Laboratory intelligent management and remote review system

Abstract

The invention discloses a laboratory intelligent management and remote review system, and relates to the field of laboratory management, and the system comprises a standard atlas database construction module which is used for generating a standard process atlas of various sample detection tasks; the sample process map generation module is used for collecting actual operation processes of experimenters and generating map representation; the anomaly detection module is used for carrying out structure comparison on the actual process map and the standard process map and identifying an abnormal process path; the abnormal graph learning module is used for classifying the abnormal process graph into an abnormal graph library and carrying out clustering and labeling processing; the abnormal evolution prediction and early warning module is used for predicting a process evolution trend based on a graph neural network model and giving out an early warning when the process evolution trend deviates from a standard; and the remote review intelligent auxiliary module is used for generating a structured review report. According to the method, structured modeling of the laboratory process, inductive precipitation of the abnormal mode and prospective early warning of the process risk are realized.

Description

Technical Field

[0001] The present invention relates to the field of laboratory management, and more specifically, to a laboratory intelligent management and remote review system. Background Art

[0002] During laboratory testing, sample transfer, processing, and report issuance typically follow a highly standardized set of procedures. For the same sample type, there are clear standards for testing tasks, step sequence, operator roles, time controls, and equipment usage. These procedures are highly reproducible and auditable. This standardization gives laboratory testing a strongly structured and constrained process, a key characteristic that distinguishes it from typical industrial or logistics processes.

[0003] However, in actual operations, due to human error, misunderstanding of the process, abnormal equipment status, or human circumvention, a series of process deviations may still occur during the sample execution process, such as missing operating steps, reversing the execution order, skipping the signature process, and issuing reports before testing. If these deviations are not identified in a timely manner, they will seriously affect the validity and traceability of the test results.

[0004] Although current laboratory management systems can record operation logs and sample status, they are mostly presented in the form of linear data structures or tables. They lack the ability to model the overall process structure, making it difficult to conduct in-depth comparisons between actual operation paths and standard processes. There is also a lack of mechanisms to systematically classify problems that have occurred, abstract them into patterns or maps, and form a knowledge accumulation of "process errors." In remote review scenarios, review experts usually rely on static records or video spot checks. They cannot quickly and accurately determine whether a process has deviated or whether there are potential risk nodes. It is also difficult to promptly identify process trends that have not yet been fully exposed but are deviating from the standard. Therefore, a new type of intelligent management and remote review system is needed to improve the controllability of the detection process, the automation of review work, and the accuracy of anomaly identification. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a laboratory intelligent management and remote review system to solve the problems mentioned in the background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: The present invention provides a laboratory intelligent management and remote review system, comprising: A standard map library construction module is used to construct standard process maps based on different sample types and detection tasks. Each map consists of multiple nodes, each of which includes standard operation steps and their constraint attributes. The order of operations is represented by directed edges between nodes. The sample flow chart generation module is used to collect the actual operation process of the experimenter and generate the actual operation flow chart; An anomaly detection module is used to compare the structure of the actual operation flow chart with the corresponding standard flow chart to identify the actual operation flow chart with anomalies; The abnormal graph learning module is used to classify the actual operation process graphs with abnormalities into the abnormal graph library, and perform structural clustering and labeling on the actual operation process graphs to form a typical violation set; The abnormal evolution prediction and early warning module is used to learn the structural characteristics and process evolution trends of the actual operation process maps in the abnormal map library based on the graph neural network model, predict the evolution risk of the current sample process, and issue early warning prompts when it deviates from the compliance process.

[0007] In some embodiments, the system further comprises: The remote review intelligent assistance module is used to generate a structured review report based on anomaly detection results and evolution risk prediction and provide it to remote review experts.

[0008] In some embodiments, the structured review report includes annotations of abnormal nodes, associated standard process maps, and references to historical abnormal maps in an associated abnormal map library.

