maximum number of training times). According to the hyperparameter coding after global search in the last training process, deploy the server security recognition model.

[0094] In the prior art, the training of deep learning models mainly adopts the gradient descent intelligent optimization algorithm for training, which may fall into local optimal values, resulting in poor server security recognition effects. Therefore, the embodiments of the present invention provide a new intelligent optimization algorithm to train the server security recognition model, so as to achieve better training effects, realize fast global optimization, and improve the security recognition ability of the server.

[0095] In a possible implementation manner, for any hyperparameter coding, according to the optimal hyperparameter coding, the full information fusion method is used to determine the search space corresponding to each hyperparameter coding, and the hyperparameter coding after determining the search space is obtained, including:

[0096] For any hyperparameter coding, obtain the information fusion coding between the hyperparameter coding and other hyperparameter codings as: Among them, represents the j-th hyperparameter coding in the t-th training process, j = 1, 2,..., P, and P represents the total number of hyperparameter codings, represents the i-th hyperparameter coding in the t-th training process, represents the hyperparameter coding corresponding information fusion coding;

[0097] Use the exponential function to perturb the information fusion coding, and the perturbed information fusion coding is: Among them, r 1 represents the first random number between [-2, 2], and r 2 represents the second random number between [-1, 1];

[0098] According to the perturbed information fusion coding and the optimal hyperparameter coding, determine the search space corresponding to the hyperparameter coding, and the hyperparameter coding after determining the search space is: Among them, represents the optimal hyperparameter coding in the t-th training process, represents the hyperparameter coding after determining the search space

[0099] The present invention uses a full information fusion method to determine the search space corresponding to each hyperparameter encoding. Based on the current optimal position, it can fuse the information of other hyperparameter encodings to determine a search area for each hyperparameter encoding, which can effectively improve the convergence speed of the algorithm. The search accuracy gradually increases in the later stage of the algorithm, which helps the algorithm to tend to be stable.

[0100] In a possible implementation manner, for the hyperparameter encoding after determining the search space, obtain the current central hyperparameter encoding, and use the sine-cosine function to determine the spiral search factor. According to the current central hyperparameter encoding and the spiral search factor, use the spiral search method to perform spiral search on the hyperparameter encoding to obtain the hyperparameter encoding after spiral search, including:

[0101] For the hyperparameter encoding after determining the search space, obtain the current central hyperparameter encoding as: Wherein, represents the current central hyperparameter encoding, represents the kth hyperparameter encoding after determining the search space in the tth training process, represents the hyperparameter encoding of the output weight parameter, and f k represents the fitness value of the hyperparameter encoding ; f sum represents the sum of the fitness values of all hyperparameter encodings after determining the search space;

[0102] Generate the spiral search angle as: θ k = απr 3 ; wherein, θ k represents the spiral search angle corresponding to the kth hyperparameter encoding after determining the search space in the tth training process, α represents the first spiral search control coefficient, and α is a constant between (5, 10); π represents the pi, and r 3 represents the third random number between (0, 1);

[0103] Generate the spiral search polar radius as: R k = θ k + βr 4 ; wherein, R k represents the spiral search polar radius corresponding to the kth hyperparameter encoding after determining the search space in the tth training process, β represents the second spiral search control coefficient, and β is a constant between (0.5, 2); r 4 represents the fourth random number between (0, 1);

[0104] According to the spiral search polar radius and the spiral search angle, generate the first spiral search factor as: Wherein, θ mDenote the spiral search angle corresponding to the hyperparameter encoding after the m-th determined search space in the t-th training process as R m Denote the spiral search polar radius corresponding to the hyperparameter encoding after the k-th determined search space in the t-th training process as γ 1m Denote the first spiral search factor corresponding to the hyperparameter encoding after the m-th determined search space in the t-th training process;

[0105] Generate the second spiral search factor according to the spiral search polar radius and the spiral search angle as: where γ 2m Denote the second spiral search factor;

[0106] Perform spiral search on the hyperparameter encoding by using the spiral search method according to the current central hyperparameter encoding, the first spiral search factor, and the second spiral search factor, and obtain the hyperparameter encoding after spiral search as: where Denote the hyperparameter encoding after the (m + 1)-th determined search space, and when m = P, Set it as the hyperparameter encoding after the 1st determined search space; Denote the hyperparameter encoding after spiral search Denote the hyperparameter encoding after the m-th determined search space in the t-th training process.

[0107] The present invention uses the spiral search method to perform spiral search on the hyperparameter encoding, which can perform spiral search on all hyperparameter encodings, effectively ensure the diversity of the hyperparameter encoding, and at the same time can also increase the possibility of the algorithm searching for the global optimal solution.

