People injury patient visiting management method and system

Abstract

The invention provides an injured patient visiting management method and system, and relates to the technical field of hospitalized patient visiting management, and the method comprises the steps: obtaining visiting request information containing the number of reservation persons and reservation time, and a plurality of historical images of a patient in the visiting time; establishing a neural network model and inputting historical image output feature vectors in the same time period, calculating similarity: calculating the similarity of adjacent feature vectors according to timestamp tags, judging whether the similarity is greater than a preset threshold, if so, sending a reservation failure signal, and if not, sending a reservation failure signal; according to the method and the device, the visiting can be started from the actual condition of the patient during visiting, so that the visiting arrangement is more rigorous.

Description

Technical Field

[0001] The present application relates to the technical field of inpatient visitation management, and in particular to a method and system for managing visitation of injured patients. Background Art

[0002] In the modern hospital management system, patient visitation management is not only related to the orderly operation of the medical environment, but also an important line of defense for protecting the health rights and interests of patients. The scientific nature and effectiveness of its management model directly affect the quality of medical services and the patient's recovery process. At present, most hospitals still use the traditional visitation appointment management method, which mainly covers two forms: manual registration and simple online appointment. In the manual registration scenario, visitors often need to line up at the front desk of the inpatient department and verbally tell the staff one by one the number of appointments, desired visiting time, and other information. The staff then uses a paper registration book to record each item, and a lot of manpower is required to verify and count the information later. The process is cumbersome and inefficient. Although the simple online appointment system uses an electronic form built on the hospital's official website or official account, the system only has basic data storage functions and cannot perform in-depth processing and intelligent analysis of massive appointment information, making it difficult to meet the complex and changing visitation management needs.

[0003] These two traditional management approaches have exposed numerous drawbacks. Regarding overall patient population control, the lack of real-time dynamic monitoring and intelligent control mechanisms often leads to crowds of visitors in wards during peak hours. These crowds not only occupy limited medical space and hinder the normal flow of medical staff, but also easily lead to poor air circulation, significantly increasing the risk of cross-infection within the hospital.

[0004] The Chinese invention patent application publication number CN116487008A provides a hospital inpatient visiting system and equipment. The patent stipulates that each patient can have a maximum of several visitors at the same time, depending on the regulations of each hospital.

[0005] However, the above patent only limits the number of visiting places for each patient according to hospital regulations, and does not consider whether the patient's condition allows family members to visit, resulting in loose visiting management. Summary of the Invention

[0006] In order to make visiting arrangements more rigorous based on the actual conditions of the patients during visits, this application provides a method and system for managing visits to injured patients.

[0007] In the first aspect, the present application provides a method for managing visits to injured patients, which adopts the following technical solutions: A method for managing visits to injured patients, comprising the following steps: Data collection: obtaining visit request information and multiple historical images of the patient during the visit time, wherein the visit request information includes the appointment time; Feature extraction: A neural network model is established for feature extraction. Historical images in the same period as the appointment time are sequentially input into the neural network model, and the feature vector corresponding to each historical image is output; Calculate similarity: Use the timestamp of the historical image as the time label of the corresponding feature vector, and calculate the similarity between the feature vectors of adjacent times based on the time label; Similarity judgment: judge whether the similarities are greater than the preset similarity threshold. If so, send a reservation failure signal; if not, execute the step of generating a reservation code; Generate appointment code: Generate a visit identification code based on the visit request information and send the visit identification code to the appointment person.

[0008] This application first collects historical images of the appointment time and visiting time in the visit request, and then uses a neural network model to extract features. It can automatically learn and extract effective features from historical images, input historical images in the same period as the appointment time into the model, extract feature vectors representing the patient's status, and then use the timestamp of the historical image as the time label of the feature vector. The similarity between the feature vectors of adjacent times is calculated based on the time label. If the similarity is greater than the preset similarity threshold, it means that the patient is not suitable to be disturbed. Sending an appointment failure signal can reduce unnecessary visits and protect the patient's rest and treatment environment. On the contrary, a visit identification code is generated based on the visit request information and sent to the appointment person end, providing a convenient identity authentication method for the appointment person, so that only people with an appointment can visit, improving the security and standardization of visit management.

[0009] Optionally, the method further comprises: constructing a GNN model using feature vectors corresponding to historical images as graph nodes and a time sequence of collecting historical images as directed edges; Feature update: Set the weight of each directed edge in the GNN model according to the interval of historical image acquisition time, input the feature vector into the GNN model, output the updated feature vector, and use the updated feature vector as the new feature vector.