[0009] In some embodiments, the standard map library construction module generates a standard process map by parsing laboratory process procedure documents or historical compliance sample process data. The constraint attributes of the node include: allowed operation time range, specified operation position role, limited equipment type and key quality control conditions; each sample type's specific detection task corresponds to a standard process map.

[0010] In some embodiments, the sample flow chart generation module is used to collect sample operation events based on laboratory management information system logs, video behavior recognition results and equipment operation records, and generate an actual operation flow chart with the same structure as the corresponding standard flow chart based on the sample operation events.

[0011] In some embodiments, the anomaly detection module uses a structural comparison algorithm based on graph edit distance to perform node mapping and path alignment on the actual operation process map and the corresponding standard process map, and identifies the actual operation process map as an anomaly when any of the following conditions is met: a) The graph edit distance is greater than the preset threshold; b) Nodes are missing, repeated, or inverted during the mapping process; c) The timestamp of the mapped node exceeds the time window allowed by the standard node; d) The operator role or device type of the mapped node does not conform to the standard node constraint attributes; Among them, the mapped nodes are nodes in the actual operation process map, and the standard nodes are nodes in the standard process map.

[0012] In some embodiments, the anomaly detection module further includes a machine learning identification submodule, which uses a trained classification model to determine whether the actual operation process map is an anomaly map, and outputs a corresponding anomaly type label and confidence score.

[0013] In some embodiments, the classification model is trained using a supervised learning method by using a plurality of actual operation process maps collected historically as training samples, labeling whether they are abnormal and the type of abnormality.

[0014] In some embodiments, the abnormal graph learning module performs graph structure embedding and similarity clustering operations on the actual operation process graphs identified as abnormal, uses a graph embedding algorithm to convert each graph into a vector representation, clusters abnormal process patterns based on vector distance, and automatically generates violation labels for each type of abnormal graph to form a typical violation set.

[0015] In some embodiments, the abnormal evolution prediction and early warning module uses a graph neural network model to perform graph embedding training on the graph structure patterns in the abnormal graph library, combines the time series node evolution characteristics to construct a process evolution trend model, and calculates the similarity change curve with the abnormal graph on the basis of the phased graph construction of the current sample process. When it is predicted that its evolution path approaches any abnormal graph category and reaches the risk threshold, it automatically generates an early warning prompt and identifies the potential risk nodes.

[0016] The advantages of the present invention over the prior art are: The present invention uses a standard graph library construction module and a sample process graph generation module to construct graph structures for standard processes and actual operation processes, respectively. This overcomes the problem that traditional linear records are difficult to capture process structure differences, and provides a unified expression and comparison basis for the operation paths between different samples. The anomaly detection module of the present invention uses the graph edit distance algorithm for structural comparison, combined with node attribute consistency verification, to achieve multi-dimensional recognition of complex deviation behaviors such as node missing, misordered, duplicated, role violations, and time anomalies. At the same time, it introduces a machine learning recognition submodule to improve the detection ability of boundary anomalies and enhance the adaptability and accuracy of the model. This invention uses an abnormal graph learning module to perform graph structure clustering and labeling on historical abnormal processes to form a typical violation set, providing basic support for subsequent rapid identification of similar problem samples, improving system interpretability and expert tracing efficiency; This invention introduces a graph neural network model to model the abnormal graph structure and process evolution characteristics. It can perform real-time graph updates and trend judgments on the currently executing process, and predict its deviation direction and abnormal probability before the process is completed, thereby intervening in advance and effectively preventing errors. This invention uses a remote review intelligent auxiliary module to integrate the identified abnormal nodes, corresponding standard process references, historical abnormality map references and warning information into a structured visual report output, significantly improving the judgment efficiency and decision reliability of review experts, and solving the problem of traditional review that "it is difficult to see and find"; The present invention can adapt to a variety of laboratory types and detection task scenarios, supports dynamic maintenance of graph structure templates and continuous optimization of learning models, has the ability of long-term operation and continuous enhancement, and is suitable for building a unified, efficient and intelligent laboratory remote supervision system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the overall functional module structure of the present invention; Figure 2 This is a flow chart of the standard atlas library construction module of the present invention; Figure 3 This is a flowchart of the actual operation flow chart generation and anomaly detection of the present invention; Figure 4 It is a processing flow chart of the abnormal evolution prediction and remote review auxiliary module of the present invention. DETAILED DESCRIPTION

