Clinical label labeling method, system and equipment based on standardized time sequence and medium

Abstract

The invention discloses a clinical tag labeling method, system and equipment based on a standardized time sequence and a medium, and relates to the technical field of clinical event fusion and dynamic tag generation, and the specific steps are as follows: obtaining clinical event data of a patient from different clinical business systems in real time, dividing cycle attribution for the clinical event data, and determining the clinical event data; and establishing a dynamic weight calculation model to allocate dynamic weight values, establishing a periodic clinical state inference model by adopting a convolutional neural network and training the periodic clinical state inference model, inputting weighted summary clinical feature vectors in a current patient period according to the trained periodic clinical state inference model, and outputting clinical tags corresponding to patients. According to the invention, clinical label labeling based on a standardized time sequence is realized, multi-source clinical event data from different clinical business systems can be collected in real time, event weights are reasonably distributed through a dynamic weight calculation model, a periodic clinical state is deduced by using a convolutional neural network, and clinical labels of patients are automatically generated.

Description

Technical Field

[0001] This invention relates to the field of clinical event fusion and dynamic label generation technology, and in particular to a clinical labeling method, system, device and medium based on standardized time sequence. Background Technology

[0002] With the advancement of smart hospital construction, medical institutions have accumulated a large amount of multi-dimensional clinical data, including electronic medical records, vital signs, medical order execution, and laboratory tests. However, this clinical data is scattered across different scheduling systems, recorded in various ways, and lacks a unified temporal semantic framework, making it difficult to support high-quality research and modeling tasks. Currently, clinical labeling methods mostly rely on manual rules or fixed window aggregation, using static time intervals (such as hospital stay days, medical order execution time periods, and examination time points) to coarsely slice clinical events and generate labels based on whether the clinical event appears within the window. However, this approach has significant limitations: existing labeling systems are mostly built based on fixed time windows or recording time points, lacking a standardized periodic structure, and failing to unify the temporal expression of different types of clinical events, resulting in inconsistent temporal semantics for the same clinical state across different data sources. Existing labeling systems typically employ coarse-grained window alignment strategies, such as dividing by day, hour, or medical order execution cycle, failing to achieve fine-grained alignment of clinical events across systems. Existing labeling systems only perform simple statistics on events, lacking the ability to dynamically adjust weights based on time distance, and failing to reflect the temporal correlation between events and the inferred cycle. The time sensitivity of labels in reflecting changes in patient condition leads to a lag in label expression. If a patient's vital signs deteriorate, laboratory indicators become abnormal, or acute changes in condition are not promptly expressed through labels, it can result in missing the optimal intervention window and causing serious consequences. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by this invention is: how to construct a clinical labeling system with standardized time periods, support for fine-grained time alignment across systems, and dynamic weight adjustment in the case of inconsistent time formats, scattered event records, and lack of temporal semantic consistency in multi-source heterogeneous clinical data, so that changes in patients' clinical status can be accurately expressed and the problems of misjudgment of status and delay in treatment caused by the lag of clinical labels can be avoided.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a clinical labeling method based on standardized time series, comprising, Connect to clinical business systems to obtain real-time clinical event data of patients from different clinical business systems; Normalization was used to unify the time of clinical event data and classify the period to obtain clinical event data within the period. Based on the clinical event data within the period, a dynamic weight calculation model is established to assign dynamic weight values ​​to the clinical event data within the period, and a weighted summary clinical feature vector within the period is obtained. Based on the weighted summary of clinical feature vectors within the cycle, a convolutional neural network is used to establish and train a cycle-based clinical status inference model. Based on the trained periodic clinical status inference model, input the weighted summary clinical feature vector of the current patient within the period, and output the corresponding clinical label of the patient.

[0006] As a preferred embodiment of the clinical labeling method based on standardized time series described in this invention, the clinical event data is normalized and divided into periods to obtain clinical event data within a period, including: The raw time of clinical event data is extracted and converted into a uniform time format using normalization. A time-series standardized model was established based on normalized clinical event data. The clinical event data is classified into periods based on the time-series standardization model to obtain the clinical event data within each period.

