Cross-hospital pathological quality risk collaborative early warning method and application thereof

Abstract

The application provides a cross-hospital pathological quality risk collaborative early warning method and application thereof, and belongs to the technical field of medical data processing. In view of the data privacy barrier and whole-link risk tracing problem in cross-hospital pathological quality control, the scheme adopts a vertical federated learning architecture, each local node extracts quality characteristics and encrypts and uploads gradients, and realizes global model training without domain data export; the whole-link feature correlation of cross-hospital circulation samples is realized by using non-sensitive attribute hash mapping; the risk contribution degree of each link is quantitatively calculated and the root cause is located by combining the Bayesian network and the marginal contribution algorithm, and then the hierarchical collaborative early warning is triggered. The application is mainly used for private compliance quality control, risk real-time early warning and responsibility tracing of the whole link of pathological diagnosis in a medical association scene.

Description

Technical Field

[0001] This invention relates to the field of medical data processing and pathology quality control technology, and in particular to a cross-hospital collaborative early warning method for pathology quality risks and its application. Background Technology

[0002] With the advancement of regional medical consortia, the cross-hospital transfer of pathology samples and mutual recognition of diagnostic results have become core supporting links in the hierarchical medical system. Pathology quality control involves the entire chain of sample collection, fixation, transportation, slide preparation, staining, slide reading, and report issuance. Quality defects in any link will propagate along the process, ultimately leading to diagnostic errors.

[0003] Currently, cross-hospital pathology quality control typically employs either a centralized data platform analysis or a single-hospital independent management model. Centralized data platform analysis attempts to aggregate pathology data from various hospitals onto a unified platform for quality risk identification and tracing; the single-hospital independent management model only identifies and handles quality issues within each hospital's internal processes independently. However, pathology data from each hospital contains a large amount of patient privacy information, and due to data security regulations, centralized aggregation analysis is difficult to implement legally and compliantly. Furthermore, the single-hospital independent management model has blind spots, failing to track the propagation path of risks during cross-hospital transfer. For example, a problem with the proper fixation of samples collected in Hospital A is often only discovered during the slide preparation stage in Hospital B, making it impossible to trace the root cause or provide early warnings of risks associated with the same batch of samples.

[0004] In addition, existing cross-campus quality problem feedback mostly relies on manual reporting, which results in a delayed response. Furthermore, the quality risk identification model trained in a single campus is limited by the sample size and scenario coverage, and its ability to generalize to the identification of rare quality defects is insufficient.

[0005] Therefore, there is an urgent need for a collaborative early warning method for pathology quality risks across different hospital campuses based on federated learning and its application, in order to solve the problems existing in the current technology. Summary of the Invention

[0006] This invention provides a cross-hospital collaborative early warning method for pathology quality risks and its application, addressing the problems of existing pathology quality control technologies, such as centralized analysis being hindered by privacy data barriers, difficulties in cross-domain tracing of risks across the entire chain, weak generalization ability of single-node models, and serious lag in collaborative early warning across multiple hospitals.

[0007] The core technology of this invention is to train a global risk identification model under the premise that the data does not leave the domain through a vertical federated learning framework, and to associate the full-link quality characteristics of cross-hospital samples based on non-sensitive attribute hash mapping, and to locate the root cause and propagation path of risk in order to trigger hierarchical collaborative early warning by combining a risk propagation probability model.

[0008] In a first aspect, the present invention provides a cross-hospital-area collaborative early warning method for pathology quality risks, the method comprising the following steps:

[0009] Local quality control nodes are deployed in each participating hospital area to standardize the extraction and mapping of quality characteristics of the entire pathology chain. The quality characteristics are sequences of physical or digital parameters that are unrelated to the diagnostic results but related to the slide preparation effect, so that the original quality data is stored in the local quality control nodes. A vertical federated learning framework is adopted, which calculates local gradients and performs homomorphic encryption on the gradients at the local quality control node, and aggregates and calculates the encrypted gradients at the coordination node to iteratively train the global pathological quality risk identification model. Based on the hash association identifier generated by the combination of non-sensitive attributes, the full-link quality characteristics of cross-hospital area transfer samples are associated, forming a sample full-link quality characteristic sequence; The contribution of each link in the full-link quality feature sequence of the sample to the final quality risk is calculated by the risk propagation probability model, and the root cause node and risk propagation path are located based on the contribution. Based on the risk level and scope of impact determined by the contribution, a cross-hospital-level collaborative early warning is triggered.

