Biological sample management-based compliance decision-making method and system

Abstract

The invention provides a compliance decision-making method and system based on biological sample management, and relates to the technical field of biological sample management, and the method comprises the steps: obtaining a multi-modal original file; the multi-modal original file is related to standardized management of a biological sample library; performing analysis processing on the multi-modal original file through a multi-modal file analysis engine to obtain modal fusion information; calling the modal fusion information, and constructing a knowledge graph for compliance management of the biological sample; obtaining an operation request of a user; the operation request comprises an original operation instruction; performing compliance verification on the operation request in combination with the knowledge graph; and performing compliance decision on the operation request according to the compliance verification result, thereby improving the accuracy and efficiency of compliance judgment, and effectively reducing the sample quality risk caused by illegal operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological sample management, and particularly relates to a compliance decision-making method and system based on biological sample management, an electronic device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] A biological sample library is a key resource platform for storing various biological samples and their associated clinical data, and its standardized management is directly related to the repeatability of research data and the success rate of clinical transformation.

[0003] In actual situations, the management organizational structure of some institutions is missing, which leads to the fact that in the process of biological sample management, process execution deviation may not be effectively and timely found, and thus sample quality risks are difficult to avoid.

[0004] Therefore, a compliance decision-making method and system based on biological sample management, an electronic device, a computer readable storage medium and a computer program product are provided. SUMMARY

[0005] In order to ensure the operation quality of the biological sample library, the present specification provides a compliance decision-making method and system based on biological sample management, an electronic device, a computer readable storage medium and a computer program product, The knowledge graph of biological sample compliance management is used to perform compliance verification on the operation request, so as to ensure the operation specification of the user and the quality of the biological sample, reduce biological sample management defects, and realize standardized management of the biological sample.

[0006] The compliance decision-making method based on biological sample management provided in the present application adopts the following technical solution, comprising: Obtaining a multi-modal original file; the multi-modal original file is related to the standardized management of the biological sample library; Performing analysis and processing on the multi-modal original file through a multi-modal file analysis engine to obtain modal fusion information; Retrieving the modal fusion information to construct a knowledge graph of biological sample compliance management; Obtaining an operation request of a user; the operation request comprises an original operation instruction; Performing compliance verification on the operation request in combination with the knowledge graph; wherein, a standard operation instruction related to the original operation instruction is found from the knowledge graph; a difference index of the original operation instruction and the standard operation instruction is calculated; and based on the size relationship between the difference index and a preset difference threshold, compliance verification is performed; Performing compliance decision-making on the operation request according to the compliance verification result.

[0007] Optionally, the multi-modal file parsing engine is configured to parse the multi-modal raw files to obtain modal fusion information, including: identifying the type of each raw file; matching a corresponding parsing strategy according to the type; parsing the raw files using the parsing strategy to obtain a parsing result of each raw file; splicing all the parsing results to generate the modal fusion information.

[0008] Optionally, the parsing result includes at least one of text information and final path encoding information. Optionally, the parsing of the raw files using the parsing strategy to obtain a parsing result of each raw file includes: parsing a first raw file into text information according to a first parsing strategy; wherein the first raw file is repaired to generate an optimized first raw file; the optimized first raw file is parsed in a first hierarchical manner to obtain text information related to the first raw file; and / or, parsing a second raw file into text information according to a second parsing strategy; wherein the second raw file is parsed in a second hierarchical manner to obtain text information related to the second raw file; and / or, parsing a third raw file into text information according to a third parsing strategy; wherein the third raw file is converted into a text sequence marked with a timestamp; the text sequence marked with the timestamp is normalized in semantics to obtain text information related to the third raw file; and / or, parsing a fourth raw file according to a fourth parsing strategy to obtain final path encoding information; wherein structured topological features are extracted from the fourth raw file; the topological features are used to encode a decision path to obtain final path encoding information.

[0009] Optionally, the modal fusion information is retrieved to construct a knowledge graph for the compliance management of biological samples, including: extracting entities of each preset level from the modal fusion information; forming a four-dimensional association network by combining the association relationships between the preset levels and the entities of the preset levels; detecting conflicts in the four-dimensional association network, and optimizing the four-dimensional association network based on the detection result to obtain the knowledge graph.

[0010] Optionally, the compliance decision for the operation request according to the compliance verification result includes: When the compliance check result is a check pass, an operation success rate of the original operation instruction is evaluated, and whether to execute the operation request is determined according to the operation success rate; When the compliance check result is a check fail, a compliance suggestion is generated for the original operation instruction.

[0011] Optionally, when the compliance check result is a check pass, the operation success rate of the original operation instruction is evaluated, and whether to execute the operation request is determined according to the operation success rate, including: An operation instruction package is constructed according to the original operation instruction and the knowledge graph; The operation instruction package is input into a rehearsal model to predict the operation success rate; When the operation success rate is greater than or equal to a preset success threshold, the original operation instruction is issued to an execution terminal.

[0012] Optionally, further comprising: Process information of a current compliance decision is obtained; A corresponding biological sample ID is obtained according to the original operation instruction; The biological sample ID is associated with the process information to generate three-dimensional traceability information.

[0013] The compliance decision system based on biological sample management provided in the application adopts the following technical solution, comprising: A file input module is configured to obtain a multi-modal original file; the multi-modal original file is related to standardized management of a biological sample library; A multi-modal file analysis engine is configured to analyze and process the multi-modal original file to obtain modal fusion information; A knowledge graph construction module is configured to retrieve the modal fusion information and construct a knowledge graph of biological sample compliance management; A request acquisition module is configured to acquire an operation request of a user; the operation request comprises an original operation instruction; A compliance control module is configured to perform compliance checking on the operation request in combination with the knowledge graph; A compliance decision module is configured to perform compliance decision on the operation request according to a compliance check result.

[0014] Optionally, the compliance control module comprises: An instruction searching sub-module is configured to search for a standard operation instruction related to the original operation instruction from the knowledge graph; A difference index calculation sub-module is configured to calculate a difference index of the original operation instruction and the standard operation instruction; The compliance verification submodule is configured to perform compliance verification based on a size relationship between the difference index and a preset difference threshold. Optionally, the multi-modal file analysis engine comprises: The type identification submodule is configured to identify the type of each original file. The strategy matching submodule is configured to match a corresponding analysis strategy according to the type. The analysis submodule is configured to analyze the original files by using the analysis strategy to obtain an analysis result of each original file. The unified semantic representation submodule is configured to splice all the analysis results to generate the modal fusion information.