[0007] This invention transforms the time fields of multi-source clinical events into a unified format through normalization processing, enabling clinical events recorded by different clinical scheduling systems to be expressed under a unified time rule, thus solving the problem of event misalignment caused by differences in time formats. By establishing a time-series standardization model, each clinical event is divided into a corresponding standardized time period, avoiding the period chaos caused by inconsistent original time granularity.

[0008] As a preferred embodiment of the standardized time-series-based clinical labeling method described in this invention, the method involves: establishing a dynamic weight calculation model based on clinical event data within a period to assign dynamic weight values ​​to the clinical event data within the period, thereby obtaining a weighted summary clinical feature vector within the period, including: A dynamic weighting calculation model is established based on clinical event data within the cycle; Initialize the dynamic weight calculation model; Based on the initialized dynamic weight calculation model, dynamic weight values ​​are assigned to the clinical event data within the period to obtain weighted clinical event data within the period. A weighted fusion method was used to process the weighted clinical event data within the period to obtain a weighted summary clinical feature vector within the period.

[0009] This invention establishes a dynamic weight calculation model that generates corresponding dynamic weights for clinical event data based on the temporal relationship between the event's occurrence time and the current period, thus reflecting temporal proximity in the event processing. Events closer to the current analysis period have a greater influence in feature construction, while events further away have a weaker influence. By weighted fusion of the weighted clinical event data, a weighted summary clinical feature vector within the period is formed, enabling the representation of multi-source and multi-type events within a unified temporal framework. This avoids feature shift problems caused by inconsistent event weights or unrecognized differences in event location.

[0010] As a preferred embodiment of the clinical labeling method based on standardized time series described in this invention, the method includes: establishing and training a periodic clinical state inference model using a convolutional neural network based on weighted summaries of clinical feature vectors within a period, including: A convolutional neural network was used to establish a periodic clinical status inference model. Historical data from disease centers were collected, and a training set was formed by weighted summaries of clinical feature vectors within a period and their corresponding real clinical labels. The periodic clinical state inference model is trained based on the training set to obtain the trained periodic clinical state inference model.

[0011] This invention employs a convolutional neural network to establish a periodic clinical state inference model and constructs a training set using weighted summaries of clinical feature vectors within a period and their corresponding real clinical labels. This enables the periodic clinical state inference model to learn representative local patterns and temporally related structures. Based on the temporal proximity, event strength relationships, and combined features among multiple event types reflected in the weighted feature vectors, the periodic clinical state inference model can generate clinical labels that correspond to the patient's true clinical state.

[0012] This invention provides a clinical labeling system based on standardized time sequence.

[0013] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a clinical labeling system based on standardized time series, comprising: an acquisition module, a period division module, a weight allocation module, a model building module, and an output module; The acquisition module is connected to the clinical business system to acquire clinical event data of patients from different clinical business systems in real time. The cycle division module uses normalization processing to unify the time of clinical event data and divide it into cycle categories to obtain clinical event data within a cycle. The weight allocation module establishes a dynamic weight calculation model based on the clinical event data within the period to allocate dynamic weight values ​​to the clinical event data within the period, thereby obtaining a weighted summary clinical feature vector within the period. The model building module is based on a weighted summary of clinical feature vectors within a cycle, and uses a convolutional neural network to build and train a cycle-based clinical state inference model. The output module is based on the trained periodic clinical status inference model. It takes the weighted summary clinical feature vector of the current patient period as input and outputs the corresponding clinical label of the patient.

[0014] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the clinical labeling method based on standardized time sequence.

[0015] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the clinical labeling method based on standardized time series.