[0010] Furthermore, the entire pathology process includes: sample collection, transportation, slide preparation, staining, and slide reading; mapping processing includes: For continuous numerical features, the Min-Max normalization algorithm is used to map them to the [0,1] interval; For the grade category features, the ordinal value mapping method is used to map them to the [0,1] interval according to the preset grade weights; For qualitative evaluation features, the confidence score output by the underlying AI quality control algorithm is used as the input feature value.

[0011] Furthermore, the gradient is homomorphically encrypted using an additive homomorphic encryption algorithm.

[0012] Furthermore, the hash association identifier is generated as a unique hash value by combining the sample collection time, the code of the department sending the sample, the sample type code, and the sample weight error range.

[0013] Furthermore, based on contribution levels, the root cause nodes of risk are identified, including: A causal topology graph of the entire pathological link is constructed using Bayesian networks, and the posterior probability of each variable in each link under the condition of the final quality risk is deduced by back-calculating Bayes' formula. The marginal contribution algorithm is used to calculate the marginal contribution increment of each feature to the total risk score under different feature combinations.

[0014] Furthermore, the marginal contribution algorithm is the Shapley value algorithm, and the contribution degree is the Shapley value; the link whose Shapley value exceeds the preset contribution degree threshold is marked as the risk root cause node.

[0015] Furthermore, the specific strategy for cross-departmental and hierarchical collaborative early warning is as follows: If the contribution is within the first preset range and the affected sample size is less than the first preset threshold, it is judged as a general risk and an early warning is sent to the person in charge of the corresponding link in the hospital where the risk occurred. If the contribution is within the second preset range and the affected sample size is greater than or equal to the first preset threshold and less than the second preset threshold, it is judged as medium risk, and an early warning is sent to the quality control departments of the hospital where the risk occurred and the receiving hospital involved. If the contribution exceeds the upper limit of the second preset interval or the affected sample size is greater than or equal to the second preset threshold, it is judged as a major risk, an early warning is sent to the medical consortium quality control center, and the subsequent processing of all samples involved is suspended.

[0016] Secondly, the present invention provides a cross-hospital-area pathology quality risk collaborative early warning system, comprising: The local node layer is deployed in each participating hospital area and is used for local quality feature extraction, hash association identifier generation, local gradient calculation, early warning reception and handling; The federal coordination layer, deployed in the regional quality control center, is used for federal model gradient aggregation, global parameter updates, hash identifier matching and verification, risk propagation path calculation, and early warning instruction distribution. The early warning application layer is used for risk visualization, early warning response process tracking, and quality statistical analysis.

[0017] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to execute the above-described cross-hospital pathology quality risk collaborative early warning method.

[0018] Fourthly, the present invention provides a readable storage medium storing a computer program, the computer program including program code for controlling a process to execute the process, the process including the cross-hospital pathology quality risk collaborative early warning method described above.

[0019] The main contributions and innovations of this invention are as follows: 1. Enhanced privacy compliance and data availability: By deploying local nodes in each hospital area to extract quality features unrelated to diagnostic results, and using a vertical federated learning framework combined with homomorphic encryption technology for model training, the original pathological data and plaintext gradients do not leave the domain, completely solving the problem of cross-hospital data barriers. Under the premise of meeting medical data security compliance requirements, the sharing and joint modeling of risk features of multiple hospitals are realized.

[0020] 2. End-to-end Risk Association and Precise Source Tracing: An end-to-end feature association mechanism based on non-sensitive attribute hash mapping was designed, which realizes the end-to-end feature splicing of samples transferred across hospitals without transmitting sensitive patient and sample identifiers; combined with the risk propagation probability model and marginal contribution algorithm, it can accurately quantify the contribution of each link to the final risk, accurately locate the root cause node and propagation path of the risk, and solve the problems of difficulty in tracing quality risks across hospitals and unclear responsibility definition.

[0021] 3. Significantly improved risk management efficiency: A hierarchical collaborative early warning mechanism based on risk contribution and impact range has been established. According to the risk level, the corresponding level of cross-hospital early warning and management process is automatically triggered, reducing the average response time of cross-hospital quality risks from hours to minutes, and significantly reducing the risk spread range and sample retesting rate.