[0015] Optionally, the analysis result comprises at least one of text information and final path encoding information. Optionally, the analysis submodule comprises: The first analysis unit is configured to analyze a first original file into text information according to a first analysis strategy. The second analysis unit is configured to analyze a second original file into text information according to a second analysis strategy. The third analysis unit is configured to analyze a third original file into text information according to a third analysis strategy. The fourth analysis unit is configured to analyze a fourth original file into final path encoding information according to a fourth analysis strategy. Optionally, the first analysis unit comprises: The document repair subunit is configured to perform document repair on the first original file to generate an optimized first original file. The first hierarchical analysis subunit is configured to perform first hierarchical analysis on the optimized first original file to obtain text information related to the first original file. Optionally, the second analysis unit comprises: The second hierarchical analysis subunit is configured to perform second hierarchical analysis on the second original file to obtain text information related to the second original file. Optionally, the third analysis unit comprises: The speech recognition subunit based on an end-to-end model is configured to perform text conversion on the third original file to obtain a timestamp-labeled text sequence. The semantic normalization subunit is configured to perform semantic normalization on the timestamp-labeled text sequence to obtain text information related to the third original file. Optionally, the fourth analysis unit comprises: Optionally, a feature extraction subunit based on a graph convolutional network is used to extract structured topological features from the fourth original file; A decision path encoder is used to encode the decision path based on the topological features to obtain the final path encoding information.

[0016] Optionally, the knowledge graph construction module includes: The entity extraction submodule is used to extract entities at various preset levels from the modal fusion information; The association network generation submodule is used to combine the association relationships between the preset levels and the entities of the preset levels to form a four-dimensional association network; The conflict detection submodule is used to perform conflict detection on the four-dimensional association network, optimize the four-dimensional association network based on the detection results, and obtain the knowledge graph.

[0017] Optionally, the compliance decision module includes: The first decision submodule is used to evaluate the success rate of the original operation instruction when the compliance verification result is that the verification is passed, and to determine whether to execute the operation request based on the success rate. The second decision submodule is used to generate compliance suggestions for the original operation instructions when the compliance verification result is that the verification fails.

[0018] Optionally, the second decision submodule includes: An operation instruction package construction unit is used to construct an operation instruction package based on the original operation instructions and the knowledge graph. An operation success rate prediction unit is used to input the operation instruction package into the pre-simulation model to predict the operation success rate. The decision unit is used to send the original operation instruction to the execution terminal when the success rate of the operation is greater than or equal to a preset success threshold.

[0019] Optionally, it also includes: a traceability module; optionally, the traceability module includes: The information acquisition submodule is used to acquire process information for the current compliance decision-making process; The ID acquisition submodule is used to acquire the corresponding biological sample ID according to the original operation instruction; The traceability association submodule is used to associate the biological sample ID with the process information to generate three-dimensional traceability information.

[0020] This specification also provides a computer device, wherein the computer device includes: Processor; and, A memory that stores computer-executable instructions, which, when executed, cause the processor to perform any of the methods described above.

[0021] This specification also provides a computer-readable storage medium that stores one or more programs / instructions that, when executed by a processor, implement any of the methods described above.

[0022] This specification also provides a computer program product, wherein the computer program product includes: a computer program / instruction, which, when executed by a processor, implements any of the methods described above.

[0023] In this invention, multimodal raw files are acquired; these multimodal raw files are related to the standardized management of biobanks; the multimodal raw files are parsed and processed by a multimodal file parsing engine to obtain modality fusion information; the modality fusion information is retrieved to construct a knowledge graph for biobank compliance management; user operation requests are obtained; these operation requests include original operation instructions; the operation requests are validated for compliance using the knowledge graph; and compliance decisions are made based on the compliance validation results. This improves the accuracy and efficiency of compliance judgment and effectively reduces sample quality risks caused by violations. Attached Figure Description

[0024] Figure 1 A schematic diagram illustrating the principle of a compliance decision-making method based on biological sample management, provided for embodiments of this specification; Figure 2 A flowchart illustrating a multimodal file parsing engine for a compliance decision-making method based on biological sample management, provided in the embodiments of this specification; Figure 3 A schematic diagram illustrating step S3 of a compliance decision-making method based on biological sample management, provided in an embodiment of this specification. Figure 4 The flowcharts for S4-S6 of a compliance decision-making method based on biological sample management provided in the embodiments of this specification are as follows: Figure 5 Example schematic diagrams of S4-S7 of a compliance decision-making method based on biological sample management provided in the embodiments of this specification; Figure 6 A flowchart illustrating a compliance decision-making method based on biological sample management, provided as an embodiment of this specification; Figure 7 A schematic diagram of the structure of a compliance decision-making system based on biological sample management, provided for embodiments of this specification; Figure 8This is a schematic diagram of the structure of an electronic device provided in an embodiment of this specification; Figure 9 This is a schematic diagram of a computer-readable storage medium provided for an embodiment of this specification. Detailed Implementation

[0025] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0026] Exemplary embodiments of the invention will now be described more fully with reference to the accompanying drawings. While conforming to the inventive concept, the features, structures, characteristics, or other details described in a particular embodiment may be combined in one or more other embodiments in a suitable manner.

[0027] The terms “and / or” or “and / or” include all combinations of any one or more of the listed items.

[0028] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0029] Figure 1 A schematic diagram illustrating the principle of a compliance decision-making method based on biological sample management, provided for embodiments of this specification, includes: S1 acquires the original files of the multimodal data; the original files of the multimodal data are related to the standardized management of the biobank. S2 uses a multimodal file parsing engine to parse and process the original multimodal file to obtain modality fusion information; S3 retrieves the modality fusion information to construct a knowledge graph for compliance management of biological samples; S4 obtains the user's operation request; the operation request includes: original operation instructions; S5 performs compliance verification on the operation request using the knowledge graph; S6 makes a compliance decision on the operation request based on the compliance verification results.

[0030] Biobanks, as key resource platforms for the centralized storage of various biological samples and their associated clinical data, play a crucial supporting role in the development of modern precision medicine. With breakthroughs in cutting-edge technologies such as genomics, single-cell sequencing, and drug target validation, biobanks are not only the material basis for disease mechanism research but also a vital infrastructure for translational medicine research. Their standardized management directly impacts the reproducibility of research data and the success rate of clinical translation.

[0031] In the current process of biological sample management, some institutions have not established independent sample management departments due to resource constraints (human resources, funds, etc.), resulting in the dispersion of management responsibilities throughout the entire sample life cycle. Under this "multi-headed management" model, there is a lack of unified standard setting, process supervision and quality control, leading to a disconnect in the operational process.

[0032] Furthermore, due to the complexity of biosample management—characterized by multiple dimensions, stages, and variables—it demands extremely high levels of professionalism and responsibility from operators. However, in the absence of dedicated oversight, operators may fail to manage biosamples according to prescribed standards due to reliance on experience, unfamiliarity with procedures, or carelessness. Such non-standard operating practices can easily lead to problems such as reduced reliability of biosamples due to improper management.

[0033] Therefore, in order to improve the standardization of biological sample management, this invention provides a compliance decision-making method based on biological sample management, specifically: S1 retrieves the original file for the multimodal data; The original documents of the multimodal approach are related to the standardized management of biobanks; specifically, the original documents include, but are not limited to, clauses related to the standardized management of biobanks, which record compliant and non-compliant operations related to the standardized management of biobanks.

[0034] The term "multimodal" as used in this invention refers to two or more different types of original files.

[0035] As a preferred method, the original files can be classified according to their original file format.

[0036] In one embodiment of the present invention, the type of the original file includes, but is not limited to: a first type, a second type, a third type, and a fourth type.

[0037] The original file formats include, but are not limited to: PDF format, audio stream-related formats, and flowchart-related formats. PDF format includes both image-based and text-based PDF formats.