[0010] This application constructs a graph neural network (GNN) model using the chronological order of historical image acquisition as directed edges to effectively capture information about changes in patient status over time. By constructing a GNN model, previously independent feature vectors are connected through a graph structure. Subsequently, this application sets the weight of each directed edge in the GNN model based on the time interval between historical image acquisitions. Images with smaller time intervals are more similar and coherent, so assigning larger weights allows the GNN model to prioritize this closely related information when updating features. Features between images with larger time intervals are likely to vary more, so smaller weights reduce reliance on information from more distant time points. Subsequently, this application inputs the original feature vector into the GNN model and outputs an updated feature vector to optimize the feature vector. The updated feature vector better integrates information from different time points, enhancing the feature representation capability and facilitating more accurate results in subsequent steps such as calculating similarity and determining patient status. For example, when determining whether a patient is suitable for visitation, a more representative feature vector can more accurately reflect whether the patient's current condition is stable or suitable for interruption. The GNN model's structure and computational approach enable it to process complex graph-structured data. For complex data such as patient historical image features, which are temporally correlated, the GNN model can better mine potential relationships and patterns. Compared to simple feature extraction methods, it can more effectively handle the interactions between features, thereby more accurately depicting the dynamic changes in a patient's condition.

[0011] Optionally, the method further includes: Segmentation and calculation: Use the instance segmentation algorithm to perform pixel-level segmentation on each historical image to obtain the instance mask of the patient's area. Based on the instance mask, calculate the area of ​​the minimum circumscribed rectangle of the patient's area and record it as the activity range; Threshold update: determine whether the activity range is smaller than a preset activity range threshold; if so, reduce the preset similarity threshold and execute the similarity determination step; if not, increase the preset similarity threshold and execute the similarity determination step.

[0012] This application uses an instance segmentation algorithm to perform pixel-level segmentation on each historical image, accurately determining the patient's location and range within the image, thereby obtaining an instance mask representing the patient's area. Based on the instance mask, the area of ​​the minimum bounding rectangle is calculated as the activity range. In the medical field, a patient's activity range often reflects their physical condition. For example, a patient with a more serious condition may have a smaller activity range, while a patient whose condition has improved may have a larger activity range. Therefore, obtaining the activity range provides a deeper understanding of the patient's condition. Subsequently, this application dynamically adjusts the preset similarity threshold based on the comparison result of the patient's activity range with a preset activity range threshold, providing greater flexibility and adaptability in determining patient status. When the patient's activity range is less than the preset activity range threshold, it indicates that the patient is in a more serious condition. In this case, reducing the preset similarity threshold can make it easier to identify subtle movements during the similarity determination step. Conversely, when the patient's activity range is larger, it indicates that the patient is in relatively good physical condition. Increasing the preset similarity threshold can avoid misjudging the patient's condition due to normal movements while the patient is resting.

[0013] This application can adjust the similarity threshold based on the patient's range of movement, making visitation decisions based on similarity more reasonable. Because the patient's range of movement is closely related to their condition, this correlation can more accurately determine whether the patient is suitable for visitation, balancing the patient's need for rest and the need for communication with the outside world, and improving the scientific and humane nature of visitation management.

[0014] Optionally, after performing the step of updating the threshold, the method further includes: GNN model update: Calculate the difference in activity ranges between any two moments, remove directed edges when the difference is greater than the preset difference threshold, reset the weight of each directed edge in the new GNN model according to the time interval of the historical image, and update the GNN model.

[0015] This application calculates the difference between any two activity ranges, and decides to retain or delete the corresponding directed edges in the GNN model based on the comparison between the difference and the preset difference threshold. When the difference between the two activity ranges is small, it means that the patient's state does not change much at the corresponding time point. Retaining the corresponding directed edges helps the model learn the feature relationship under similar states, so that the GNN model tries not to learn too jumpy or irrelevant feature connections. After the new GNN model is obtained through the above operations, the weight of each directed edge is reset according to the time interval of the historical image, and the attention of the GNN model to the features at different time points is further adjusted. This application dynamically adjusts the structure and weight of the GNN model based on the difference in the patient's activity range, which can enable the GNN model to better adapt to the state change patterns of different patients. The changes in the activity ranges of different patients are different. Through this adaptive adjustment, the GNN model can respond to various situations more flexibly, improve the processing capabilities and analysis accuracy of different patient data, and thus improve the judgment accuracy based on feature analysis in the visit management method.