[0018] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0019] This intelligent laboratory management and remote review system aims to achieve intelligent management and remote review support for laboratory operational processes through graph technology, anomaly detection, machine learning, and graph neural networks. The system primarily consists of multiple functional modules, including standard graph library construction, sample flow graph generation, anomaly detection, anomaly graph learning, anomaly evolution prediction and early warning, and intelligent remote review assistance.

[0020] Specifically, as shown in FIG1 , the system of the present invention includes: A standard map library construction module is used to construct standard process maps based on different sample types and detection tasks. Each map consists of multiple nodes, each of which includes standard operation steps and their constraint attributes. The order of operations is represented by directed edges between nodes. The sample flow chart generation module is used to collect the actual operation process of the experimenter and generate the actual operation flow chart; An anomaly detection module is used to compare the structure of the actual operation flow chart with the corresponding standard flow chart to identify the actual operation flow chart with anomalies; The abnormal graph learning module is used to classify the actual operation process graphs with abnormalities into the abnormal graph library, and perform structural clustering and labeling on the actual operation process graphs to form a typical violation set; The abnormal evolution prediction and early warning module is used to learn the structural characteristics and process evolution trends of the actual operation process maps in the abnormal map library based on the graph neural network model, predict the evolution risk of the current sample process, and issue early warning prompts when it deviates from the compliance process.

[0021] More specifically, the Standard Spectrum Library Construction module is used to generate standard process maps for different sample types and detection tasks. These maps are represented as directed graphs, where each node corresponds to a standard operation step, and nodes are connected by directed edges to reflect the order of operations. Each node also has constraint attributes that define the compliance requirements of the operation.

[0022] The construction of a standard process map begins with a structured representation of laboratory operational procedures. For example, for a sample test in a chemical laboratory, a standard process map might include the following nodes: "Sample Receipt," "Sample Pretreatment," "Instrumental Analysis," "Data Processing," and "Result Review." These nodes are connected sequentially by directed edges, for example, from "Sample Receipt" to "Sample Pretreatment" and then to "Instrumental Analysis." Constraint attributes for each node include the following: The allowed operation time frame, such as "sample preparation" must be completed within 12 hours after sample receipt.

[0023] Designated operating position roles, such as "instrument analysis", can only be performed by personnel holding instrument operation certificates.

[0024] The equipment type is limited, for example, "instrumental analysis" requires the use of a spectrophotometer model "X-2000".

[0025] Key quality control conditions, such as "data processing", must ensure that the data calibration error is less than 0.5%.

[0026] As shown in Figure 2, a standard process map can be generated in two ways. The first is to parse the laboratory's standard operating procedures (SOPs). The system uses natural language processing to extract operational steps and constraints from the SOPs. For example, the SOP might stipulate that "samples must be registered immediately upon receipt and pre-processing must be completed within 24 hours." Based on this, the system generates a map containing the "Sample Receipt" and "Sample Pre-processing" nodes, and sets a time constraint for the "Sample Pre-processing" node.

[0027] The second approach is to analyze historical compliant sample process data. By mining the historical operation records of compliant samples in the laboratory management information system, the system can identify common operation paths and constraints. For example, after analyzing the operation records of 1,000 compliant samples, the system found that the "instrumental analysis" step is usually performed after the "sample pretreatment" step and takes no more than two hours, thus generating a corresponding standard process map.

[0028] Assume that a laboratory needs to test the heavy metal content in a water sample. A standard process map may include the following nodes and constraints: Node 1: Sample reception (time range: registration within 1 hour after receipt, role: sample administrator).

[0029] Node 2: Sample pretreatment (timeframe: completed within 24 hours after receipt, equipment: ultrasonic cleaner).