[0108] In a possible implementation manner, for the hyperparameter encoding after spiral search, randomly match the first collaborative search encoding and the second collaborative search encoding for the hyperparameter encoding, and perform interval search on the hyperparameter encoding according to the first collaborative search encoding and the second collaborative search encoding by using the adaptive collaborative search method based on the encoding position, and obtain the hyperparameter encoding after interval search, including:

[0109] For the hyperparameter encoding after spiral search, randomly match different first collaborative search encodings and second collaborative search encodings

[0110] Obtain the first collaborative search encoding corresponding fitness value f R1 and the second collaborative search encoding corresponding fitness value f R2 ;

[0111] According to the first collaborative search encoding The corresponding fitness value f R1 And the second collaborative search encoding The corresponding fitness value f R2 , perform interval search on the hyperparameter encoding, and the hyperparameter encoding after interval search is obtained as:

[0112]

[0113] Wherein, Represents the d-th dimension parameter of the first collaborative search encoding Of Represents the d-th dimension parameter of the second collaborative search encoding, Represents the d-th dimension parameter of the hyperparameter encoding after the n-th spiral search, n = 1, 2,..., P, P represents the total number of hyperparameter encodings, d = 1, 2,..., D, D represents the total dimension of hyperparameters, Represents the fitness value corresponding to the hyperparameter encoding after the n-th spiral search, Represents the d-th hyperparameter of the hyperparameter encoding after the n-th interval search.

[0114] The present invention adopts an adaptive collaborative search method based on the encoding position to perform interval search on the hyperparameter encoding, which can effectively search the region between several hyperparameter encodings, can improve the algorithm search speed in the early stage of the algorithm, improve the search accuracy in the later stage of the algorithm, and can effectively avoid collisions during the search process and search more unfamiliar regions.

[0115] In a possible implementation manner, for the hyperparameter encoding after interval search, a Levy reverse flight strategy is adopted to perform global search on the hyperparameter encoding, and the hyperparameter encoding after global search is obtained, including:

[0116] For the hyperparameter encoding after interval search, obtain the reverse search encoding corresponding to the h-th hyperparameter encoding as: Wherein, h = 1, 2,..., P, P represents the total number of hyperparameter encodings;

[0117] For the hyperparameter encoding after interval search, obtain the Levy search information corresponding to the h-th hyperparameter encoding as: Wherein, Levy represents a random step size generated by Levy flight, and σ represents a flight step size control factor, Represents the optimal hyperparameter encoding, Represents the hyperparameter encoding after the h-th interval search;

[0118] According to the Levy search information and the reverse search encoding, perform global search on the hyperparameter encoding, and the hyperparameter encoding after global search is obtained as: Among them, represents the hyperparameter encoding after global search

[0119] Determine whether the fitness value corresponding to the hyperparameter encoding after global search increases. If so, accept this global search; otherwise, reject this global search.

[0120] The embodiment of the present invention adopts the Levy reverse flight strategy to perform global search on the hyperparameter encoding, which can not only effectively assist the algorithm to jump out of the local optimal solution, but also ensure that the search is always towards the optimal region. While effectively ensuring the convergence speed of the algorithm, it can improve the global search ability, thereby achieving the improvement of the server security recognition effect.

[0121] Optionally, after each search, out-of-bounds processing can be performed on the hyperparameter encoding. For example, the hyperparameters of the overrun dimension can be pulled back to the nearest boundary value, or the hyperparameters of the overrun dimension can be randomly initialized within their corresponding upper and lower limits, so as to ensure the effectiveness of the training process.

[0122] Those skilled in the art will readily think of other embodiments of the present invention after considering the specification and the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not disclosed in the present invention. It should be understood that the present invention is not limited to the exact structure described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A method for online reading of medical terminal images, characterized in that: include: Establish communication connections between the front-end server and the hospital LAN and cloud PACS respectively, and cache the medical terminal images obtained from the hospital LAN to the front-end server to achieve preliminary information synchronization; Upload all medical terminal images in the front-end server to the cloud PACS for storage, and update the timeliness of the medical terminal images cached in the front-end server; Obtaining an image data query request transmitted by a device held by a doctor, and querying a front-end server according to the image data query request to obtain a medical terminal image query result and network traffic characteristics during the query process; wherein the medical terminal image query result includes whether the target medical terminal image is queried or not; Identify the network traffic characteristics through a pre-deployed server security identification model to determine a service security operation identification result; wherein the service security operation identification result includes security operation or unsecurity operation; When the service safety operation identification result is that the operation is unsafe, the front-end server performs safety assurance processing and ends the current online reading process; When the service safety operation identification result is safe operation, the target medical terminal image is pulled from the front-end server and the cloud PACS according to the medical terminal image query result, and the target medical terminal image is transmitted back to the doctor's device for online display.