[0016] Optionally, in the step of updating the GNN model: In the step of updating the GNN model: a support vector machine classification algorithm is also used to classify all differences to obtain differences corresponding to the hyperplane, and the differences corresponding to the hyperplane are recorded as first data, differences greater than the first data are recorded as second data, and differences less than the first data are recorded as third data; different hierarchical labels are added to the graph nodes corresponding to the first data, second data, and third data respectively; according to the hierarchical labels, the weights of the directed edges of adjacent levels in the GNN model are increased and the weights of the directed edges across levels are reduced, and the GNN model is updated.

[0017] This application uses a support vector machine classification algorithm to classify the activity range differences and divides the differences into three categories (first data, second data, and third data), which can more carefully analyze the degree of change in the patient's status. The differences in different categories reflect the different situations of the patient's status change, providing a basis for the subsequent differentiated processing of graph nodes in the GNN model. For example, a larger difference (second data) may indicate that the patient's status has changed significantly, while a smaller difference (third data) indicates that the status is relatively stable. This classification method helps the model capture the relationship between features more accurately. Subsequently, this application adds hierarchical labels to the graph nodes corresponding to different categories of differences in the GNN model, and adjusts the weights of directed edges according to the hierarchical labels (increasing the weights of directed edges in adjacent levels and reducing the weights of directed edges across levels), further optimizing the model structure. Nodes in adjacent levels often represent relatively continuous changes in patient status. Increasing the weights between them can enhance the GNN model's capture of the relationship between such continuously changing states; while reducing the weights of cross-level directed edges can reduce the GNN model's attention to irrelevant or jumpy state changes, thereby making the feature learning of the GNN model more reasonable and effective, and improving the model's ability to represent patient status characteristics.

[0018] By adopting the above scheme, this application can enable the model to better adapt to the diversity of patient status changes. Different patients may have different status change patterns. By classifying the difference and adjusting the model accordingly, the model can handle various situations more flexibly and improve the accuracy of analyzing different patient data. In visitation management, a more accurate model can more reasonably determine whether a patient is suitable for visitation, improving the scientificity and rationality of management.

[0019] Optionally, the method further comprises the step of continuous active time calibration; Continuous active time calibration: Calculate the proportion of similarities less than a preset similarity threshold in all similarities, record the proportion as activity, select time periods with activity greater than a preset activity threshold and continuous time greater than a preset active time threshold, and calibrate them as continuous active time; After executing the step of similarity determination and before executing the step of generating a reservation code, the method further includes: Time judgment: judge whether the appointment time is within the interval of continuous active time. If so, execute the step of generating an appointment code. If not, send an appointment failure signal; Appointment time recommendation: When sending an appointment failure signal, the continuous active time closest to the appointment time is sent as the recommended appointment time.

[0020] This application defines activity by calculating the proportion of similarities less than a preset similarity threshold in all similarities, providing a new quantitative indicator for evaluating the degree of change in the patient's status. Activity can comprehensively reflect the stability of the patient's status at different time points. For example, a higher activity means that the patient's status has changed significantly at multiple time points, and may be in an unstable or recovery stage; a lower activity indicates that the patient's status is relatively stable. When the activity is greater than the preset activity threshold, the period is marked as continuous active time. After that, this application determines whether the appointment time is within the interval of continuous active time. If so, an appointment code is generated. Otherwise, an appointment failure signal is sent, and the continuous active time closest to the appointment time is fed back to the appointment personnel as the recommended appointment time, so that they can understand the situation in a timely manner and make corresponding adjustments, thereby improving the efficiency and satisfaction of visits, and also improving the experience of both patients and visitors.

[0021] Optionally, in the step of generating the appointment code, at least one visit identification code is generated. When there are multiple visit identification codes, each identification code is different from the remaining identification codes.

[0022] In a hospital environment with complex personnel flow, the uniqueness of each visitor identification code ensures that only those holding the corresponding valid identification code can enter a specific visiting area, reducing the risk of unauthorized personnel entering and ensuring patient safety and ward order. Different visitor identification codes enable hospital administrators to more clearly record and manage the information of each visitor. When tracing visitation details within a specific time period, the identification code can quickly and accurately locate the corresponding visitor and related information, such as visiting times and appointment information. This helps improve management efficiency and promptly identify and address potential issues, such as illegal visitation. This application can generate at least one visitor identification code and ensure its uniqueness when multiple are used, meeting different visitation scenarios and needs. For example, patients with special medical conditions may require multiple different types of visitors (such as doctors, family members, caregivers, etc.). Each visitor can have a unique visitor identification code for targeted management and control. Furthermore, the number and usage of identification codes can be flexibly adjusted according to the specific visitation situation.

[0023] Optionally, the visiting identification code includes a first identification code for entering the ward area and a second identification code for leaving the ward area, the effectiveness time of the first identification code is later than the start time of the appointment time, and the expiration time of the second identification code is earlier than the end time of the appointment time.