[0030] Node 3: Instrumental analysis (timeframe: completed within 2 hours after pretreatment, equipment: atomic absorption spectrometer, role: analyst).

[0031] This information is organized into a directed graph structure, which serves as a benchmark for subsequent process comparisons.

[0032] In a further embodiment, the sample flow chart generation module is responsible for collecting the actual operation conditions of the experimenter on the specific sample and generating an actual operation flow chart that is consistent with the standard flow chart structure.

[0033] As shown in Figure 3, the collection of actual operational data relies on multiple sources. First, through the laboratory management information system's logs, the system can obtain sample receipt time, distribution status, and operation records. Second, using video behavior recognition technology, cameras within the laboratory monitor the movements of laboratory personnel in real time. For example, the system can identify when laboratory personnel begin "sample preparation" or use a specific instrument for "instrument analysis." Finally, equipment operation records provide detailed information on instrument usage, such as start time and operating parameters.

[0034] Based on the collected data, the system maps the operation events to the nodes of the standard process map in chronological order to generate the actual operation process map. Taking the heavy metal detection of water samples as an example, the system may record: Node 1: Sample reception (actual time: 2023-10-01 09:00, operator: Zhang).

[0035] Node 2: Sample pretreatment (actual time: 2023-10-01 10:00, equipment: ultrasonic cleaner).

[0036] Node 3: Instrumental analysis (actual time: 2023-10-01 12:30, equipment: atomic absorption spectrometer).

[0037] The generated actual operation flow chart maintains the same node and directed edge structure as the standard flow chart, but records the actual execution status to facilitate subsequent comparison and analysis.

[0038] As shown in Figure 4, in a further embodiment, the anomaly detection module identifies anomalies by comparing the structure of the actual operation flow graph with the standard flow graph. Its core algorithm is based on the graph edit distance. In other embodiments, machine learning techniques can be used or supplemented.

[0039] Structural alignment is divided into three steps: 1. During the node mapping process, the nodes of the actual operation process map are matched with the nodes of the standard process map. For example, the actual "Sample Preparation" node is mapped to the standard "Sample Preparation" node.

[0040] 2. During the path alignment process, based on the node mapping, the directed edge paths of the two graphs are compared to check whether there are any missing, duplicated or incorrect sequences.

[0041] 3. During the attribute verification process, each mapping node is checked to see if its attributes meet the standard requirements, such as whether the time exceeds the limit and whether the operator role is compliant.

[0042] The calculation of graph edit distance is used to quantify the difference between two graphs. Assume that the standard flow graph is The actual operation flow chart is ,in G Represents a graph, Represents a collection of nodes, Represents a directed edge set, subscript s and a They represent the standard flow chart and the actual operation flow chart respectively.

[0043] The graph edit distance GED is defined as G a Convert to G s The minimum number of operations required, including inserting, deleting, and replacing nodes or edges. The formula is as follows:

[0044] The abnormality is determined based on any of the following conditions: The graph edit distance is more than 5, which is preferred; Nodes are missing (e.g., "Instrumental Analysis" is missing), repeated (e.g., "Sample Preparation" is performed twice), or the order is reversed (e.g., "Instrumental Analysis" is before "Sample Preparation"); The node time exceeds the standard range (e.g. “sample preparation” is completed 48 hours after receipt); The operator role or equipment does not meet the requirements (e.g., "instrumental analysis" is performed by an intern rather than an analyst).

[0045] To enhance detection capabilities, in another embodiment, the present invention introduces a machine learning recognition submodule. This submodule uses historical operation flow graphs as training samples, annotating them with abnormalities and the type of abnormality (e.g., "time limit exceeded," "step missing"). A random forest model is used for training. The model input is the graph's feature vector (including the number of nodes, edges, attribute deviations, etc.), and the output is an abnormality label and a confidence score (ranging from 0 to 1). The training process includes: During data preprocessing, the graph is converted into a feature vector.