[0022] 4. Enhanced Model Generalization and Rare Defect Recognition Capabilities: By leveraging federated learning to aggregate quality feature gradients from multiple campuses to train a global model, the limitations of single-campus models in terms of sample size and scene coverage are overcome. This significantly improves the accuracy of rare quality defect recognition and effectively adapts to diverse operational procedures and equipment differences across campuses.

[0023] Details of one or more embodiments of the present invention are set forth in the following drawings and description, so that other features, objects and advantages of the invention will be more readily understood. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is an architecture diagram of a cross-hospital pathology quality risk collaborative early warning system according to an embodiment of the present invention; Figure 2 This is a flowchart of the training process for the vertical federated global risk identification model according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the path of transmission of pathological quality risks according to an embodiment of the present invention; Figure 4 This is a flowchart of the cross-hospital-area pathology quality risk classification and collaborative early warning and response process according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the cross-institute sample full-link feature association principle according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0026] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0027] Example 1 This embodiment applies to a regional medical consortium comprising three tertiary hospitals, two of which are general hospitals and one is a pathology diagnostic center. For example... Figure 1 As shown in the figure, this embodiment provides a cross-hospital pathology quality risk collaborative early warning system based on federated learning, which adopts a three-layer architecture: local node layer, federated coordination layer and early warning application layer.

[0028] The local node layer is deployed in each participating hospital area and is responsible for local quality feature extraction, hash association identifier generation, local gradient calculation, early warning reception and handling; the federated coordination layer is deployed in the regional quality control center and is responsible for federated model gradient aggregation, global parameter update, hash identifier matching and verification, risk propagation path calculation and early warning instruction distribution; the early warning application layer is directed to quality control personnel and medical consortium management personnel in each hospital area and is used for risk visualization, early warning handling process tracking and quality statistical analysis.

[0029] Based on the above system, the cross-hospital-area pathology quality risk collaborative early warning method of this embodiment specifically includes the following steps: Step 1: Deploy local quality control nodes in each participating hospital area to standardize and extract and map the quality characteristics of the entire pathology chain, so that the original quality data is stored in the local quality control nodes.

[0030] Quality characteristics refer to standardized sequences of physical or digital parameters extracted from the entire pathology process that are unrelated to diagnostic results but highly correlated with slide preparation quality. This step removes image morphology and identity markers containing patient privacy, retaining only numerical characteristics reflecting process quality. The entire pathology process includes sample collection, transportation, slide preparation, staining, and slide reading. For example: 1) Sample collection process (6 dimensions): collection time, collection site code, fixative ratio, fixation time, sample volume, and qualifications of the collection personnel; 2) Transfer process (4 dimensions): average transfer temperature, transfer time, number of shaking cycles, and type of transfer container; 3) Slicing process (7 dimensions): slice thickness, flatness, dewaxing degree, baking time, baking temperature, slicer model, operator qualifications; 4) Staining process (5 dimensions): staining depth, contrast, background clarity, staining solution batch, operator qualifications; 5) Image reading process (5 dimensions): Image reading physician qualifications, review status, consistency rate between initial diagnosis and review, report duration, and equipment model.

[0031] This embodiment unifies the end-to-end quality features into 27 dimensions. To eliminate the aforementioned differences in physical dimensions and adapt to federated computing, a mapping process is performed on the extracted features. The specific mapping process includes: 1. For continuous numerical features, the Min-Max normalization algorithm is used to map them to the [0,1] interval. Examples include fixed time and sample size in the data acquisition stage, average transport temperature and transport time in the transport stage, and slice thickness and baking time in the slide preparation stage. Taking fixed time as an example, its relative position within the industry standard range is mapped to [0,1] to eliminate differences in different physical dimensions.

[0032] 2. For grade category features, the ordinal value mapping method is used to map them to the [0,1] interval according to the preset grade weights. For example, the qualifications of data collection personnel, operators, and radiologists are assigned equally spaced scores according to their professional titles (such as junior, attending, deputy director, and director) and normalized to reflect the contribution weight of human resources to quality risk control.

[0033] 3. For qualitative evaluation features, the confidence score output by the underlying AI quality control algorithm is used as the input feature value. For example, the flatness and dewaxing degree in the film production stage, and the staining depth, contrast, and background clarity in the staining stage are directly used as the input features of the local control node by the confidence score in the [0,1] interval output by the underlying algorithm.

[0034] Step 2: Using a vertical federated learning framework, the global pathological quality risk identification model is iteratively trained by calculating local gradients and homomorphically encrypting the gradients at the local quality control node, and then aggregating and calculating the encrypted gradients at the coordination node.