[0016] The beneficial effects of this invention are as follows: By establishing a time-series standardization model, this invention correctly assigns clinical events to standardized periods. Combined with an exponentially decaying weight function, it dynamically adjusts the influence intensity of historical events, highlighting the clinical value of recent data. By constructing a weighted summary clinical feature vector within a period and inputting it into a period state inference model for state inference, interpretable clinical labels are generated. By unifying the label time point assignment rules, the consistency of label time is ensured. Ultimately, this invention integrates time standardization, tolerance calibration, dynamic weighting, and intelligent inference to output structured clinical labels, thereby improving the research value and practical performance of clinical data. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall process of a clinical labeling method based on standardized time sequence according to an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0020] Example 1, referring to Figure 1 As one embodiment of the present invention, this embodiment provides a clinical labeling method based on standardized time series, including: It should be noted that with the continuous advancement of smart hospital construction and the widespread application of multi-source medical information systems, clinical data exhibits significant multi-source heterogeneity. Inconsistent time recording formats, varying granularities, and mixed time zones across different systems make it difficult to accurately align clinical events in the time dimension. Existing clinical label generation methods largely rely on manual backtracking or simple rules, commonly suffering from issues such as missing time period definitions, cross-system event deviations leading to incorrect period assignments, and imbalanced event weights that fail to reflect differences in clinical importance. This results in poor temporal consistency of generated labels and an inability to dynamically reflect the evolving trends of patient conditions, easily causing delays and misjudgments in clinical assessment.

[0021] Therefore, to address the issues of existing clinical labeling methods lacking standardized periodic structures and dynamic weight adjustment capabilities, the following steps (S1-S5) are implemented: real-time acquisition of clinical event data from different clinical business systems; normalization processing and period classification to obtain clinical event data within a period, thus solving the problem of the lack of standardized periodic structures in existing methods; establishment of a dynamic weight calculation model to obtain a weighted summary clinical feature vector within a period, thus solving the problem of the lack of dynamic weight adjustment capabilities in existing methods; and establishment and training of a periodic clinical state inference model using a convolutional neural network, inputting the current patient's weighted summary clinical feature vector within a period, and outputting the patient's corresponding clinical label.

[0022] S1: Connects to the clinical business system to obtain real-time clinical event data of patients from different clinical business systems; S2: Normalization is used to unify the time of clinical event data and classify the period to obtain clinical event data within the period; S3: Based on the clinical event data within the period, establish a dynamic weight calculation model to assign dynamic weight values ​​to the clinical event data within the period, and obtain the weighted summary clinical feature vector within the period; S4: Based on the weighted summary of clinical feature vectors within the cycle, a convolutional neural network is used to establish and train a cycle-based clinical status inference model. S5: Based on the trained periodic clinical status inference model, input the weighted summary clinical feature vector of the current patient within the period, and output the corresponding clinical label of the patient.

[0023] Example 2, an embodiment of the present invention, provides a clinical labeling method based on standardized time sequence, based on the previous embodiment, including: In step S1, the system connects to the clinical business system to obtain real-time clinical event data of patients from different clinical business systems, including the following steps: The system accesses raw data from various systems, including Hospital Information System (HIS) (containing patient basic information, medical orders, and billing data), Electronic Medical Record System (EMR) (containing structured and unstructured data such as chief complaints, present medical history, and surgical records), Laboratory Information System (LIS) (containing results and timestamps of blood routine, biochemistry, and other laboratory tests), Image Storage and Transmission System (PACS) (containing DICOM format CT and MRI images and reports), and monitoring equipment (containing real-time vital signs data such as heart rate, blood oxygen, and respiratory rate), via HL7 / FHIR protocol, RESTful API interface, and direct database connection. It also supports OAuth2.0 authentication and HTTPS encrypted transmission to ensure the security and compliance of data access.

[0024] In step S2, normalization processing is used to unify the time of clinical event data and classify it into period categories to obtain clinical event data within a period, including the following steps B1-B3: B1: Extract the raw time of clinical event data and convert it into a unified time format using normalization.

[0025] The raw timestamps of extracted clinical event data were normalized to a uniform UTC time.

[0026] For example, a patient had their temperature measured at 15:30 (Beijing time) on October 21, 2025. 15:30 is the raw time of the event, which includes time zone information (CST, i.e., UTC+8). We need to convert this raw time to standard Coordinated Universal Time (UTC). Beijing time is 8 hours ahead of UTC, so subtracting 8 hours gives us the UTC time: 2025-10-21 07:30. 07:30 is the normalized UTC time.