[0046] During the model training process, 80% of the historical data is used for training and 20% for validation, and the parameters are adjusted to optimize the accuracy.

[0047] During the application process, the model outputs the abnormality probability for the new graph, such as 0.9, which means a high probability of abnormality.

[0048] Assume that in the actual operation process map, "instrument analysis" is performed by unauthorized personnel and the time exceeds the standard by 2 hours. The system identifies the attribute anomaly through structural comparison. The machine learning sub-module further confirms that its anomaly probability is 0.95 and marks it as "role violation" and "time limit exceeded".

[0049] The anomaly graph learning module classifies the identified anomaly graphs into the anomaly graph library and performs structural clustering and labeling.

[0050] Anomaly graphs (graphs in the anomaly graph library) are first converted into vector representations using a graph embedding algorithm (such as GraphSAGE) to capture their structural features. Next, K-means clustering is performed based on the Euclidean distance between vectors. The number of clusters, K, can be set to 5 to 10, depending on the number of anomaly graphs. Each type of anomaly graph is assigned a violation label, such as "missing step" or "time limit exceeded."

[0051] Clustering results form typical violation clusters. For example, the system might discover that one type of anomaly graphs all lack the "result review" step, while another type is often anomalies due to "operator role mismatch." These clusters provide reference for subsequent anomaly identification.

[0052] In a further embodiment, the abnormal evolution prediction and warning module uses a graph neural network (GNN) to predict the evolution risk of the current sample process and issue a warning when it deviates from the compliance process.

[0053] In this paper, the anomaly evolution prediction and early warning module employs graph neural network models, such as graph convolutional networks (GCNs), to train the flow chart structures in the anomaly graph library to extract the distribution patterns and evolutionary characteristics of different anomaly types within the graph structure. During training, the model takes as input the adjacency matrix and node features of each anomaly graph. Node features include attributes such as operation type, execution time, operator role, and device type. The GCN model aggregates information from adjacent nodes layer by layer to extract local and global structural features of the process flow. Ultimately, it outputs a graph embedding vector representing the entire graph, which can be used to measure the structural similarity between different flow charts.

[0054] To introduce evolutionary trend factors, the system further combines the time sequence of node appearance and historical state changes to extract time series features, such as key node delay trends, role change frequency, and constraint violation occurrence time. These time features are used together with graph embedding vectors to construct a process evolution trend model, thereby improving the ability to predict process trajectories that are "not yet completed but have deviated."

[0055] In actual operation, the system will construct a phased graph based on the completed operations before the sample process is completed, and perform graph embedding on the graph. The system compares the embedding vector of the phased graph with the embedding vectors of various abnormal process graphs in the abnormal graph library, calculates the similarity, and preferably uses cosine similarity to obtain a matching score ranging from 0 to 1. The higher the similarity value, the closer the current process structure is to a certain type of abnormal graph. When the similarity of a certain abnormal type exceeds the set threshold (such as 0.8), the system believes that the current process has a high risk of deviation and is about to evolve into this abnormal category. It will immediately issue a warning prompt and mark the key risk nodes in the process.

[0056] For example, in a heavy metal detection process, if the current process reaches the "Instrument Analysis" node but lacks the "Quality Control Confirmation" node, and the execution time is later than the standard time window, the system will compare its spectrum with the archived "Test Failure" anomaly spectrum and find that the similarity is 0.85, exceeding the threshold. At this time, the system not only indicates that there is a risk of "Quality Control Missing" in the sample process, but also highlights the potential problem node, guiding management personnel to intervene in time to prevent the further spread of process errors.

[0057] In a further embodiment, the present invention also includes a remote review intelligent assistance module that integrates anomaly detection and risk prediction results to generate a structured review report. This report includes anomaly node annotations (e.g., violations in the "Instrument Analysis" role), corresponding standard process maps, references to historical anomaly maps, and risk prediction information. The report is presented in a visual format, supporting anomaly highlighting and chart display. The report can also be generated using a large language model.