[0035] like Figure 2 As shown, the coordinating node sends the initial model parameters to each local node; Each local node uses local data to calculate the model gradient. Gradient calculation refers to the local nodes in each hospital using private pathology quality data to locally train the model and generate a gradient vector that reflects the direction of improvement of model parameters. The gradient is encrypted using an additive homomorphic encryption algorithm and then uploaded to the coordinating node. In this embodiment, the Paillier homomorphic encryption algorithm is specifically used, and the key length is set to 2048 bits, so that the coordinating node can only perform aggregation and accumulation operations on the gradient in the encrypted state, and cannot know the specific gradient value of any single campus. The coordinating node aggregates and calculates the encryption gradient, updates the global model parameters, and then distributes the results to each local node. Iterate the above process until the model converges. The initial learning rate is set to 0.01, the batch size is 64, the number of iterations is set to 100, and the global model's AUC value reaches 0.964 after training.

[0036] Step 3: Based on the hash association identifier generated by the combination of non-sensitive attributes, associate the full-link quality characteristics of the cross-hospital area transfer samples to form a full-link quality characteristic sequence of the samples.

[0037] like Figure 5 As shown, for samples transferred across hospital campuses, the outgoing hospital combines the non-sensitive attributes of the samples. In this embodiment, the combined attributes are: collection time accurate to the minute, 3-digit code of the sending department, 2-digit code of the sample type, and sample weight error range of ±0.5g. After combination, a unique hash value is generated by a hash algorithm. In this embodiment, the SHA-256 algorithm is used to generate a 32-bit hash value as a temporary association identifier, which is synchronized to the receiving hospital along with the sample transfer slip. The receiving hospital uses the same algorithm to generate a hash value for the corresponding attributes of the received sample and matches it with the hash value in the transfer slip.

[0038] If a match is found, the features of the sample in the outgoing and receiving hospital areas are logically concatenated to achieve feature association across hospital processes. All association processes only transmit hash values ​​and do not transmit any sensitive identifiers of patients or samples.

[0039] Step 4: Calculate the contribution of each link in the sample's full-link quality feature sequence to the final quality risk using a risk propagation probability model, and locate the root cause node and risk propagation path based on the contribution.

[0040] like Figure 3 As shown, for samples identified as having quality risks, a causal topology graph of the entire pathological pathway is constructed using a Bayesian network, with the final quality risk as the target node Y and the characteristics of each process as condition nodes. When the terminal identifies a quality defect (Y=1), the posterior probability of each variable in each stage is calculated backwards using Bayes' theorem, under the condition of the final quality risk. The process involves identifying suspicious links from a macro-causal chain; then, using a marginal contribution algorithm, the marginal contribution increment of each link's features to the total risk score under different feature combinations is calculated.

[0041] In this embodiment, the marginal contribution algorithm specifically adopts the Shapley Value algorithm, where the Shapley value of link i in the entire link is... The calculation formula is as follows:

[0042] Where N represents the set of all extracted pathological quality features used in the evaluation of the entire chain in this scheme (i.e., the "general coalition" in game theory), which includes the feature parameters of all links; |N| represents the total number of features contained in set N; i represents the target feature link whose risk contribution is currently being calculated; S represents any feature subset in set N that does not contain target link i (i.e., the "local coalition" in game theory); |S| represents the number of features contained in the current feature subset S; The eigenvalue function (i.e., the payoff function) represents, in this scheme, the quality risk assessment score predicted by the model when only the quality features of a subset S are input into the global pathological quality risk identification model. This represents the marginal risk contribution increment of link i, that is, the specific value that causes the total quality risk score to increase or decrease after the addition of the actual features of the target link i, which originally only had the effect of feature subset S. The probability weighting coefficient representing the occurrence of a specific feature subset S is mathematically based on the following: Given all features in random permutations, |N|! is the total number of permutations of all stage features, |S|! is the number of permutations of features within subset S, and (|N|-|S|-1)! is the number of permutations of features remaining excluding subset S and stage i. This coefficient ensures that the contribution of each stage under various combinations of preceding features is fairly and evenly weighted.

[0043] Calculate the contribution of each step, i.e., the Shapley value. Then, the links whose Shapley values ​​exceed the preset contribution threshold are marked as risk root cause nodes. In this embodiment, the preset contribution threshold is set to 0.3. Finally, the shortest directed edge from the root cause node to the final risk node is constructed according to the process flow order to form a risk propagation path and clarify the risk transmission probability of each node.