[0027] B2: Establish a time-series standardized model based on normalized clinical event data.

[0028] B3: Based on the time-series standardization model, the period of clinical event data is divided to obtain the clinical event data within the period.

[0029] In this embodiment of the application, the specific steps for establishing the time series normalization model in step B2 are as follows: The time tolerance windows for each clinical event data type are established, as shown in Table 1: Table 1 Time Tolerance Table for Clinical Event Data

[0030] Based on normalized clinical event data, an anchor time T_anchor and a cycle length ΔT_cycle are defined to form a standardized cycle [T_anchor, T_anchor + ΔT_cycle]. The normalized UTC time is compared with the extended interval [T_anchor - Δt, T_anchor + ΔT_cycle + Δt]. If it is within the range, it is assigned to that cycle; otherwise, it is marked as abnormal.

[0031] In an optional implementation, the time-series normalization model established in step B2 can also employ a sliding time window method. A time window length is set, and the window slides along the time axis at a set step size, which can be equal to or a fraction of the window length. The UTC-normalized event times are matched with each sliding window interval. If an event time falls within a certain sliding window range, it is assigned to that window; if an event exceeds any window range, it is marked as an anomaly or noise.

[0032] In another optional implementation, the time series standardization model established in step B2 can also employ a timestamp quantization method. A uniform time quantization precision is set, mapping the normalized timestamps to quantization units, and the obtained quantization index is used as the event's attribution period. If the quantization result exceeds a reasonable range (e.g., consecutive missing values, abnormal spans), it is marked as an abnormal event. Events are then batch-aligned according to the quantization number to form a standardized time series.

[0033] It should be noted that by setting time tolerance windows for various types of clinical event data and constructing standardized periods using anchor times and period lengths, events from different sources and with different recording frequencies can be time-assigned based on the same period model. Using extended intervals to match the normalized UTC times ensures that events near period boundaries can be identified within the tolerance range, avoiding period misclassification due to recording delays or acquisition errors. Events outside the extended interval are marked as abnormal to prevent erroneous timestamps from interfering with the expression of period features.

[0034] In step S3, based on the clinical event data within the period, a dynamic weight calculation model is established to assign dynamic weight values ​​to the clinical event data within the period, resulting in a weighted summary clinical feature vector within the period, including the following steps C1-C4: C1: Establish a dynamic weight calculation model based on clinical event data within the cycle.

[0035] C2: Initialize the dynamic weight calculation model.

[0036] Initialize the decay coefficient of the dynamic weight calculation model and benchmark weights The attenuation coefficient is set by clinical experts based on the characteristics of the disease; for example, a larger value is used for rapidly changing sepsis, while a smaller value is used for stable chronic disease indicators. The baseline weight is pre-configured based on the clinical importance of the label. During runtime, the system calculates the dynamic effective weight of the clinical event data through exponential decay by combining the difference between the time of the labeled event and the current analysis time with the attenuation coefficient and the baseline weight, thus obtaining weighted clinical event data within the period.

[0037] C3: Based on the initialized dynamic weight calculation model, assign dynamic weight values ​​to the clinical event data within the period to obtain weighted clinical event data within the period.

[0038] C4: A weighted fusion method is used to process the weighted clinical event data within the period to obtain a weighted summary clinical feature vector within the period.

[0039] Based on all weighted clinical event data within the current period, the data indicators of the same type of clinical event are weighted and fused.

[0040] For example, if body temperature is recorded multiple times within a cycle, the system will not simply count the number of fevers, but will instead calculate a weighted average of each temperature reading and its corresponding weight to generate a continuous value representing the overall fever intensity of that cycle, such as "weighted average body temperature: 38.2°C". After this processing, all indicators together constitute a structured numerical feature vector, in the format of: [weighted average body temperature, weighted average heart rate, highest weighted systolic blood pressure, weighted average white blood cell count, ...], comprehensively depicting the patient's physiological state within that cycle.