[0058] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A laboratory intelligent management and remote review system, characterized by: include: A standard map library construction module is used to construct standard process maps based on different sample types and detection tasks. Each map consists of multiple nodes, each of which includes standard operation steps and their constraint attributes. The order of operations is represented by directed edges between nodes. The sample flow chart generation module is used to collect the actual operation process of the experimenter and generate the actual operation flow chart; An anomaly detection module is used to compare the structure of the actual operation flow chart with the corresponding standard flow chart to identify the actual operation flow chart with anomalies; The abnormal graph learning module is used to classify the actual operation process graphs with abnormalities into the abnormal graph library, and perform structural clustering and labeling on the actual operation process graphs to form a typical violation set; The abnormal evolution prediction and early warning module is used to learn the structural characteristics and process evolution trends of the actual operation process maps in the abnormal map library based on the graph neural network model, predict the evolution risk of the current sample process, and issue early warning prompts when it deviates from the compliance process.

2. The laboratory intelligent management and remote review system according to claim 1 is characterized in that: The system further comprises: The remote review intelligent assistance module is used to generate a structured review report based on anomaly detection results and evolution risk prediction and provide it to remote review experts.

3. The laboratory intelligent management and remote review system according to claim 2 is characterized in that: The structured review report includes abnormal node annotations, associated standard process maps, and references to historical abnormal maps in the associated abnormal map library.

4. The laboratory intelligent management and remote review system according to claim 1 is characterized in that: The standard map library construction module generates a standard process map by parsing laboratory process procedure documents or historical compliance sample process data. The constraint attributes of the node include: allowed operation time range, specified operation position role, limited equipment type and key quality control conditions; each sample type’s specific detection task corresponds to a standard process map.

5. The laboratory intelligent management and remote review system according to claim 1 is characterized in that: The sample flow chart generation module is used to collect sample operation events based on the laboratory management information system log, video behavior recognition results and equipment operation records, and generate an actual operation flow chart with the same structure as the corresponding standard flow chart based on the sample operation events.

6. The laboratory intelligent management and remote review system according to claim 1 is characterized in that: The anomaly detection module uses a structural comparison algorithm based on graph edit distance to perform node mapping and path alignment on the actual operation process map and the corresponding standard process map, and identifies the actual operation process map as an anomaly when any of the following conditions are met: a) The graph edit distance is greater than the preset threshold; b) Nodes are missing, repeated, or inverted during the mapping process; c) The timestamp of the mapped node exceeds the time window allowed by the standard node; d) The operator role or device type of the mapped node does not conform to the standard node constraint attributes; Among them, the mapped nodes are nodes in the actual operation process map, and the standard nodes are nodes in the standard process map.

7. The laboratory intelligent management and remote review system according to claim 1 is characterized in that: The anomaly detection module further includes a machine learning identification submodule, which uses a trained classification model to determine whether the actual operation process map is an anomaly map and outputs a corresponding anomaly type label and confidence score.

8. The laboratory intelligent management and remote review system according to claim 7, characterized in that: The classification model is trained using a supervised learning method by using multiple actual operation process maps collected historically as training samples, marking whether they are abnormal and the type of abnormality.

9. The laboratory intelligent management and remote review system according to claim 1 is characterized in that: The abnormal graph learning module performs graph structure embedding and similarity clustering operations on the actual operation process graphs identified as abnormal, uses a graph embedding algorithm to convert each graph into a vector representation, clusters abnormal process patterns based on vector distance, and automatically generates violation labels for each type of abnormal graph to form a typical violation set.

10. The laboratory intelligent management and remote review system according to claim 1, characterized in that: The abnormal evolution prediction and early warning module uses a graph neural network model to perform graph embedding training on the graph structure patterns in the abnormal graph library, and combines the time series node evolution characteristics to construct a process evolution trend model. Based on the phased graph construction of the current sample process, it calculates the similarity change curve with the abnormal graph. When its evolution path is predicted to approach any abnormal graph category and reaches the risk threshold, it automatically generates an early warning prompt and identifies the potential risk node.

Images