[0024] This application divides visitor identification codes into two types: entry and exit, with strict limits on their validity and expiration times. This allows precise control over the length of a visitor's stay in a ward. This allows visitors to enter and exit wards only during their scheduled appointments and within a more precise timeframe, reducing premature entry and late departures. This maintains proper order in the ward and ensures patients have a relatively stable, uninterrupted rest and treatment environment. For example, this prevents visitors from entering the ward too early and gathering to wait, or from lingering long after their scheduled appointment, potentially disrupting subsequent arrangements. By clearly distinguishing between entry and exit identification codes, this adds a layer of authentication and management, allowing hospital administrators to more accurately track and record each visitor's entry and exit times and behavior, facilitating tracing and investigation when issues arise. Furthermore, different identification codes reduce the risk of misuse or abuse, as misusers would need to obtain both valid entry and exit identification codes and use them within the corresponding timeframes, making it more difficult. This improves the security of visitor management. Precisely controlling visiting times helps hospitals more effectively manage and utilize ward resources. This can reduce resource waste and conflicts caused by uncertainty in visiting hours. For example, subsequent nursing procedures and examination arrangements can be better coordinated with visiting hours, improving overall medical service efficiency. Furthermore, for wards with multiple patients, this can reduce interference between visiting times and improve ward utilization efficiency.

[0025] Optionally, the method further includes: Obtain historical data: Obtain training data, the training data including: the usage time of all first identification codes during the historical visiting time, the usage time of all second identification codes, and the remaining visiting quota at each moment, where the remaining visiting quota is equal to the preset number threshold minus the number of visitors who have made reservations; Modeling and training: Establish a BI-LSTM model, use training data to train the BI-LSTM model, and obtain the trained BI-LSTM model; Personnel prediction: Input the appointment time into the trained BI-LSTM model and output the predicted remaining visitation slots.

[0026] This application builds a decision-making system based on historical data by collecting the usage times of the first and second identification codes during historical visiting hours, as well as the remaining visiting slots at each moment. This data reflects the patterns and patterns of past visitation activities, providing a solid foundation for subsequent predictions. For example, by analyzing the usage times of the identification codes, we can understand the peak hours for visitors to enter and exit the hospital, thereby better predicting future visitation patterns. Subsequently, this application makes predictions based on this extensive historical data, reducing errors caused by subjective judgment and making the allocation of visiting slots more scientific and reasonable, thereby improving the rationality of hospital visitation allocation and the efficiency of visitation management. The LSTM model itself excels at processing time series data and can capture long-term dependencies in the data. The bidirectional LSTM model further combines past and future information, making it highly effective for predicting the remaining visiting slots with time-series characteristics. The predicted remaining visiting slots can provide important decision-making basis for appointment holders and hospital management. Appointments can be made based on the predicted results to rationally schedule visiting times, reducing the problem of invalid appointments when the remaining slots are insufficient.

[0027] In the second aspect, this application provides a visit management system for injured patients, which adopts the following technical solutions: A visiting management system for injured patients, comprising: a processor and a memory; The memory stores a computer-readable storage medium; When the processor processes the computer program stored on the computer-readable storage medium, the method described above is implemented.

[0028] In summary, this application includes at least one of the following beneficial technical effects: 1. This application first collects historical images from the appointment time and visiting time in the visit request. It then uses a neural network model for feature extraction, automatically learning and extracting effective features from historical images. Historical images from the same time period as the appointment time are input into the model to extract feature vectors representing the patient's status. The timestamps of the historical images are then used as time tags for the feature vectors. The similarity between feature vectors from adjacent times is calculated based on the time tags. If the similarity is greater than a preset similarity threshold, it indicates that the patient is unsuitable for interruption. Sending an appointment failure signal can reduce unnecessary visits and protect the patient's rest and treatment environment. Otherwise, a visit identification code is generated based on the visit request information and sent to the appointment person, providing a convenient identity verification method for the appointment person, ensuring that only those with an appointment can visit, thereby improving the security and standardization of visit management.