[0041] All clinical events (symptoms, signs, laboratory indicators) within the current period are dynamically weighted using a time-decay model and then fused into a standardized weighted summary clinical feature vector for the current period, expressed as: … , in, This is a weighted summary of clinical feature vectors within the period. For weighted average body temperature, For weighted average heart rate, The highest weighted systolic blood pressure, This is a weighted average white blood cell count; all other similar indicators are processed in the same way. It is a continuous representation of the patient's physiological state within the current standardized cycle.

[0042] In this embodiment of the application, the specific steps for establishing the dynamic weight calculation model in step C1 are as follows: The dynamic weight calculation model is expressed as follows: , in, This represents the weight of the clinical event at time t, where t is the time when the clinical event occurs. As the baseline weight for this event, The decay coefficient controls the rate at which the weight decreases over time. For the current analysis time, The time difference between the time the event occurred and the current analysis time. It is an exponential function. Indicates a standardized time period.

[0043] In an optional implementation, the dynamic weight calculation model established in step C1 can also employ a weight allocation method based on time window hierarchy. The time axis is divided into intervals according to the period length, with the period divided into near-end windows, middle windows, and far-end windows. The occurrence time of events is mapped to the corresponding time windows, and events are assigned weights corresponding to the window level. Events closer to the center window receive higher weight levels, while events farther from the center receive lower weight levels.

[0044] In another optional implementation, the dynamic weight calculation model established in step C1 can also employ a time-based clustering weight allocation method. This method clusters events within a period based on their occurrence times, grouping events with similar times into the same cluster center. Based on the positional relationship between the cluster center and the current period's reference time, a weight value is assigned to each cluster, giving larger weights to clusters closer to the reference time. This cluster weight is then applied to all events within each cluster.

[0045] It should be noted that the design of the dynamic weight calculation model in this invention is based on three core principles, aligning with the natural progression of disease and the timeliness of intervention in clinical medicine. For example, recent indicators of acute events are far more important than outdated indicators, while chronic disease management focuses on recent data to assess current control status; these principles are also implicit in many clinical guidelines. Mathematically, the classic exponential decay function is employed, ensuring that the weights decrease monotonically and smoothly over time. Furthermore, the parameters of the dynamic weight calculation model can be validated and optimized using real-world data, ensuring consistency between the model and actual clinical data.

[0046] In step S4, based on the weighted summary of clinical feature vectors within the cycle, a convolutional neural network is used to establish and train a cycle-based clinical state inference model, including the following steps D1-D3: D1: A convolutional neural network is used to establish a periodic clinical status inference model.

[0047] D2: Collect historical data from disease centers, and form a training set consisting of weighted summaries of clinical feature vectors within the period and corresponding real clinical labels.

[0048] The system collects historical data from multiple disease centers and constructs a training set consisting of clinical feature vectors weighted and aggregated over a standardized period and their corresponding real clinical labels.

[0049] D3: Train the periodic clinical state inference model based on the training set to obtain the trained periodic clinical state inference model.

[0050] During the training phase, the periodic clinical state inference model learns nonlinear patterns from the weighted aggregated clinical feature vectors within a period and outputs the predicted probability distribution of each clinical state category.

[0051] In this embodiment of the application, the specific steps for establishing the periodic clinical state inference model in step S4 are as follows: A lightweight CNN model is used to establish the periodic clinical state inference model, and the expression is: , , in, This represents the class probability distribution of input sample i, where i is the sample index. Lightweight CNN model, Let C be the weighted summary clinical feature vector of sample i within the period, and let C be the total number of state categories. Indicates that sample i belongs to the first... The predicted probability of the class state category.

[0052] In an optional implementation, step S4, establishing the periodic clinical state inference model, can also employ a periodic state inference method based on recurrent neural networks. The weighted clinical feature vectors within a period are input into the recurrent neural network structure in chronological order. The recurrent neural network sequentially reads the features and establishes dependencies between adjacent time points. The memory units of the recurrent structure retain key temporal context information, enabling the model to determine the clinical state corresponding to the current period based on historical feature sequences. Finally, the output layer generates the periodic-level state classification results.