[0044] Step 5: Based on the risk level and scope of impact determined by the contribution, trigger cross-hospital-level collaborative early warning.

[0045] like Figure 4 As shown, the specific strategy for cross-departmental and hierarchical collaborative early warning is as follows: If the contribution is within the first preset range and the number of affected samples is less than the first preset threshold, it is judged as a general risk. Only the person in charge of the corresponding link in the hospital where the risk occurred will be notified and asked to provide feedback on the handling results within 2 hours. In this embodiment, the first preset range is [0.3, 0.5), and the first preset threshold is 10 samples. If the contribution is within the second preset range and the number of affected samples is greater than or equal to the first preset threshold and less than the second preset threshold, it is judged as medium risk. An early warning is simultaneously pushed to the quality control departments of the hospital where the risk occurred and the receiving hospital involved, along with a risk propagation path diagram. The joint investigation across hospitals is required to be completed within 4 hours. In this embodiment, the second preset range is [0.5, 0.7), and the second preset threshold is 50 samples. If the contribution exceeds the upper limit of the second preset interval (greater than or equal to 0.7), or if the affected sample size is greater than or equal to the second preset threshold (50 samples), it is judged as a major risk. The system will push the highest level warning to the regional medical consortium quality control center, initiate a full-chain risk investigation, and suspend the subsequent processing of all samples of the same batch or with the same characteristics to block errors.

[0046] After the system in this embodiment was online for 6 months, the re-examination rate of cross-hospital pathology samples decreased from 9.2% to 1.1%, the average risk response time was shortened from 22 hours to 12 minutes, the root cause localization accuracy reached 95.1%, and no data leakage incidents occurred.

[0047] Example 2 This embodiment provides a preferred implementation method for cross-hospital collaborative identification of the rare quality defect "incomplete dewaxing" in paraffin sections. This defect occurs in fewer than 30 cases per hospital per year, and the accuracy rate of conventional single-hospital models is only 52%.

[0048] In this embodiment, the five participating hospitals accumulated a total of 217 incompletely dewaxed samples and 12,000 normal samples. The data from each hospital was stored independently and not aggregated.

[0049] The local quality control node extracts the qualitative evaluation feature of the degree of dewaxing during the film production process, and uses the confidence score output by the underlying AI quality control algorithm as the input feature value.

[0050] A vertical federated learning framework was used to train a global pathological quality risk identification model, specifically a vertical federated logistic regression model. Positive samples were weighted at 10 to address the extreme imbalance of sample classes, and the number of iterations was set to 80 rounds. Since the logistic regression model involves polynomial approximation operations, the CKKS homomorphic encryption algorithm was used instead of the Paillier algorithm to improve encryption efficiency. The precision was set to 16 bits, supporting approximate homomorphic calculations on floating-point numbers.

[0051] After training, the global model achieved an accuracy of 79.2% and a recall of 76.8% in identifying incomplete dewaxing defects, which are 27.2% and 31.5% higher than the average level of a single hospital, respectively. It can effectively identify this rare quality defect and avoid diagnostic errors caused by incomplete dewaxing.

[0052] Example 3 This embodiment also provides an electronic device, see reference. Figure 6 It includes a memory 404 and a processor 402, wherein the memory 404 stores a computer program and the processor 402 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0053] Specifically, the processor 402 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement embodiments of the present invention.

[0054] Memory 404 may include a mass storage device for data or instructions. For example, and not limitingly, memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 404 may include removable or non-removable (or fixed) media. Where appropriate, memory 404 may be internal or external to a data processing device. In a particular embodiment, memory 404 is non-volatile memory. In a particular embodiment, memory 404 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0055] The memory 404 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 402.

[0056] The processor 402 reads and executes computer program instructions stored in the memory 404 to implement any of the cross-hospital collaborative early warning methods for pathology quality risks in the above embodiments.

[0057] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408, wherein the transmission device 406 is connected to the processor 402, and the input / output device 408 is connected to the processor 402.

[0058] The transmission device 406 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 406 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0059] Input / output device 408 is used to input or output information.

[0060] Example 4 This embodiment also provides a readable storage medium storing a computer program, the computer program including program code for controlling a process to execute the process, the process including the cross-hospital pathology quality risk collaborative early warning method according to Embodiment 1.