[0038] Specifically, the original PDF files mainly record the terms related to the sample library standardization process.

[0039] The first type is used to characterize the original file as an image-based PDF. Image-based PDFs are digital images of paper documents, storing image data (pixel matrices). Text is represented by pixels within the image rather than editable text encoding; therefore, they are not editable. Original files of the first type include PDF files obtained through scanning. That is, paper documents are converted into images using a scanner, mobile phone camera, or document scanner, and then multiple pages of images are packaged into a PDF file. In one embodiment of the invention, the original file of the first type is a scanned copy with handwritten annotations.

[0040] The second type is used to indicate that the original file is a text-based PDF. Text-based PDFs are electronic document format conversions that simultaneously store text encoding and layout information, and are editable. Original files of this second type include PDF files converted from Word documents. That is, text, layout, images, tables, and other elements in a Word document are directly converted to PDF using the "Save As PDF" function in office software such as Word.

[0041] The third type is used to characterize the original file as an audio stream-related format. Original files in audio stream-related formats can be video / audio recordings related to the standardized management of biobanks, such as training recordings, compliance reminders, and on-site operational guidance dialogues, used to supplement the limitations of static descriptions in written terms.

[0042] The fourth type is used to characterize original documents in flowchart-related formats. Flowchart-related original documents are graphical records related to standardized processes in the sample library, such as standardized operation flowcharts. They visually represent key standardized nodes and operational specifications.

[0043] This invention provides a data foundation for subsequent analysis by comprehensively collecting multimodal raw files related to the standardized management of biobanks.

[0044] S2 uses a multimodal file parsing engine to parse and process the original multimodal file to obtain modality fusion information; S21 builds a multimodal file parsing engine; The multimodal file parsing engine is used to perform semantic parsing on raw files of various formats.

[0045] In one embodiment of the present invention, the multimodal file parsing engine includes: A feature extraction subunit based on graph convolutional networks is used to extract structured topological features from the third type of original files; The decision path encoder is used to acquire the output of a graph convolutional network and encode its decision paths.

[0046] The document repair subunit is used to repair the original files of the first type. A hierarchical parser is used to acquire and parse the output of the document repair subunit; and / or, to acquire and parse the second type of raw file; The speech recognition subunit based on the end-to-end model is used to acquire the fourth type of raw file and perform text conversion; The semantic normalization subunit is used to obtain the output of the speech recognition subunit based on the end-to-end model and perform semantic normalization.

[0047] The unified semantic representation submodule is used to obtain the outputs of the decision path encoder, the hierarchical parser, and the semantic normalization subunit, and then concatenate them.

[0048] The terms entity library is used to acquire and store the output of the unified semantic representation submodule.

[0049] By building a specialized multimodal file parsing engine, parsing efficiency and accuracy are improved, and parsing of multiple file formats is supported.

[0050] S22 identifies the type of each original file; S221 determines the format of the original file based on its file extension; S222 performs type identification based on the format of the original file; S222-1 When the original file is in PDF format, the first or second type is identified by combining the file information; Since both image-based and text-based PDFs use the .PDF extension, it's impossible to determine the first or second type solely based on the extension. Further analysis of the file information is required.

[0051] Considering that the original content of an image-based PDF file is solely images and lacks a text layer, while a text-based PDF file embeds text fonts, in one embodiment of this invention, file type classification can be based on font information. Specifically, the file information is the metadata of the original file. When the original file is in PDF format, its metadata is read; if the original file's metadata does not contain any font information, the original file is identified as an image-based PDF and marked as a first type. If the original file's metadata includes font names, the original file is identified as a text-based PDF and marked as a second type.

[0052] Considering that image-based PDF files typically have large file sizes (usually measured in MB) if they contain high-resolution scans, while text-based PDF files, which store text encoding, generally have smaller file sizes (usually measured in KB, unless a large number of high-definition images are embedded), in another embodiment of the present invention, file type classification can be based on the size of the original file. Specifically, the file information is the size of the original file. When the original file is in PDF format, its size is read; if the size of the original file is greater than a preset size threshold, the original file is identified as an image-based PDF and marked as a first type. If the size of the original file is less than or equal to the preset size threshold, the original file is identified as a text-based PDF and marked as a second type.

[0053] In practical use, the first and second types can be divided based on the metadata and / or the size of the original file. If the division result based on the metadata of the original file is inconsistent with the division result based on the size of the original file, it should be submitted to manual review and confirmation.

[0054] This invention addresses the unique characteristics of the PDF format by further differentiating between image-based and text-based PDFs through metadata or file size, thereby improving parsing accuracy.

[0055] S222-2 If the original file is in an audio stream-related format, then the original file is marked as a third type.

[0056] Audio stream formats include, but are not limited to, audio and video formats. Audio formats include, but are not limited to, .mp3, .wav, and .aac formats. Video formats include, but are not limited to, .mp4 and .wmv formats.

[0057] S222-3 When the original file is in a flowchart-related format, the type of the original file is marked as the fourth type.

[0058] Flowchart formats include, but are not limited to, image formats and flowchart formats. Image formats include, but are not limited to, .jpg, .jepg, and .bmp formats. Flowchart formats include, but are not limited to, .vsdx formats.

[0059] If the original file is identified as not being in PDF, audio stream, or flowchart format, it is determined that the original file is in a format that the multimodal file parsing engine cannot parse, and a warning is issued to facilitate manual intervention.

[0060] In one embodiment of the present invention, the type can also be manually marked by a person when collecting or uploading the original file.

[0061] In another embodiment of the present invention, when uploading original files, the type of the original files can be classified and marked based on different upload buttons.

[0062] S23 Matches the corresponding parsing strategy according to the type; In one embodiment of the present invention, the first type corresponds to the first parsing strategy; the second type corresponds to the second parsing strategy; the third type corresponds to the third parsing strategy; and the fourth type corresponds to the fourth parsing strategy.

[0063] The first type of original file is used as the first original file; the second type of original file is used as the second original file; the third type of original file is used as the third original file; and the fourth type of original file is used as the fourth original file.

[0064] This invention improves parsing efficiency and accuracy by matching appropriate parsing strategies to different types of files.

[0065] S24 uses the parsing strategy to parse the original file and obtains the parsing result for each original file; The parsing result includes at least one of text information and final path encoding information; like Figure 2 As shown, step S24 specifically includes: S241 parses the first original file into text information according to the first parsing strategy; S241-1 performs document repair on the first original file through the document repair subunit to generate an optimized first original file; Construct a document repair model; the preferred document repair model is a generative adversarial network (GAN) based on a U-Net generator and a PatchGAN discriminator.

[0066] In one embodiment of the present invention, the first original file is split into at least one document image in page order; the document images are traversed to determine the current document image. The document image is used as input to the document restoration model. The encoder performs multi-layer convolutions to progressively extract high-level semantic features from the current document image. These features include the overall layout of the document image and contextual information about text regions. The spatial resolution of the current document image is progressively reduced (e.g., from 512×512→256×256→…→32×32) using convolutions or max pooling. The decoder starts from the low-resolution feature map and progressively restores the spatial resolution based on transposed convolutions (e.g., from 32×32→64×64→…→512×512). After each upsampling step, the feature maps from the corresponding layers of the encoder are concatenated via skip connections, injecting low-level high-resolution details (such as text stroke edges) into the high-level semantic features to avoid blurring the restoration result. Finally, the restored document image is output. ,in, , For the learnable parameters, the loss function is: .