[0029] 2. This application uses an instance segmentation algorithm to perform pixel-level segmentation on each historical image, accurately determining the patient's location and range within the image, thereby obtaining an instance mask representing the patient's area. Based on the instance mask, the area of ​​the minimum bounding rectangle is calculated as the activity range. In the medical field, a patient's activity range often reflects their physical condition. For example, a patient with a more serious condition may have a smaller activity range, while a patient whose condition has improved may have a larger activity range. Therefore, obtaining the activity range provides a deeper understanding of the patient's condition. Subsequently, this application dynamically modifies the preset similarity threshold based on the comparison result between the patient's activity range and a preset activity range threshold, providing greater flexibility and adaptability in determining patient status. When the patient's activity range is less than the preset activity range threshold, it indicates that the patient is in a more serious condition. In this case, reducing the preset similarity threshold can make it easier to identify subtle movements during the similarity determination step. Conversely, when the patient's activity range is larger, it indicates that the patient is in relatively good physical condition. Increasing the preset similarity threshold can avoid misjudging the patient's condition due to normal movements while the patient is resting. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of Example 1 of the present application; Figure 2 This is a flow chart of Example 2 of the present application; Figure 3 This is a flowchart of Example 3 of the present application. DETAILED DESCRIPTION

[0031] The following combination Figures 1 to 3 This application is described in further detail.

[0032] Visiting hours: refers to the time stipulated by the hospital for visiting injured patients.

[0033] Appointment time: refers to the time when the family members of the injured patient make an appointment to visit the injured patient. The family members of the injured patient will choose the expected time as the appointment time within the visiting hours.

[0034] Example 1: This example discloses a method for managing visits to injured patients. Figure 1The method includes: S11 data collection, S12 visit judgment, firstly, data collection is performed to obtain visit request information including the number of appointment personnel and appointment time, as well as multiple historical images of the patient during the visiting time; visit judgment includes the steps of feature extraction, similarity calculation, similarity judgment and appointment code generation, wherein feature extraction is to establish a neural network model and input historical images of the same time period to output feature vectors, similarity calculation is to calculate the similarity of adjacent feature vectors based on timestamp labels, and similarity judgment is to judge whether the similarity is greater than a preset threshold. If so, an appointment failure signal is sent, otherwise the step of generating an appointment code is executed, and finally, an appointment code is generated, that is, a visit identification code is generated according to the visit request information and sent to the appointment personnel end. This embodiment includes the following steps: S11 data collection, obtaining visit request information and multiple historical images of the patient during the visiting time, wherein the visit request information includes the appointment time.

[0035] These historical images contain rich information about the patient's status, such as their activity status and physical condition, providing a crucial basis for subsequent visit decisions based on image analysis. For example, by comparing images taken at different times, we can understand the changing trends in the patient's condition. When the patient is awake, adjacent historical images show greater differences, while when the patient is resting, the differences are smaller.

[0036] In other embodiments, the visiting request information also includes: the number of people with appointments, by calculating the sum of the number of people who have made appointments and the number of people with appointments within the appointment time (recorded as the total number of people), it is determined whether the total number of people is greater than the number threshold of visitors allowed set by the hospital, and then it is determined whether the people with appointments can successfully make an appointment this time. If the total number of people is greater than the number threshold, the appointment fails; if the total number of people is not greater than the number threshold, the appointment is successful, and S12 visiting judgment is executed.

[0037] S12 visit judgment includes S121 feature extraction, S122 similarity calculation, S123 similarity judgment and S124 generation of appointment code.

[0038] S121 Feature Extraction: This embodiment leverages the powerful feature learning capabilities of the neural network model to analyze and process historical images from the same time period as the appointment. Trained with extensive data, the neural network model automatically identifies key features related to the patient's condition in historical images and converts these images into corresponding feature vectors.

[0039] S122 calculates similarity, using the timestamps of historical images as time tags and associating them with corresponding feature vectors. By calculating the similarity between feature vectors at adjacent times, the degree of change in the patient's status over time is measured. A higher similarity indicates that the patient's status at adjacent time points is more similar, and the patient's status during that time period is relatively stable; conversely, a lower similarity indicates that the patient's status has changed significantly.

[0040] S123: Similarity determination: The calculated similarity is compared with a preset similarity threshold. If all similarities are greater than the preset similarity threshold, it indicates that the patient is resting. The patient's status is set to unsuitable for visiting, thereby reducing the risk of disturbing the patient's rest due to visiting. At this time, a reservation failure signal is sent. If any similarity is less than the preset similarity threshold, the patient's status has changed, indicating that the patient is now suitable for visiting. The next step is to generate a reservation code.

[0041] S124 generates an appointment code. After determining that the patient is suitable for visitation, it generates a visitor identification code based on the initially collected visitor request information and sends it to the person who made the appointment. The visitor identification code serves as a credential for visitors to enter the ward. This approach not only protects the visiting rights of those with legitimate appointments, but also effectively manages visitors, ensuring an orderly visit process.