[0053] In another optional implementation, the periodic clinical state inference model established in step S4 can also employ an attention-based periodic state inference method. This method encodes the weighted clinical feature vectors within the period into temporal positions, ensuring each feature possesses identifiable temporal position information. The correlation between features at different temporal positions is calculated using a self-attention mechanism, enabling the model to automatically determine the contribution of each feature to the periodic state based on this correlation. Finally, the attention-aggregated features are fed into a feedforward network to generate the periodic clinical state prediction result.

[0054] It should be noted that by employing a convolutional neural network to establish a periodic clinical state inference model, discriminative feature representations can be extracted based on local correlation patterns and temporal proximity relationships contained in the weighted aggregated clinical feature vectors within a period. A sliding process is applied to the feature sequence to capture different event combinations and their relative positions within the period, thus forming feature patterns corresponding to different clinical states. Training the model based on real clinical labels eliminates the reliance on manual rules or single feature statistical methods in the periodic state determination process.

[0055] In step S5, based on the trained periodic clinical status inference model, the weighted summary clinical feature vector of the current patient within the current period is input, and the corresponding clinical label of the patient is output, including the following steps E1-E3: E1: Based on the trained periodic clinical status inference model, input the weighted summary clinical feature vector of the current patient within the current period.

[0056] E2: The periodic clinical state inference model outputs the predicted probability distribution of each clinical state category.

[0057] It should be noted that the periodic clinical status inference model can flexibly configure threshold and weight strategies, and the final output of the periodic clinical status inference model is a clear periodic clinical label with confidence evidence.

[0058] For example, suppose a patient's weighted summary clinical feature vector is collected within a standardized period, as shown in the following expression: , Input the trained periodic clinical state inference model, the expression is: , The output is: [(Sepsis, 0.85), (Heart Failure, 0.10), (No Abnormalities, 0.05)], which means that the cycle clinical state inference model judges that the cycle is most likely in a sepsis state with a confidence level of 85%.

[0059] E3: Based on the predicted probability distribution of each clinical state category in the output, obtain the corresponding clinical label for the patient.

[0060] In this embodiment of the application, the specific steps for outputting the clinical label corresponding to the patient in step S5 are as follows: Clinical labels are generated using a standardized medical data structure based on the predicted probability distribution of each clinical state category.

[0061] It should be noted that the clinical labels output by this system are encapsulated using the Observation resource format in the FHIR standard. Each label corresponds to an Observation instance, explicitly recording the clinical state represented by the label. Standard terminology such as SNOMED CT is used for encoding, and the corresponding standardized time period is represented by the effectivePeriod field. The confirmation status of the label, such as final or preliminary, and the decision confidence level are attached to the Observation as components.

[0062] It should also be noted that Observation resources are associated with specific patients and optionally with medical or disease episode information. All fields conform to the data types and encoding schemes defined by the FHIR specification, ensuring that tags can be directly exchanged, parsed, and used across different healthcare systems without additional conversion.

[0063] Define unified label time point assignment rules based on the generated clinical labels; For clinical events occurring within a specific time period, the generated clinical tags must be linked to a specific time location to ensure data comparability across patients and institutions. This invention defines a unified rule for tag time point assignment based on the different nature of the events. When an examination or observation occurs continuously within a standardized time period, the clinical tag triggered by the examination result should be assigned to the standard period in which the event was first confirmed, and the time anchor point of that period should be used as the starting point for the tag's effective time.

[0064] Based on the unified label time point attribution rules, clinical labels are aligned with bed events.

[0065] According to the unified tag time point assignment rules, if the start time of an event falls within a certain standard period, the tag generated by the event belongs to this period and its effective time is the start point of the period. The end time of the tag is handled flexibly depending on the specific situation. If the known state lasts for multiple periods, the complete time interval is marked. If it is only an instantaneous judgment, its duration is assumed to be the same as the current period, or it is dynamically extended by the system based on subsequent observations.