[0067] At the same time, the current document image With the repaired document image As input to the PatchGAN discriminator, it determines the authenticity of each local region in the image to improve the accuracy of the restoration.

[0068] The repaired document images are merged according to page order to obtain the optimized first original file.

[0069] By restoring document images, noise in scanned documents can be eliminated and the accuracy of OCR recognition can be improved.

[0070] S241-2 performs a first-level parsing on the optimized first original file to obtain text information related to the first original file; The hierarchical parser includes a first hierarchical parsing subunit; the optimized first original file is input into the first hierarchical parsing subunit for text recognition; Since the first original document contains image-based pages (such as scanned copies or PDFs generated from photographs), each page needs to undergo OCR processing. In one embodiment of the present invention, the content of each page in the optimized first original document is sequentially subjected to OCR text recognition according to the page order in the first original document, thereby converting the optimized first original document into text information.

[0071] S242 parses the second original file into text information according to the second parsing strategy; The second original file is subjected to a second layer of parsing to obtain text information related to the second original file; Layered parsing refers to parsing the document according to the hierarchical structure of the PDF standard, from the bottom byte stream to the high-level semantic information, to extract or manipulate the internal objects of the document.

[0072] In one embodiment of the present invention, the first original file is input into the first hierarchical parsing subunit for parsing. The execution flow is similar to steps S241-2, and will not be described again here.

[0073] In another embodiment of the present invention, the hierarchical parser further includes a second hierarchical parsing subunit; the second original file is input into the second hierarchical parsing subunit for text extraction; specifically: The process begins by reading the header of the second original file, locating the cross-reference table, and the trailer dictionary. The header confirms the PDF version of the second original file. The cross-reference table provides the offsets of all objects. The trailer dictionary provides the reference to the root object (Catalog). Based on the root object ( / Root), the page tree ( / Pages) is located, and page nodes ( / Page) are recursively traversed to associate page resources (e.g., fonts / Font, content streams / Contents, images / XObject). For content streams ( / Contents), drawing instructions (such as BT / ET marker text blocks and Tj / TJ output text) are parsed, and font encoding is used to convert the original instructions into readable text. For image objects, it is determined whether they are flowcharts; if so, they are captured and used as the fourth original file. The readable text is then summarized to obtain the text information corresponding to the second original file.

[0074] S243 parses the third original file into text information according to the third parsing strategy; S243-1 uses an end-to-end model-based speech recognition subunit to perform text conversion on the third original file, obtaining a timestamp-marked text sequence; In one embodiment of the present invention, the third original file is converted into the input format required by the end-to-end model; the speech-to-text mapping is directly learned through a single network, and the character probability distribution or attention weights are output; the output probability distribution is converted into a text sequence through greedy search or Beam Search decoding, and an alignment path is generated, i.e., the frame range corresponding to each character.

[0075] S243-2 performs semantic normalization on the text sequence marked with timestamps to obtain text information related to the third original file; Semantic normalization maps non-standard expressions to general, standardized written language, ensuring text consistency.

[0076] In one embodiment of the present invention, a semantic standardization database is pre-constructed; the semantic standardization database needs to store a one-to-one correspondence or many-to-one relationship between non-standard words and standard words. The semantic standardization database determines several reference entries; each reference entry includes: one standard word and at least one non-standard word. During semantic standardization, the timestamp-marked text sequence is separated into text content and timestamp; the text content is traversed, and word segmentation is performed; the semantic standardization database is queried word by word to find non-standard words in the text content and replace them with the corresponding standard words. The processed text content is then used as text information.

[0077] S244 parses the fourth original file according to the fourth parsing strategy to obtain the final path encoding information; S244-1 extracts structured topological features from the fourth original file; A graph convolutional network is used to parse the node relationships and flow logic in the fourth original file, extracting structured topological features, specifically: Corresponding each operation step in the fourth original file to the diagram structure A node; if the fourth original file includes Steps, node set ,in Indicates the first One operation step.

[0078] Map the flow relationships (such as sequence, branching, and looping) between the operation steps in the fourth original file to directed edges in the graph; if the steps Post-execution flow steps Then add directed edges. The set of edges It is used to reflect the logical control flow of the process (such as if-else branches, while loops, etc.).

[0079] graph structure Convert to matrix form; specifically, construct... Directed adjacency matrix If an edge exists ,but ,otherwise, Construct the degree matrix Degree matrix It is a diagonal matrix. (Out of bounds) or (in-degree).

[0080] By performing local aggregation operations on the adjacency matrix, the topological features of nodes are extracted layer by layer through a 3-layer GCN, i.e. Specifically: The first layer of features is output by aggregating the direct neighbor information of each node through an adjacency matrix. , ;in, This is used to represent the addition of self-loops while preserving the node's own information; yes The degree matrix; It is a learnable weight matrix with a size of , This is the first layer output dimension; It's an activation function. Based on the first layer of features, it aggregates information from more distant neighborhoods (such as neighbors of neighbors) and outputs the second layer of features. , ;in, It is the second-level weight matrix, with a size of , This is the second output dimension. It further aggregates global structural information to output the final topological features of each node. , ,in, It is the third layer weight matrix, with a size of , Output dimensions for the target. Final topological features. , is the topological feature matrix of each node, where each row... Represents a node Its topological features integrate structural information from itself and its global neighborhood.

[0081] S244-2 Encodes the decision path based on the topological features to obtain the final path encoding information.

[0082] Topological features are obtained through a decision path encoder, and then decision path encoding is performed.

[0083] In one embodiment of the present invention, a decision path is identified from topological features; the decision path is the specific execution trajectory from the starting point to the ending point in a process, consisting of a series of nodes (operation steps) arranged sequentially, and may contain decision points (such as conditional branches and loop control). The decision path is then converted into a sequence of path nodes. ; Path node sequence Each node in the table is replaced with its corresponding node topology feature. The initial feature sequence of the decision path is obtained. If the decision path contains decision points (such as branch nodes) This involves fusing the semantic information of the decision conditions with node features. The resulting node feature sequence is then... Input a sequence model (such as LSTM or Transformer) to capture the sequential dependencies and long-distance interactions of the decision path, obtaining the target node feature sequence. Then, through a fully connected layer or attention mechanism, fuse the sequence representation of the target node feature sequence with meta-information such as the start point, end point, and length of the path to obtain the final path encoding information. .

[0084] S25 concatenates all the parsing results to generate the modality fusion information; All text information is aggregated to obtain the final text information; this final text information is then encoded using the BERT base model to obtain BERT semantic features. The final path encoding information is used as a structural feature of the GCN. ; BERT semantic features and GCN structural features Dimensional unification is performed, and then the unified dimensions are concatenated to obtain the final 384-dimensional unified semantic representation. This serves as modality fusion information. In one embodiment of the invention, 768-dimensional BERT semantic features are projected onto 192-dimensional BERT semantic features through a linear layer; GCN structural features are then used as... The output is 192-dimensional; 192-dimensional BERT semantic features and 192-dimensional GCN structural features By concatenating the elements, a 384-dimensional vector is obtained. .