2. The method for online reading of medical terminal images according to claim 1, characterized in that: Upload each medical terminal image in the front-end server to the cloud PACS for storage, and update the medical terminal images cached in the front-end server, including: Upload every medical terminal image in the front-end server to the cloud PACS for storage; For any medical terminal image cached in the front-end server, determine whether the timeliness of the medical terminal image is invalid. If so, perform backup determination on the medical terminal image. Otherwise, do not process the medical terminal image. Determine whether the medical terminal image is backed up in the cloud PACS. If so, delete the medical terminal image from the front-end server. Otherwise, back up the medical terminal image to the cloud PACS and delete the medical terminal image from the front-end server.

3. The method for online reading of medical terminal images according to claim 2, characterized in that: Also includes: When the storage capacity of the front-end server exceeds a preset capacity threshold, if new medical terminal images are cached from the hospital LAN to the front-end server, the medical terminal images stored the longest in the front-end server will be deleted.

4. The method for online reading of medical terminal images according to claim 1, characterized in that: Obtain the image data query request transmitted by the device held by the doctor, and query the front-end server according to the image data query request to obtain the medical terminal image query results and the network traffic characteristics during the query process, including: Obtain the image data query request transmitted by the doctor's device and simultaneously obtain the network traffic characteristics during the query process; A keyword matching method is used to query patient information associated with medical terminal images, and a query is performed on a front-end server according to the image data query request to determine a medical terminal image query result; Among them, if the medical terminal image corresponding to the image data query request exists in the front-end server, the medical terminal image query result is determined to be that the target medical terminal image is queried, otherwise the medical terminal image query result is determined to be that the target medical terminal image is not queried.

5. The method for online reading of medical terminal images according to claim 1, characterized in that: The network traffic characteristics are identified through a pre-deployed server security identification model to determine the service security operation identification result, including: The network traffic characteristics are constructed as the input of a pre-deployed server security identification model, and the actual output of the pre-deployed server security identification model is obtained to obtain a service security operation identification result.

6. The medical terminal image online reading method according to claim 5, characterized in that: The pre-deployment method of the server security identification model includes: A deep learning model is used to build a server security identification model, and the hyperparameters of the server security identification model are initialized to obtain multiple different hyperparameter encodings; For any hyperparameter encoding, a training data set is used to obtain the fitness value corresponding to each hyperparameter encoding, and the optimal hyperparameter encoding is determined according to the fitness value corresponding to each hyperparameter encoding; For any hyperparameter encoding, according to the optimal hyperparameter encoding, the full information fusion method is used to determine the search space corresponding to each hyperparameter encoding, and the hyperparameter encoding after the search space is determined is obtained; For the hyperparameter code after determining the search space, obtain the current center hyperparameter code, and use the sine and cosine functions to determine the spiral search factor, and use the spiral search method to perform a spiral search on the hyperparameter code according to the current center hyperparameter code and the spiral search factor to obtain the hyperparameter code after the spiral search; For the hyperparameter code after the spiral search, the first collaborative search code and the second collaborative search code are randomly matched for the hyperparameter code, and the hyperparameter code is interval searched according to the first collaborative search code and the second collaborative search code, and an adaptive collaborative search method based on the code position is adopted to obtain the hyperparameter code after the interval search; For the hyperparameter encoding after the interval search, the Levy reverse flight strategy is used to perform a global search on the hyperparameter encoding to obtain the hyperparameter encoding after the global search; The search space, spiral search, interval search, and global search are repeatedly executed until the training end conditions are met. The server security identification model is deployed according to the hyperparameter encoding after the global search in the last training process.