[0061] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0062] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0063] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figures 2-5 Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0064] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] The above embodiments are merely illustrative of several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. A cross-hospital collaborative early warning method for pathology quality risks, characterized in that, Includes the following steps: Local quality control nodes are deployed in each participating hospital area to standardize the extraction and mapping of quality characteristics of the entire pathology chain. The quality characteristics are sequences of physical or digital parameters that are unrelated to the diagnostic results but related to the slide preparation effect, so that the original quality data is stored in the local quality control nodes. A vertical federated learning framework is adopted, in which local gradient calculation is performed at the local quality control node and the gradient is homomorphically encrypted. The encrypted gradient is aggregated and calculated at the coordination node, and the global pathological quality risk identification model is iteratively trained. Based on the hash association identifier generated by the combination of non-sensitive attributes, the full-link quality characteristics of cross-hospital area transfer samples are associated, forming a sample full-link quality characteristic sequence; The contribution of each link in the full-link quality feature sequence of the sample to the final quality risk is calculated by the risk propagation probability model, and the root cause node and risk propagation path are located based on the contribution. Based on the risk level and scope of impact determined by the contribution, a cross-hospital-level collaborative early warning is triggered.

2. The cross-hospital-area pathology quality risk collaborative early warning method as described in claim 1, characterized in that, The entire pathology process includes: sample collection, transportation, slide preparation, staining, and slide reading; the mapping process includes: For continuous numerical features, the Min-Max normalization algorithm is used to map them to the [0,1] interval; For the grade category features, the ordinal value mapping method is used to map them to the [0,1] interval according to the preset grade weights; For qualitative evaluation features, the confidence score output by the underlying AI quality control algorithm is used as the input feature value.

3. The cross-hospital-area pathology quality risk collaborative early warning method as described in claim 1, characterized in that, The gradient is homomorphically encrypted using an additive homomorphic encryption algorithm.

4. The cross-hospital-area pathology quality risk collaborative early warning method as described in claim 1, characterized in that, The hash association identifier is generated by combining the sample collection time, the code of the department sending the sample, the sample type code, and the sample weight error range, and then using a hash algorithm to generate a unique hash value.

5. The cross-hospital-area pathology quality risk collaborative early warning method as described in claim 1, characterized in that, Based on the contribution level, the root cause nodes of the risk are located, including: A causal topology graph of the entire pathological link is constructed using Bayesian networks, and the posterior probability of each variable in each link under the condition of the final quality risk is deduced by back-calculating Bayes' formula. The marginal contribution algorithm is used to calculate the marginal contribution increment of each feature to the total risk score under different feature combinations.

6. The cross-hospital-area pathology quality risk collaborative early warning method as described in claim 5, characterized in that, The marginal contribution algorithm is the Shapley value algorithm, and the contribution degree is the Shapley value; the link whose Shapley value exceeds the preset contribution degree threshold is marked as the risk root cause node.

7. The cross-hospital-area pathology quality risk collaborative early warning method as described in any one of claims 1 to 6, characterized in that, The specific strategy for cross-hospital-level collaborative early warning is as follows: If the contribution is within the first preset range and the affected sample size is less than the first preset threshold, it is determined to be a general risk, and an early warning is sent to the person in charge of the corresponding link in the hospital where the risk occurred. If the contribution is within the second preset range and the affected sample size is greater than or equal to the first preset threshold and less than the second preset threshold, it is determined to be of medium risk, and an early warning is sent to the quality control departments of the hospital where the risk occurred and the receiving hospital involved. If the contribution exceeds the upper limit of the second preset interval or the affected sample size is greater than or equal to the second preset threshold, it is determined to be a major risk, an early warning is sent to the medical consortium quality control center, and the subsequent processing of all samples involved is suspended.

8. A cross-hospital-area pathology quality risk collaborative early warning system, characterized in that, include: The local node layer is deployed in each participating hospital area and is used for local quality feature extraction, hash association identifier generation, local gradient calculation, early warning reception and handling; The federal coordination layer, deployed in the regional quality control center, is used for federal model gradient aggregation, global parameter updates, hash identifier matching and verification, risk propagation path calculation, and early warning instruction distribution. The early warning application layer is used for risk visualization, early warning response process tracking, and quality statistical analysis.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to execute the cross-hospital pathology quality risk collaborative early warning method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, the computer program including program code for controlling a process to execute the process, the process including the cross-hospital-area pathology quality risk collaborative early warning method according to any one of claims 1 to 7.

Images