[0085] This invention is based on a multimodal file parsing engine. It performs structured semantic parsing on cross-format files such as PDF files, scanned documents, flowcharts, and voice recordings, and outputs modal fusion information with unified semantic representation.

[0086] S3 retrieves the modality fusion information to construct a knowledge graph for compliance management of biological samples; The knowledge graph is a four-dimensional relational network, including: standard layer, operation layer, resource layer and risk layer entity nodes.

[0087] S31 extracts entities at each preset level from the modal fusion information; The preset layers include: standard layer, operation layer, resource layer and risk layer.

[0088] S311 constructs a standard layer based on modal fusion information; The entity types in the standard layer include: ISO standard clauses and policy entries. In one embodiment of the present invention, attribute values ​​such as clause number, clause content, issuing agency, and effective date are extracted from the modal fusion information using a named entity recognition model and used as entities in the standard layer.

[0089] S312 constructs an operation layer based on modal fusion information; The entity types of the operation layer include: standard operating procedure steps (SOP steps) and equipment instructions; in one embodiment of the present invention, the attribute values ​​of operation attributes such as step number, operation description, execution equipment, and time requirements are extracted as entities of the operation layer.

[0090] S313 constructs a resource layer based on modal fusion information; The entity types of the resource layer include: equipment information, personnel information, and sample data. In one embodiment of the present invention, the attribute values ​​of resource attributes such as equipment ID, equipment type, personnel name, position, sample number, and sample type are extracted as entities of the resource layer.

[0091] S314 constructs a risk layer based on modal fusion information; The risk layer includes: deviation events and corrective actions; in one embodiment of the present invention, the attribute values ​​of risk attributes such as event ID, event description, occurrence time, scope of impact, corrective actions, and responsible persons are extracted as entities of the risk layer.

[0092] S32 combines the relationships between the preset levels and the entities of the preset levels to form a four-dimensional relationship network; S321 constructs the association relationship between preset levels by defining the semantic association between cross-level entities; Establish cross-layer relationships between the clause layer, operation layer, equipment layer, and risk layer. The pre-defined layer relationships include, but are not limited to: the standard layer. The relationship between the operation layer and the operation layer Resource layer relationships, resource layer Relationships within the risk layer, risk layer The relationships within the operational layer.

[0093] like Figure 3 As shown, specifically: S321-1 Constructing the Standard Layer The relationships within the operational layer; Standard layer The operational layer relationships are used to characterize the requirement relationships (has_requirement) to determine the normative constraints that clauses impose on operations. In one embodiment of the invention, the requirement relationships between standard layer entities (e.g., clauses) and operational layer entities (e.g., SOP steps) can be extracted based on preset rules (e.g., execute xx) or a supervised learning model (e.g., relation extraction BERT). Specifically, for example, the operation "temperature monitoring" and "daily recording of liquid nitrogen tank temperature" are required to be performed according to clause ISO 20387 5.3.2.

[0094] S321-2 Constructing the Operation Layer Resource layer relationships; Operational layer The resource layer relationships are used to characterize execution relationships (executes) to identify the resources explicitly invoked in the operation steps. For example, in the operation "daily record liquid nitrogen tank temperature", the equipment involved is liquid nitrogen tank A3, and the sensor invoked is PT100-047.

[0095] S321-3 Constructing the Resource Layer The relationships within the risk layer; resource layer The risk layer's correlations are used to characterize risk trigger relationships (triggers_risk) to identify risks caused by resource anomalies or improper use. Specifically, for example, in the case of "using sensor PT100-047 to record the temperature of liquid nitrogen tank A3," there is a deviation event: temperature exceeding the limit by >3℃.

[0096] S321-4 Constructing a Risk Layer The relationships within the operational layer; Risk layer The operational layer relationships are used to characterize risk remediation relationships (requires_fix) in order to determine the specific operational requirements for risk rectification. For example, a deviation event of exceeding the temperature limit will trigger the operation "start backup cooling".

[0097] Of course, after determining the actions to rectify the risk, the relevant clauses for the actions to rectify the risk will be determined based on the relationship between the standard layer and the operational layer (e.g., ISO 20387 7.1.4). Then, according to the relationship between the standard layer and the operational layer, the specific SOP steps will be determined.

[0098] In another embodiment of the invention, other cross-layer relationships can also be constructed, such as: Operational layer The risk layer's correlation is used to characterize the risk cause relationship (may_cause) in order to identify potential risk causes during operation (e.g., uncalibrated equipment may lead to measurement deviation).

[0099] S322 stores the entities extracted from each level and their cross-level relationships in a graph database, forming a four-dimensional relational network.

[0100] This invention defines semantic relationships between entities across layers to form a four-dimensional relationship network, enabling comprehensive association of standard terms, operational procedures, resource information, and risk events.

[0101] S33 performs conflict detection on the four-dimensional association network, optimizes the four-dimensional association network based on the detection results, and obtains the knowledge graph.

[0102] The four-dimensional association network is verified in real time through a conflict detection algorithm to ensure the consistency and accuracy of the knowledge graph.

[0103] S331 uses a pre-trained language model (e.g., Transformer) to vectorize the terms, resulting in low-dimensional dense vectors. , .

[0104] S332 constructs several clause pairs to be validated; Search for the vector of the clause to be verified from the entities at the operational level. Find all standard clause vectors from the entities in the standard layer. ; Based on the vector of terms to be verified and all standard clause vectors Construct several clause pairs to be verified; each clause pair includes: a clause vector to be verified. A standard clause vector That is, standard clause vector. The number of clauses is the same as the number of clause pairs.

[0105] S333 calculates the conflict score between the clause pairs to be verified. Among them, conflict score ; S334 If there is a conflict between the clauses, the score is... > If a conflict threshold is preset and the labels of the two clause vectors in a clause pair are different, then the two entities are marked as conflicting. The preset conflict threshold can be adjusted according to actual conditions. In one embodiment of the present invention, the preset conflict threshold is preferably 0.8. That is, ; S335 updates the primary four-dimensional correlation network based on the verification results of collision detection; Based on the above method, semantic conflicts between SOPs and the main standard are detected in real time. When the SOP version changes, expired associations are automatically disconnected and new paths are established.

[0106] This invention establishes a dynamic mapping between standard clauses and actual operational scenarios in the sample library based on knowledge graphs, so as to facilitate compliance verification of biological sample operations.

[0107] S4 retrieves the user's operation request; The operation request includes: the original operation instruction; like Figure 4As shown, compliance verification is performed after obtaining the user's input operation request.

[0108] S5 uses the knowledge graph to perform compliance verification on the operation request, ensuring that the operation complies with the standardized management requirements of the biobank.

[0109] S51 searches the knowledge graph for standard operation instructions related to the original operation instruction; Key features are determined based on the original operation instructions; key features include, but are not limited to, temperature information, time information, and biological sample information.

[0110] Based on the key features, relevant target standard clauses are searched from the knowledge graph; through the association between the standard layer and the operation layer, standard operation instructions are matched based on the target standard clauses to provide a basis for subsequent calculation of difference indicators.