7. The medical terminal image online reading method according to claim 6, characterized in that: For any hyperparameter encoding, according to the optimal hyperparameter encoding, the full information fusion method is used to determine the search space corresponding to each hyperparameter encoding, and the hyperparameter encoding after the search space is determined is obtained, including: For any hyperparameter encoding, the information fusion encoding between the hyperparameter encoding and other hyperparameter encodings is obtained as follows: in, represents the jth hyperparameter encoding in the tth training process, j = 1, 2, ..., P, P represents the total number of hyperparameter encodings, represents the i-th hyperparameter encoding during the t-th training process, Represents hyperparameter encoding Corresponding information fusion coding; The information fusion code is disturbed by an exponential function, and the information fusion code after disturbance is obtained as follows: Wherein, r1 represents a first random number between [-2, 2], and r2 represents a second random number between [-1, 1]; According to the information fusion coding and the optimal hyperparameter coding after the disturbance, the search space corresponding to the hyperparameter coding is determined, and the hyperparameter coding after the search space is determined is obtained as follows: in, represents the optimal hyperparameter encoding during the t-th training process, Represents the hyperparameter encoding after determining the search space 8. The medical terminal image online reading method according to claim 7, characterized in that: For the hyperparameter coding after determining the search space, the current center hyperparameter coding is obtained, and the spiral search factor is determined by using the sine and cosine functions. According to the current center hyperparameter coding and the spiral search factor, the hyperparameter coding is spirally searched by using the spiral search method to obtain the hyperparameter coding after the spiral search, including: For the hyperparameter encoding after determining the search space, the current central hyperparameter encoding is obtained as: in, represents the current central hyperparameter encoding, represents the hyperparameter encoding after the kth determined search space during the tth training process, Represents hyperparameter encoding The output weight parameter, and f k Represents hyperparameter encoding The fitness value, f sum Represents the sum of fitness of all hyperparameter encodings after determining the search space; Generate a spiral search angle of: θ k =απr3; where θ k represents the spiral search angle corresponding to the hyperparameter encoding after the kth determined search space in the tth training process, α represents the first spiral search control coefficient, and α is a constant between (5,10); π represents pi, and r3 represents the third random number between (0,1); Generate a spiral search radius of: R k =θ k +βr4; where R k represents the spiral search radius corresponding to the hyperparameter encoding after the kth determined search space in the tth training process, β represents the second spiral search control coefficient, and β is a constant between (0.5, 2); r4 represents the fourth random number between (0, 1); According to the spiral search pole diameter and the spiral search angle, a first spiral search factor is generated as: Among them, θ m represents the spiral search angle corresponding to the hyperparameter encoding after the mth determined search space during the tth training process, R m represents the spiral search radius corresponding to the hyperparameter encoding after the kth determined search space during the tth training process, γ 1m represents the first spiral search factor corresponding to the hyperparameter encoding after the mth determined search space during the tth training process; According to the spiral search pole diameter and the spiral search angle, a second spiral search factor is generated as: Among them, γ 2m represents the second spiral search factor; According to the current central hyperparameter code, the first spiral search factor and the second spiral search factor, a spiral search method is used to perform a spiral search on the hyperparameter code, and the hyperparameter code after the spiral search is obtained as follows: in, represents the hyperparameter encoding after the m+1th determined search space, and when m=P, Set to the hyperparameter encoding after the first determined search space; Represents the hyperparameter encoding after spiral search Represents the hyperparameter encoding after the mth determined search space in the tth training process.

9. The medical terminal image online reading method according to claim 8, characterized in that: For the hyperparameter code after the spiral search, the first collaborative search code and the second collaborative search code are randomly matched for the hyperparameter code, and the hyperparameter code is interval searched according to the first collaborative search code and the second collaborative search code, and an adaptive collaborative search method based on the code position is adopted to obtain the hyperparameter code after the interval search, including: For the hyperparameter encoding after the spiral search, randomly match different first collaborative search encodings for the hyperparameter encoding and the second collaborative search code Get the first collaborative search code The corresponding fitness value f R1 and the second collaborative search code The corresponding fitness value f R2 ; According to the first collaborative search code The corresponding fitness value f R1 and the second collaborative search code The corresponding fitness value f R2 , perform interval search on the hyperparameter encoding, and obtain the hyperparameter encoding after interval search as: in, Indicates the first collaborative search code The d-th dimension parameter of represents the d-th dimension parameter of the second collaborative search encoding, represents the d-th dimension parameter of the hyperparameter encoding after the n-th spiral search, n = 1, 2, ..., P, P represents the total number of hyperparameter encodings, d = 1, 2, ..., D, D represents the total dimension of the hyperparameter, Represents the fitness value corresponding to the hyperparameter encoding after the nth spiral search, represents the d-th dimension hyperparameter of the hyperparameter encoding after the n-th interval search.

10. The medical terminal image online reading method according to claim 9, characterized in that: For the hyperparameter encoding after the interval search, the Levy reverse flight strategy is used to perform a global search on the hyperparameter encoding to obtain the hyperparameter encoding after the global search, including: For the hyperparameter encoding after the interval search, the reverse search encoding corresponding to the hth hyperparameter encoding is obtained as: Where h = 1, 2, ..., P, P represents the total number of hyperparameter encodings; For the hyperparameter encoding after the interval search, the Levy search information corresponding to the hth hyperparameter encoding is obtained as follows: Where Levy represents the random step size generated by Levy flight, σ represents the flight step size control factor, represents the optimal hyperparameter encoding, represents the hyperparameter encoding after the hth interval search; According to the Levy search information and the reverse search code, a global search is performed on the hyperparameter code, and the hyperparameter code after the global search is obtained as follows: in, Represents the hyperparameter encoding after global search Determine whether the fitness value corresponding to the hyperparameter encoding after the global search increases. If so, accept the global search, otherwise reject the global search.

Images