[0042] This embodiment first collects data to obtain visiting request information such as the number of people who have made appointments, appointment times, and multiple historical images of the patient during the visiting time; in visiting judgment, the historical images are first extracted through a neural network model to obtain feature vectors, and then the similarity of adjacent feature vectors is calculated based on the timestamps. It is compared with a preset similarity threshold and a reservation failure signal is sent if it exceeds the threshold. Otherwise, a visit identification code is generated according to the visiting request information and sent to the person who has made the appointment, thereby realizing scientific and intelligent visiting management.

[0043] Example 2: Reference Figure 2 The difference between this embodiment and embodiment 1 is that the method further includes: S20 constructs a model, using the feature vectors corresponding to historical images as graph nodes and the time sequence of collecting historical images as directed edges to build a GNN model.

[0044] S21 feature update, sets the weight of each directed edge in the GNN model according to the interval of historical image acquisition time, inputs the feature vector into the GNN model, outputs the updated feature vector, and uses the updated feature vector as the new feature vector.

[0045] This step leverages the unique architecture and principles of the Graph Neural Network (GNN) model to deeply explore the temporal correlations and potential relationships between historical image features, enabling the feature vector to more accurately reflect the patient's status. Specifically: The feature vectors corresponding to historical images are used as graph nodes, and each feature vector is assigned a node identity within the graph structure. This allows previously independent feature vectors to establish connections through the graph structure. Furthermore, directed edges are constructed according to the chronological order in which the historical images were collected. The direction of the directed edge represents the flow of time. This construction method integrates the time dimension into the graph structure, intuitively reflecting the dynamic process of patient status changes over time. For example, at different stages of a patient's disease progression, image features at adjacent time points are connected by directed edges, clearly showing the trajectory of the disease's changes.

[0046] The weight of each directed edge in the GNN model is set based on the time intervals between historical image acquisitions. Images acquired with a shorter time interval reflect relatively minor changes in patient status and have higher correlations between features. Therefore, these directed edges are given larger weights, allowing the GNN model to prioritize these relationships when processing. On the other hand, images acquired with longer time intervals may have more significant feature changes and relatively lower correlations, resulting in smaller directed edge weights. This weighting mechanism allows the GNN model to more reasonably balance the importance of features at different time points, highlighting the impact of recent, less-changed features on current status judgment.

[0047] The feature vectors from the S121 feature extraction are input into the constructed GNN model. Based on the graph structure and edge weights, the GNN model performs information transfer and feature aggregation on the feature vectors. During information transfer, each node (feature vector) receives information from neighboring nodes and updates it based on its own information. Feature aggregation further integrates this information to generate a new feature representation. The updated feature vectors are ultimately output, denoted as new feature vectors. These new feature vectors comprehensively consider the interrelationships between features at different time points and provide a more comprehensive and accurate description of the patient's status compared to the original feature vectors.

[0048] Then, in S122 , similarity is calculated between new feature vectors at adjacent times according to the time tags.

[0049] In other embodiments, the method further comprises: S22 segmentation and calculation: The instance segmentation algorithm performs detailed pixel-level processing on each historical image, accurately distinguishes the area where the patient is located, and generates an instance mask. Based on this mask, the minimum circumscribed rectangular area of ​​the patient area is calculated and defined as the activity range. This step uses quantitative values ​​to intuitively reflect the size of the patient's activity space in the image, providing key basic data for subsequent analysis. For example, the activity range of patients in the early stages of postoperative recovery is smaller, while the activity range of patients whose condition improves will increase accordingly.

[0050] S23 threshold update: The calculated activity range is compared with the preset activity range threshold for judgment. If the activity range is less than the preset threshold, it means that the patient is in a stage where he cannot move freely. In this case, the preset similarity threshold is reduced, and the judgment standard in the similarity judgment step is lowered, increasing the time available for appointment. Conversely, if the activity range is greater than the preset activity range threshold, it means that the patient is in an active state. The preset similarity threshold is appropriately increased to improve the distinction between resting and awake periods.

[0051] S24 GNN model update, calculate the difference between any two activity ranges, and determine whether each difference is less than the preset difference threshold. If so, retain the directed edge corresponding to the difference in the GNN model; if not, delete the directed edge corresponding to the difference in the GNN model.

[0052] By calculating the difference between any two ranges of motion, the system captures the magnitude of changes in a patient's range of motion. Each difference is evaluated based on a preset difference threshold. Directed edges in the GNN model corresponding to differences below the threshold are retained, while edges exceeding the threshold are deleted. This streamlines the GNN model structure, retaining relatively stable directed edges. By iteratively traversing all differences and reweighting directed edges based on historical image time intervals, the GNN model focuses more on adjacent nodes with similar states, providing a more accurate model foundation for feature updates.