[0111] S52 calculates the difference index between the original operation instruction and the standard operation instruction; As a preferred approach, compliance quantification verification of the original operation instructions is performed based on KL divergence. The difference in probability distribution is used to quantify the compliance distance between the original operation instructions and the standard operation instructions, so as to quickly determine the initial compliance.

[0112] Specifically, difference indicators ;in, This represents the probability distribution of the original operation instructions (such as the execution probability of operation steps, parameter selection preferences, etc.). This represents the probability distribution of standard operating instructions.

[0113] S53 performs compliance verification based on the relationship between the difference index and the preset difference threshold; If the difference index is less than the preset difference threshold, the verification passes and the user's operation request is deemed compliant.

[0114] If the difference index is greater than or equal to the preset difference threshold, the verification fails and the user operation is deemed to be in violation (non-compliant user operation).

[0115] The preset difference threshold is preferably 0.05. That is, .

[0116] S6 makes a compliance decision on the operation request based on the compliance verification results.

[0117] Based on the compliance verification results, make compliance decisions on operation requests to ensure that the execution or modification of operation requests complies with the standardized management requirements of biobanks.

[0118] S61 When the compliance verification result is that the verification is passed, the success rate of the original operation instruction is evaluated, and the operation request is executed based on the success rate. S611 constructs an operation instruction package based on the original operation instructions and the knowledge graph; The operation instruction package includes: original operation instructions, associated equipment, and acceptance criteria.

[0119] In one embodiment of the present invention, the original operation instructions and associated devices are extracted from the operation request; the acceptance criteria are searched from the knowledge graph based on the original operation instructions; and the operation instruction package is generated by combining the original operation instructions, associated devices, and acceptance criteria.

[0120] 612 Input the operation instruction package into the pre-simulation model to predict the operation success rate; The pre-visualization model is used to predict the success rate of operations in the Unity3D virtual environment.

[0121] S612-1 Constructing a virtual sample library; Real-world operation scenarios are reproduced in Unity3D by integrating elements such as sensors, devices, and human interaction to build a virtual sample library that covers normal, abnormal, and boundary scenarios.

[0122] S612-2 executes operation instruction packages in a virtual sample library via Monte Carlo simulation. The operation process is repeated to obtain the single simulation success rate for each step. The single simulation success rate is the product of the conditional probabilities of each action in the entire operation cycle (assuming that each step is independent).

[0123] Single simulation success rate = ;in, For the first The status of the step (such as device parameters, environment variables); For corresponding actions (such as operation commands, parameter adjustments).

[0124] S612-3 calculates the average success rate of all single simulations as the operational success rate; operational success rate .

[0125] This invention verifies the actual success rate of operations through virtual simulation, thereby reducing the risk of operational errors.

[0126] S613 When the success rate of the operation is greater than or equal to the preset success threshold, the original operation instruction is sent to the execution terminal.

[0127] After the user's electronic signature is confirmed, the corresponding user operation request is executed.

[0128] S614 When the success rate of the operation is less than the preset success threshold, manual review is initiated.

[0129] S62 When the compliance verification result is that the verification fails, a compliance suggestion is generated for the original operation instruction.

[0130] In one embodiment of the present invention, the target standard clause is regarded as a conflict clause; the conflict clause is located, the standard operating instructions are found, and compliance suggestions are given based on the differences between the standard operating instructions and the original operating instructions.

[0131] By summarizing conflicting clauses and historical cases, a deviation analysis report is generated. Combining the deviation analysis report with a knowledge graph, revised compliance recommendations are generated for users' reference.

[0132] This invention connects to a knowledge graph and performs compliance checks on user operation requests, then outputs approval instructions or deviation correction schemes.

[0133] To improve query and tracing performance, it also includes: S7 generating 3D tracing information; specifically: Obtain process information for the current compliance decision; obtain the corresponding biological sample ID based on the original operation instruction; associate the biological sample ID with the process information to generate three-dimensional traceability information.

[0134] The preferred three-dimensional traceability information is an audit traceability matrix. The audit traceability matrix is ​​a three-dimensional chain of evidence, with each audit traceability matrix associated with a biological sample ID.

[0135] The audit traceability matrix includes a timeline (X-axis), a clause version axis (Y-axis), and a chain of evidence axis (Z-axis). The timeline (X-axis) records the request time of the operation request, the compliance decision time, and the actual execution time. For example, it records "Operation instruction generation timestamp → Execution completion time." The clause version axis (Y-axis) records the relevant standard entities and operational entities. For example, it records "ISO 20387:2025 → Organization SOP v3.2 → Supplementary Clauses." The chain of evidence axis (Z-axis) records the complete association information between data and operations, including but not limited to: operation logs, transportation record screenshots, and user electronic signatures. For example, it records "Knowledge graph nodes / user confirmation records / equipment execution logs." All information in the timeline (X-axis), clause version axis (Y-axis), and chain of evidence axis (Z-axis) is located on the same spatiotemporal coordinate system.

[0136] Preferably, the audit traceability matrix can be stored in the operation log database. During traceability, the corresponding audit traceability matrix can be retrieved from the operation log database based on the biological sample ID. The audit traceability matrix can also be broken down into standardized operational process nodes (e.g., receiving → sorting → storing → outbound) according to the biological sample's flow process, linking different processes for the same biological sample.

[0137] In one embodiment of the present invention, the temperature exposure integral can also be calculated and recorded based on actual operation. To quantify the extent to which products are exposed to unsuitable temperature environments, and to record quality control evidence such as freeze-thaw cycles, centrifugal G-value curves, and quality inspection reports.

[0138] This invention uses biological sample ID as the core identifier to establish a mapping relationship between sample ID and other dimensions of data, supporting rapid location and backtracking of the complete operation chain from any dimension starting point, thereby improving query and tracing effects.

[0139] In one embodiment of the present invention, compliance verification and correction can also be performed based on conflict detection. For example... Figure 5 As shown, the user's operation request includes: weekly thermometer calibration. After verification based on the knowledge graph, the target standard instruction is matched. The target standard instruction includes matching clauses and matching operations. The matching clauses include: ISO 5.3.2: Daily calibration. The matching operations (organizational SOP) include: weekly calibration. Conflict detection is performed on the matching clauses and matching operations to obtain a conflict score. Conflict detection is similar to step S33 and will not be described again here.

[0140] Based on the relationship between the conflict score and the preset conflict threshold, the compliance of a user's operation request is determined. Specifically, if the conflict score is greater than the preset conflict threshold (0.7), the user's operation request is deemed non-compliant and rejected. When a conflict arises between the matching terms and the matching operation, the operation layer needs to be updated. This involves providing a correction suggestion based on the matching terms, which is then sent to the user as a compliance recommendation. The correction suggestion includes: determining the modified operation instruction (changing it to daily calibration), providing the referenced terms (referencing ISO 5.3.2), and updating the operation version (updating the SOP to v3.3). An audit traceability matrix is ​​recorded. The audit traceability matrix includes: Time axis (X-axis): 2025-06-21 14:30; Terms version axis (Y-axis): ISO 5.3.2 vs. SOP 4.1; Evidence chain axis (Z-axis): Electronic signature confirmation of user ID203.