[0066] In an optional implementation, the clinical tags output in step S5 can also be generated using the HL7 general message structure. The core elements contained in the clinical tags are mapped to standardized fields supported by the message structure. The tag content is encoded and hierarchically organized according to preset data organization rules to adapt to the data exchange requirements between different systems. The structured tag data is encapsulated into standardized message objects that can be parsed by the medical information system to support cross-system tag sharing, querying, and storage.

[0067] In another optional implementation, the clinical tags output in step S5 can also be documented using the Clinical Document Architecture (CDA) format. Based on the status type corresponding to each clinical tag, a clinical document fragment containing an entry-level structure is generated for each tag. The tag's attribute information is mapped to the document's metadata and main content. The content structure, field naming, and hierarchical relationships are standardized according to document specifications, ensuring that the tag documents meet the requirements for medical document exchange and archiving. Finally, a standardized document structure that can be recognized and parsed by the electronic medical record system is generated, enabling the universal storage and retrieval of tags in the clinical business system.

[0068] It should be noted that by encapsulating the generated clinical tags using the Observation resource format in the FHIR standard, a unified data structure can be used to express the status category, time attribute, and associated clinical events of the clinical tags, giving them clear semantic boundaries and field meanings. This encapsulation method also ensures that clinical tags maintain a consistent structural form across different clinical business systems, facilitating parsing, storage, and retrieval by multiple systems. Furthermore, structured encapsulation allows tag content to establish referential relationships with other medical information resources.

[0069] In summary, this invention achieves a unified representation of multi-source clinical event data from different clinical business systems within a standardized time-series framework by constructing a unified time normalization mechanism. By establishing a dynamic weight calculation model, it can generate time-sensitive weighted features. Through the construction of a periodic clinical state inference model, it realizes automated inference and label generation of clinical states. By employing a structured label encapsulation method, the generated clinical labels possess a unified semantic format and shareable characteristics, enabling direct parsing and invocation across multiple systems.

[0070] Example 3 is an embodiment of the present invention, which provides a clinical labeling system based on standardized time series, including: a data acquisition module, a period division module, a weight allocation module, a model building module, and an output module; The acquisition module is connected to the clinical business system to acquire clinical event data of patients from different clinical business systems in real time. The cycle division module uses normalization processing to unify the time of clinical event data and divide it into cycle categories to obtain clinical event data within a cycle. The weight allocation module establishes a dynamic weight calculation model based on the clinical event data within the period to allocate dynamic weight values ​​to the clinical event data within the period, thereby obtaining a weighted summary clinical feature vector within the period. The model building module is based on a weighted summary of clinical feature vectors within a cycle, and uses a convolutional neural network to build and train a cycle-based clinical state inference model. The output module is based on the trained periodic clinical status inference model. It takes the weighted summary clinical feature vector of the current patient period as input and outputs the corresponding clinical label of the patient.

[0071] This embodiment also provides an electronic device applicable to a clinical labeling method based on standardized time sequence, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the clinical labeling method based on standardized time sequence proposed in the above embodiment.

[0072] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a clinical labeling method based on standardized time sequence as proposed in the above embodiments.

[0073] The storage medium proposed in this embodiment and the clinical labeling method based on standardized time sequence proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0074] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A clinical labeling method based on standardized time series, characterized in that: include, Connect to clinical business systems to obtain real-time clinical event data of patients from different clinical business systems; Normalization was used to unify the time of clinical event data and classify the period to obtain clinical event data within the period. Based on the clinical event data within the period, a dynamic weight calculation model is established to assign dynamic weight values ​​to the clinical event data within the period, and a weighted summary clinical feature vector within the period is obtained. Based on the weighted summary of clinical feature vectors within the cycle, a convolutional neural network is used to establish and train a cycle-based clinical status inference model. Based on the trained periodic clinical status inference model, input the weighted summary clinical feature vector of the current patient within the period, and output the corresponding clinical label of the patient.