[0053] The support vector machine classification algorithm is used to classify all differences to obtain the differences corresponding to the hyperplane. The differences corresponding to the hyperplane are recorded as the first data, the differences greater than the first data are recorded as the second data, and the differences less than the first data are recorded as the third data. Different hierarchical labels are added to the graph nodes corresponding to the first data, second data, and third data respectively; according to the hierarchical labels, the weights of the directed edges of adjacent levels are increased and the weights of the directed edges across levels are reduced, and the GNN model is updated so that the updated GNN model can pay more attention to similar graph nodes.

[0054] This step uses a support vector machine classification algorithm to classify all activity range differences. Using the differences corresponding to the hyperplane as the boundary, the differences are divided into three categories: first data, second data, and third data, representing different degrees of state change. Subsequently, hierarchical labels are added to the graph nodes corresponding to the different categories of differences. Based on the hierarchical labels, the weights of directed edges are adjusted. The weights of adjacent hierarchical edges are increased to strengthen the transmission of features that continuously change the patient's state. The weights of cross-hierarchical edges are reduced to weaken the transmission of features with large activity ranges. This allows the GNN model to be updated, making the updated GNN model pay more attention to the continuously changing graph nodes.

[0055] In S21 feature update, the feature vector is input into the updated GNN model.

[0056] Example 3: Reference Figure 3 The difference between this embodiment and embodiment 1 is that the method further includes: S31 Continuous Active Time Calibration: This step defines activity by calculating the proportion of similarities less than a preset similarity threshold in all similarities. Similarity represents the degree of similarity between historical images at different time points. A similarity less than the preset similarity threshold means that the patient's status has changed significantly at these time points. The activity indicator comprehensively reflects the dynamic changes in the patient's status. The higher the proportion, the more frequent the patient's status changes and the higher the activity; otherwise, the lower the proportion. For example, if most of the similarities are less than the preset threshold over a period of time, then the activity will be high, indicating that the patient is awake.

[0057] Time periods with activity levels greater than a preset activity threshold and duration greater than a preset active time threshold are selected and marked as continuous active time. In other words, the system determines whether similarity levels less than a preset similarity threshold are continuous. If they are continuous, indicating that the patient was awake during this period, this period is marked as continuous active time.

[0058] After executing S123 similarity determination and before executing S124 generating a reservation code, the method further includes: S32 time judgment, judging whether the reservation time is within the interval of continuous active time, if so, executing S124 to generate a reservation code; if not, sending a reservation failure signal.

[0059] S33 appointment time recommendation: when sending an appointment failure signal, the continuous active time closest to the appointment time is sent as the recommended appointment time.

[0060] In S124, when generating the appointment code, this step will generate at least one visit identification code. When there are multiple identification codes, each identification code is unique and different from other identification codes. In addition, the visit identification code is divided into a first identification code for entering the ward area and a second identification code for leaving the ward area. The effective time of the first identification code is later than the start time of the appointment time, which can ensure that visitors do not enter the ward too early to disturb the patients; the expiration time of the second identification code is earlier than the end time of the appointment time, which can avoid visitors from staying in the ward after the appointment time, which helps to maintain the order of the ward and the rest environment of the patients.

[0061] In other embodiments, the method further comprises: S41 acquires historical data and collects training data, including the usage times of all first and second identification codes during historical visiting hours, as well as the remaining visitation quota at each time. The remaining visitation quota is calculated by subtracting the number of visitors who have made reservations from a preset threshold. This training data records detailed information about past visitation activities.

[0062] S42 modeling and training: Build a Bidirectional Long Short-Term Memory (BI-LSTM) model, a deep learning model that effectively processes time series data. It considers both past and future information, making it highly effective for predicting data with temporal characteristics. Using the acquired training data, the BI-LSTM model is trained to learn patterns and patterns in historical visitation data, thereby adjusting the model's parameters to accurately predict future visitation patterns.

[0063] S43 personnel prediction: input the appointment time into the trained BI-LSTM model, output the predicted remaining visiting places, and feed back the predicted remaining visiting places to the appointment personnel.

[0064] The appointment time is input into the trained BI-LSTM model. Based on the learned patterns and patterns, the BI-LSTM model outputs a predicted number of remaining visitation slots and feeds this information back to the appointment maker. This prediction allows the appointment maker to more effectively arrange visitation plans, while the hospital can also proactively allocate and manage resources based on the forecast.

[0065] Example 4: This example discloses a visit management system for injured patients, the system comprising: processor and memory; The memory stores a computer-readable storage medium; When the processor processes the computer program stored on the computer-readable storage medium, the method described above is implemented.