[0141] To facilitate distributed data storage and traceability across multiple institutions, this invention utilizes blockchain technology for evidence preservation. Specifically, operation logs, entity version hash values, and other data can be stored on the blockchain evidence preservation module to achieve evidence fixation. Furthermore, other information such as knowledge graphs can also be stored or accessed through the blockchain evidence preservation module.

[0142] In one embodiment of the present invention, the invention can be applied to compliance decisions in the biological sample receiving process. Specifically: The original documents are obtained, including: "Sample Receiving Operation Procedure_v3.0.pdf" containing text and flowcharts, and "ISO 20387:2025" with handwritten annotations; based on the output of the multimodal document parsing engine, key clauses and flowcharts are obtained; the key clause is: when receiving frozen samples, the transport temperature record must be checked, and if it is >-150℃ for 2 consecutive hours, the sample will be rejected; the flowchart generates a receiving decision tree containing 6 decision nodes. A knowledge graph is constructed based on the key clauses and flowcharts. At 14:05 on 2025-06-21, a user's operation request is obtained, which includes: receiving lung cancer samples transported in liquid nitrogen (transport record: -120℃ for 3 consecutive hours).

[0143] After comparing the operation request with the knowledge graph, a validation failure was detected at 14:07 on June 21, 2025, with the decision to reject the request. The relevant clause was ISO 20387 7.2.1. The conflict analysis indicated that the transport temperature exceeded the limit (threshold: -150℃). Following the validation failure, compliance recommendations were provided at 14:10 on June 21, 2025, including: 1. Completing the "Sample Rejection Form"; 2. Notifying the logistics provider to retrieve the samples; 3. Updating the supplier evaluation report.

[0144] Record three-dimensional traceability information, including: Timeline (X-axis): 2025-06-21 14:05 (request time) → 14:07 (decision time) → 14:10 (execution time). Clause version axis (Y-axis): ISO 20387 7.2.1 → Organization SOP5.3. Chain of evidence axis (Z-axis): Operation log + screenshot of transportation record + user electronic signature.

[0145] To achieve synchronized policy updates under a multi-agency distributed deployment, in one embodiment of the present invention, a federated learning mechanism is constructed, based on Paillier homomorphic encryption gradient aggregation, to enable rapid updates of standard information at each edge node.

[0146] Set up several edge nodes Edge nodes are institutions responsible for the compliance management of biometric samples, directly participating in the entire policy update process. Generate Paillier key pairs (public key...). Private key Distribute to all edge nodes Standard information for initializing the local model. Set the learning rate The standard information includes policy updates.

[0147] For each edge node Using local data Calculate current standard information loss Backpropagation yields the gradient. ; Calculation unit standard information During local training, standard information is trained and updated locally based on current standard information and unit standard information. .

[0148] edge nodes Use the Paillier key pair (where the public key is) ) information on current standards Encryption is performed to obtain the encrypted standard information. , ; encrypt the standard information As feedback on the new policy, it is uploaded to the server to ensure gradient privacy during transmission.

[0149] Collect encrypted standard information from all edge nodes For each encrypted standard message Decryption is performed to obtain the decrypted standard information. The server aggregates the encrypted standard information from all edge nodes to obtain global standard information. , The global standard information is broadcast as a policy update packet to all edge nodes; each edge node updates its local current standard information. Achieve policy consistency across multiple nodes. Store global standard information in a policy version repository. Based on a federated learning update mechanism, the repository can also be used to retrieve corresponding standard information as needed.

[0150] This invention enables synchronized policy updates in a distributed multi-agency deployment through a federated learning mechanism, ensuring the consistency and timeliness of policies / clauses across agencies.

[0151] like Figure 6As shown, this invention relates to a file input layer, an intelligent decision-making layer, an output application layer, and a data storage layer. The file input layer is used to acquire multimodal raw files and execute the multimodal file parsing engine. The intelligent decision-making layer is used to perform knowledge graph construction, compliance verification, and federated learning updates. The output application layer is used for compliance decision-making, audit attribution matrix construction, and data transmission and reception at edge nodes. The data storage layer includes: blockchain notarization, a policy version repository, and an operation log repository.

[0152] This invention achieves semantic parsing and compliance control through a full-chain management approach encompassing document parsing, knowledge construction, decision control, and traceability auditing. It is suitable for scenarios requiring strict adherence to biosample management standards, such as hospital biobanks, drug development centers, and epidemiological research institutions.

[0153] This invention is based on the intersection of bioinformatics, natural language processing, and quality management engineering. It addresses three core issues: multimodal semantic understanding, clause-operation mapping, and real-time compliance control. By integrating graph neural networks, knowledge graphs, and federated learning techniques, it achieves intelligent parsing and implementation supervision of standardized documents for biobanks based on deep learning.

[0154] Figure 7 A schematic diagram of a compliance decision-making system based on biological sample management, provided for embodiments of this specification, includes: The file input module 710 is used to acquire the original multimodal files; the original multimodal files are related to the standardized management of the biobank. A multimodal file parsing engine 720 is used to parse and process the original multimodal file to obtain modality fusion information; The knowledge graph construction module 730 is used to retrieve the modality fusion information and construct a knowledge graph for biosample compliance management. The request acquisition module 740 is used to acquire the user's operation request; the operation request includes: original operation instructions; The compliance control module 750 is a compliance control intelligent agent used to perform compliance verification on the operation request in conjunction with the knowledge graph; The compliance decision module 760 is used to make compliance decisions on the operation request based on the compliance verification results.

[0155] Optionally, the compliance control module 750 includes: The instruction lookup submodule is used to search for standard operation instructions related to the original operation instruction from the knowledge graph; The difference index calculation submodule is used to calculate the difference index between the original operation instruction and the standard operation instruction; The compliance verification submodule is used to perform compliance verification based on the relationship between the difference index and the preset difference threshold. Optionally, the multimodal file parsing engine 720 includes: The type identification submodule is used to identify the type of each original file; The strategy matching submodule is used to match the corresponding parsing strategy according to the type. The parsing submodule is used to parse the original file using the parsing strategy, and obtain the parsing result for each original file. The unified semantic representation submodule is used to concatenate all the parsing results to generate the modality fusion information.

[0156] Optionally, the parsing result includes at least one of text information and final path encoding information; Optionally, the parsing submodule includes: The first parsing unit is used to parse the first original file into text information according to the first parsing strategy; The second parsing unit is used to parse the second original file into text information according to the second parsing strategy; The third parsing unit is used to parse the third original file into text information according to the third parsing strategy; The fourth parsing unit is used to parse the fourth original file according to the fourth parsing strategy to obtain the final path encoding information; Optionally, the first parsing unit includes: The document repair subunit is used to repair the first original file and generate an optimized first original file. The first hierarchical parsing subunit performs first hierarchical parsing on the optimized first original file to obtain text information related to the first original file; Optionally, the second parsing unit includes: The second hierarchical parsing subunit is used to perform a second hierarchical parsing on the second original file to obtain text information related to the second original file; and / or, Optionally, the third parsing unit includes: The speech recognition subunit based on the end-to-end model is used to perform text conversion on the third original file to obtain a timestamp-marked text sequence; The semantic normalization subunit is used to perform semantic normalization on the text sequence marked by the timestamp to obtain text information related to the third original file; Optionally, the fourth parsing unit includes: Optionally, a feature extraction subunit based on a graph convolutional network is used to extract structured topological features from the fourth original file; A decision path encoder is used to encode the decision path based on the topological features to obtain the final path encoding information.