2. The clinical labeling method based on standardized time sequence as described in claim 1, characterized in that: The normalization process is used to unify the time of clinical event data and classify it into periods, resulting in clinical event data within a period, including: The raw time of clinical event data is extracted and converted into a uniform time format using normalization. A time-series standardized model was established based on normalized clinical event data. The clinical event data is classified into periods based on the time-series standardization model to obtain the clinical event data within each period.

3. The clinical labeling method based on standardized time sequence as described in claim 2, characterized in that: The process involves establishing a dynamic weighting calculation model based on clinical event data within a given period to assign dynamic weight values ​​to the clinical event data within that period, resulting in a weighted aggregated clinical feature vector for that period, including: A dynamic weighting calculation model is established based on clinical event data within the cycle; Initialize the dynamic weight calculation model; Based on the initialized dynamic weight calculation model, dynamic weight values ​​are assigned to the clinical event data within the period to obtain weighted clinical event data within the period. A weighted fusion method was used to process the weighted clinical event data within the period to obtain a weighted summary clinical feature vector within the period.

4. The clinical labeling method based on standardized time sequence as described in claim 3, characterized in that: The step of establishing and training a periodic clinical state inference model using a convolutional neural network based on a weighted summary of clinical feature vectors within a period includes: A convolutional neural network was used to establish a periodic clinical status inference model. Historical data from disease centers were collected, and a training set was formed by weighted summaries of clinical feature vectors within a period and their corresponding real clinical labels. The periodic clinical state inference model is trained based on the training set to obtain the trained periodic clinical state inference model.

5. The clinical labeling method based on standardized time sequence as described in claim 4, characterized in that: The process of inferring the periodic clinical status based on the trained model involves inputting a weighted summary clinical feature vector for the current patient's period and outputting the patient's corresponding clinical label, including: Based on the trained periodic clinical status inference model, input the weighted summary clinical feature vector of the current patient's period; The periodic clinical status inference model outputs the predicted probability distribution of each clinical status category; Based on the predicted probability distribution of each clinical state category, the corresponding clinical label for the patient is obtained.

6. The clinical labeling method based on standardized time sequence as described in claim 5, characterized in that: The process of obtaining the patient's corresponding clinical label based on the predicted probability distribution of each clinical state category includes: Based on the predicted probability distribution of each clinical state category in the output, clinical labels are generated using a standardized medical data structure. Define unified label time point assignment rules based on the generated clinical labels; Based on the unified label time point attribution rules, clinical labels are aligned with bed events.

7. A clinical labeling method based on standardized time sequence as described in claim 6, characterized in that: The step of establishing a dynamic weight calculation model based on clinical event data within the cycle includes: The dynamic weight calculation model is expressed as follows: , in, This represents the weight of the clinical event at time t, where t is the time when the clinical event occurs. As the baseline weight for this event, The decay coefficient controls the rate at which the weight decreases over time. For the current analysis time, The time difference between the time the event occurred and the current analysis time. It is an exponential function. Indicates a standardized time period.

8. A clinical labeling system based on standardized time series, employing a clinical labeling method based on standardized time series as described in any one of claims 1-7, characterized in that, include: The system includes a data acquisition module, a period division module, a weight allocation module, a model building module, and an output module. The acquisition module is connected to the clinical business system to acquire clinical event data of patients from different clinical business systems in real time. The cycle division module uses normalization processing to unify the time of clinical event data and divide it into cycle categories to obtain clinical event data within a cycle. The weight allocation module establishes a dynamic weight calculation model based on the clinical event data within the period to allocate dynamic weight values ​​to the clinical event data within the period, thereby obtaining a weighted summary clinical feature vector within the period. The model building module is based on a weighted summary of clinical feature vectors within a cycle, and uses a convolutional neural network to build and train a cycle-based clinical state inference model. The output module is based on the trained periodic clinical status inference model. It takes the weighted summary clinical feature vector of the current patient period as input and outputs the corresponding clinical label of the patient.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the clinical labeling method based on standardized time sequence according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the clinical labeling method based on standardized time sequence according to any one of claims 1 to 7.

Images