[0066] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for managing visits to injured patients, characterized in that: include: Data collection: obtaining visit request information and multiple historical images of the patient during the visit time, wherein the visit request information includes the appointment time; Feature extraction: A neural network model is established for feature extraction. Historical images in the same period as the appointment time are sequentially input into the neural network model, and the feature vector corresponding to each historical image is output; Calculate similarity: Use the timestamp of the historical image as the time label of the corresponding feature vector, and calculate the similarity between the feature vectors of adjacent times based on the time label; Similarity judgment: judge whether the similarities are greater than the preset similarity threshold. If so, send a reservation failure signal; if not, execute the step of generating a reservation code; Generate appointment code: Generate a visit identification code based on the visit request information and send the visit identification code to the appointment person.

2. The method for managing visits to injured patients according to claim 1, characterized in that: The method further includes: constructing a GNN model using feature vectors corresponding to historical images as graph nodes and the time sequence of collecting historical images as directed edges; Feature update: Set the weight of each directed edge in the GNN model according to the interval of historical image acquisition time, input the feature vector into the GNN model, output the updated feature vector, and use the updated feature vector as the new feature vector.

3. The method for managing visits to injured patients according to claim 2, characterized in that: The method further comprises: Segmentation and calculation: Use the instance segmentation algorithm to perform pixel-level segmentation on each historical image to obtain the instance mask of the patient's area. Based on the instance mask, calculate the area of ​​the minimum circumscribed rectangle of the patient's area and record it as the activity range; Threshold update: determine whether the activity range is smaller than a preset activity range threshold; if so, reduce the preset similarity threshold and execute the similarity determination step; if not, increase the preset similarity threshold and execute the similarity determination step.

4. The method for managing visits to injured patients according to claim 3, characterized in that: After performing the step of updating the threshold, the method further includes: GNN model update: Calculate the difference in activity ranges between any two moments, remove directed edges when the difference is greater than the preset difference threshold, reset the weight of each directed edge in the new GNN model according to the time interval of the historical image, and update the GNN model.

5. The method for managing visits to injured patients according to claim 4, characterized in that: The method further comprises: In the step of updating the GNN model: a support vector machine classification algorithm is also used to classify all differences to obtain differences corresponding to the hyperplane, and the differences corresponding to the hyperplane are recorded as first data, differences greater than the first data are recorded as second data, and differences less than the first data are recorded as third data; different hierarchical labels are added to the graph nodes corresponding to the first data, second data, and third data respectively; according to the hierarchical labels, the weights of the directed edges of adjacent levels in the GNN model are increased and the weights of the directed edges across levels are reduced, and the GNN model is updated.

6. The method for managing visits to injured patients according to any one of claims 1 to 5, characterized in that: The method further comprises the steps of continuous active time calibration; Continuous active time calibration: Calculate the proportion of similarities less than a preset similarity threshold in all similarities, record the proportion as activity, select time periods with activity greater than a preset activity threshold and continuous time greater than a preset active time threshold, and calibrate them as continuous active time; After executing the step of similarity determination and before executing the step of generating a reservation code, the method further includes: Time judgment: judge whether the appointment time is within the interval of continuous active time. If so, execute the step of generating an appointment code. If not, send an appointment failure signal; Appointment time recommendation: When sending an appointment failure signal, the continuous active time closest to the appointment time is sent as the recommended appointment time.

7. The method for managing visits to injured patients according to claim 6, characterized in that: In the step of generating the reservation code, at least one visit identification code is generated. When there are multiple visit identification codes, each identification code is different from the remaining identification codes.

8. The method for managing visits to injured patients according to claim 7, characterized in that: The visiting identification code includes a first identification code for entering the ward area and a second identification code for leaving the ward area. The effective time of the first identification code is later than the start time of the appointment time, and the expiration time of the second identification code is earlier than the end time of the appointment time.

9. The method for managing visits to injured patients according to claim 8, characterized in that: The method further comprises: Obtain historical data: Obtain training data, the training data including: the usage time of all first identification codes during the historical visiting time, the usage time of all second identification codes, and the remaining visiting quota at each moment, where the remaining visiting quota is equal to the preset number threshold minus the number of visitors who have made reservations; Modeling and training: Establish a BI-LSTM model, use training data to train the BI-LSTM model, and obtain the trained BI-LSTM model; Personnel prediction: Input the appointment time into the trained BI-LSTM model and output the predicted remaining visitation slots.

10. A visit management system for injured patients, characterized in that: include: processor and memory; The memory stores a computer-readable storage medium; When the processor processes the computer program stored on the computer-readable storage medium, the method according to any one of claims 1 to 9 is implemented.

Images