[0157] Optionally, the knowledge graph construction module 730 includes: The entity extraction submodule is used to extract entities at various preset levels from the modal fusion information; The association network generation submodule is used to combine the association relationships between the preset levels and the entities of the preset levels to form a four-dimensional association network; The conflict detection submodule is used to perform conflict detection on the four-dimensional association network, optimize the four-dimensional association network based on the detection results, and obtain the knowledge graph.

[0158] Optionally, the compliance decision module 760 includes: The first decision submodule is used to evaluate the success rate of the original operation instruction when the compliance verification result is that the verification is passed, and to determine whether to execute the operation request based on the success rate. The second decision submodule is used to generate compliance suggestions for the original operation instructions when the compliance verification result is that the verification fails.

[0159] Optionally, the second decision submodule includes: An operation instruction package construction unit is used to construct an operation instruction package based on the original operation instructions and the knowledge graph. An operation success rate prediction unit is used to input the operation instruction package into the pre-simulation model to predict the operation success rate. The decision unit is used to send the original operation instruction to the execution terminal when the success rate of the operation is greater than or equal to a preset success threshold.

[0160] Optionally, it also includes: a traceability module; optionally, the traceability module includes: The information acquisition submodule is used to acquire process information for the current compliance decision-making process; The ID acquisition submodule is used to acquire the corresponding biological sample ID according to the original operation instruction; The traceability association submodule is used to associate the biological sample ID with the process information to generate three-dimensional traceability information.

[0161] The functions of the system in this embodiment have been described in the above method embodiments. Therefore, for any parts not detailed in this embodiment, please refer to the relevant descriptions in the foregoing embodiments, which will not be repeated here.

[0162] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this specification. The electronic device includes a memory 801 and a processor 802. The memory 801 is used to store computer-executable instructions. When the computer-executable instructions are executed by the processor 802, they can implement the steps of the above-described method embodiments.

[0163] Figure 9 This is a schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of this specification. The computer-readable storage medium 900 stores one or more computer programs, which, when executed by a processor, can implement the steps of the above-described method embodiments.

[0164] This specification also provides a computer program product, including a computer program / computer executable instructions, which, when executed by a processor, can implement the steps of the above-described method embodiments.

[0165] Those skilled in the art will understand that all or part of the processes in the above method embodiments can be implemented by a computer program instructing related hardware. When the computer program is executed, it may include the processes of the embodiments of the above methods.

[0166] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A compliance decision-making method based on biological sample management, characterized in that, include: Obtain the original files of the multimodal samples; the original files of the multimodal samples are related to the standardized management of the biobank. The original multimodal file is parsed and processed by a multimodal file parsing engine to obtain modality fusion information; Retrieve the modality fusion information to construct a knowledge graph for compliance management of biological samples; Obtain the user's operation request; The operation request includes: the original operation instruction; The operation request is validated for compliance using the knowledge graph; wherein, a standard operation instruction related to the original operation instruction is searched from the knowledge graph; a difference index between the original operation instruction and the standard operation instruction is calculated; and compliance validation is performed based on the relationship between the difference index and a preset difference threshold. The operation request is subject to compliance decision based on the compliance verification results.

2. The method as described in claim 1, characterized in that, The process of parsing the original multimodal file using a multimodal file parsing engine to obtain modality fusion information includes: Identify the type of each original file; Match the corresponding parsing strategy according to the type; The original files are parsed using the aforementioned parsing strategy to obtain the parsing results for each original file. All the parsing results are concatenated to generate the modality fusion information.

3. The method as described in claim 2, characterized in that, The parsing result includes at least one of text information and final path encoding information; The process of parsing the original file using the parsing strategy to obtain the parsing result for each original file includes: The first original file is parsed into text information according to the first parsing strategy; wherein, document repair is performed on the first original file to generate an optimized first original file; the optimized first original file is then subjected to a first layered parsing to obtain text information related to the first original file; and / or, The second original file is parsed into text information according to the second parsing strategy; wherein, the second original file is subjected to a second hierarchical parsing to obtain text information related to the second original file; and / or, The third original file is parsed into text information according to the third parsing strategy; wherein, the third original file is converted into a text sequence with timestamps; the text sequence with timestamps is semantically normalized to obtain text information related to the third original file; and / or, The fourth original file is parsed according to the fourth parsing strategy to obtain the final path encoding information; wherein, structured topological features are extracted from the fourth original file; and the decision path is encoded according to the topological features to obtain the final path encoding information.

4. The method as described in claim 1, characterized in that, The process of retrieving modality fusion information and constructing a knowledge graph for biosample compliance management includes: Extract entities at each preset level from the modal fusion information; By combining the relationships between the preset levels and the entities at the preset levels, a four-dimensional relationship network is formed; Conflict detection is performed on the four-dimensional association network, and the network is optimized based on the detection results to obtain the knowledge graph.

5. The method as described in claim 1, characterized in that, The compliance decision-making process for the operation request based on the compliance verification results includes: When the compliance verification result is that the verification is passed, the success rate of the original operation instruction is evaluated, and the operation request is executed based on the success rate. When the compliance verification result is that the verification fails, a compliance suggestion is generated for the original operation instruction.

6. The method as described in claim 5, characterized in that, When the compliance verification result is successful, the success rate of the original operation instruction is evaluated, and the operation request is executed based on the success rate, including: Construct an operation instruction package based on the original operation instructions and the knowledge graph; The operation instruction package is input into the pre-simulation model to predict the operation success rate; When the success rate of the operation is greater than or equal to the preset success threshold, the original operation instruction is sent to the execution terminal.

7. The method as described in claim 1, characterized in that, Also includes: Obtain information about the current compliance decision-making process; Obtain the corresponding biological sample ID according to the original operation instructions; The biological sample ID is associated with the process information to generate three-dimensional traceability information.

8. A compliance decision-making system based on biological sample management, characterized in that, include: The file input module is used to acquire the original files of the multimodal samples; the original files of the multimodal samples are related to the standardized management of the biobank. A multimodal file parsing engine is used to parse and process the original multimodal files to obtain modality fusion information; The knowledge graph construction module is used to retrieve the modality fusion information and construct a knowledge graph for compliance management of biological samples. The request retrieval module is used to retrieve user operation requests; The operation request includes: the original operation instruction; The compliance control module is used to perform compliance verification on the operation request in conjunction with the knowledge graph; wherein, it searches for standard operation instructions related to the original operation instruction from the knowledge graph; calculates the difference index between the original operation instruction and the standard operation instruction; and performs compliance verification based on the relationship between the difference index and a preset difference threshold. The compliance decision module is used to make compliance decisions on the operation request based on the compliance verification results.

9. A computer device, characterized in that, The computer device includes: Processor; and, A memory storing computer-executable instructions, which, when executed, cause the processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs / instructions, which, when executed by a processor, implement the method as described in any one of claims 1-7.

Images