Medical assistant system and method for internal medicine emergency patients based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based medical assistant system and method for emergency internal medicine patients, which relates to the field of medical technology. The method collects infusion physiological status data in real time by setting up multiple sensor groups; performs data preprocessing and standardization and normalization through an AI preprocessing strategy to form a standard infusion physiological status data set; inputs the standard infusion physiological status data set into the constructed infusion time prediction model, and combines nonlinear attenuation modeling to achieve high-precision prediction of the patient's infusion completion time prediction value Tinf, and divides emergency internal medicine patients who are about to complete infusion through a preliminary comparative evaluation mechanism, thereby identifying the patient group whose infusion is about to be completed in advance, achieving accurate early warning and advance planning of fluid exchange tasks, and significantly reducing the incidence of adverse events such as infusion emptying, blood return, and venous tube blockage.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to an artificial intelligence-based medical assistant system and method for emergency internal medicine patients. Background Art

[0002] This research involves the field of artificial intelligence medical assistance, particularly within the research scope of intelligent medical dispatch systems. More specifically, it involves an AI-based medical assistance method for emergency internal medicine patients. This method is primarily applied to the management of high-frequency nursing procedures, primarily intravenous infusion therapy, in emergency internal medicine settings. This is particularly relevant in situations where concentrated patient populations and numerous infusion tasks create challenges in scheduling nursing resources. This assistance method enables intelligent collection of patient physiological status data during emergency internal medicine infusions, dynamic prediction of the infusion process, rational allocation of nurse tasks, and system-level load assessment.

[0003] At present, in the emergency infusion area of ​​the internal medicine department of the hospital, nurses need to frequently go back and forth to the patient's bedside to perform tasks such as infusion inspections, changing medications, and handling abnormalities. Especially during peak hours, due to the concentrated number of patients and asynchronous infusion cycles, some nurses are called by multiple patients in a short period of time, resulting in the problem of excessive local nursing load. In addition, the existing infusion inspection scheduling usually relies on manual experience and judgment, lacks real-time data support, and cannot effectively predict which patients are about to complete the infusion or whether there are potential risks, which can easily lead to delayed response to fluid changes, excessive infusion, and even secondary medical risks such as venous reflux. This "task stacking" problem seriously affects the quality and efficiency of emergency care and increases the manpower pressure on nurses.

[0004] The above phenomenon stems primarily from three core deficiencies: first, a lack of real-time, quantifiable monitoring of a patient's current infusion status; second, an inability to extract effective indicators from historical nursing behaviors to dynamically assess nurse tasks; and third, the current allocation of nurse tasks lacks a systematic, holistic view and an intelligent scheduling optimization mechanism. When these factors combine, they often lead to instantaneous peaks in infusion tasks during certain time periods, which in turn can lead to the following abnormal consequences: some patients remain unattended for extended periods after their infusion, making them susceptible to blood backflow or tube blockage; some nurses become overloaded, leading to fatigue and increased error rates; and overall system scheduling becomes unbalanced, manifesting as local overload and global inefficiency. These consequences not only impact the continuity and safety of emergency treatment but can also lead to medical disputes due to delayed resolution, hindering the overall improvement of hospital service quality. Summary of the Invention

[0005] In response to the deficiencies of the existing technology, the present invention provides an artificial intelligence-based medical assistant system and method for internal medicine emergency patients, which solves the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: comprising the following steps:

[0007] S1. By setting up a sensor group, real-time infusion physiological status data is collected, and a wireless communication network is set up to transmit the infusion physiological status data to the AI ​​scheduling server;

[0008] S2. Connect the nursing task platform to the AI ​​scheduling server, extract nursing behavior data in real time, and pre-process the nursing behavior data and infusion physiological status data to obtain a standard nursing behavior data set and a standard infusion physiological status data set;

[0009] S3. Build an infusion time prediction model, extract a standard infusion physiological state data set and input it into the infusion time prediction model, calculate and output the infusion completion time prediction value Tinf, and set up a preliminary evaluation mechanism to trigger the scheduling optimization mechanism;

[0010] S4. After the scheduling optimization mechanism is triggered, the nursing scheduling interference function Hs is calculated based on the predicted value of the infusion completion time Tinf and the standard nursing behavior data set, and the nursing threshold F2 is set to perform nursing evaluation, and the task diversion strategy is executed based on the evaluation results;

[0011] S5. After the task diversion strategy is executed, the global pressure index Osys is calculated and output based on the nursing scheduling interference function Hs of all current nurses, and the pressure threshold Oth is set for secondary comparative evaluation to determine the scheduling load balance.

[0012] Preferably, said S1 includes S11 and S12;

[0013] S11. Install a sensor group on the internal medicine emergency infusion patient and the infusion device to collect the infusion physiological status data of the internal medicine emergency infusion patient in real time;

[0014] The sensor group includes a wristband pulse sensor, an infrared thermal imager, a drug physical database API, and a smart infusion pump sensor;

[0015] The infusion physiological state data includes micro-pressure fluctuation curve, blood flow area heat map sequence, pharmacological characteristic data set, total set volume Vset and injection volume Vinjected;

[0016] The micro-pressure fluctuation curve is obtained by having an internal medicine emergency infusion patient wear a wristband pulse sensor, attached to the patient's venipuncture area, and detecting the micro-pressure fluctuation of the local tissue every 5 seconds during the infusion process;

[0017] The blood flow area heat map sequence is collected and acquired by an infrared thermal imager fused with near-infrared reflective blood flow imaging NIRS;

[0018] The pharmacological characteristic dataset automatically captures the drug code in the doctor's order through the drug physical database API, and extracts the pharmacological characteristic dataset based on the drug code, wherein the pharmacological characteristic dataset includes the molecular weight M of the drug used by the i-th patient, the dynamic viscosity u of the solvent used by the i-th patient, and the molecular formula of the drug used by the i-th patient;

[0019] The set volume Vset and the injected volume Vinjected are collected and acquired by the intelligent infusion pump sensor;

[0020] S12. By connecting the network communication modules of all sensor groups to the intranet LoRa / MQTT gateway, the real-time collected infusion physiological status data will be wirelessly transmitted to the constructed AI scheduling server.

[0021] Preferably, said S2 includes S21 and S22;

[0022] S21. Set up a medical care platform API in the AI ​​scheduling server, integrate the AI ​​scheduling server with the nursing task platform, and extract nursing behavior data in real time;

[0023] The nursing behavior data includes the path complexity Lj between the i-th patient and the j-th nurse ij , the task operation time of the i-th patient and the j-th nurse Cz ij and the number of abnormal states of the i-th patient Yc i ;

[0024] The path complexity Lj between the i-th patient and the j-th nurse ij By using RFID tags worn by nurses and UWB indoor positioning, combined with the path node graph algorithm, the shortest time path between the nurse and the patient is calculated;

[0025] The task operation time of the i-th patient and the j-th nurse is Cz ij Obtained by automatically recording the difference between the task start time and the completion feedback time for regression analysis;

[0026] The number of abnormal states of the i-th patient Yc i The number of times the infusion blockage, backflow and alarm conditions are detected by the infusion pump is obtained;

[0027] S22. Construct an AI preprocessing strategy in the AI ​​scheduling server, and preprocess the acquired nursing behavior data and infusion physiological status data based on the AI ​​preprocessing strategy to obtain a standard nursing behavior data set and a standard infusion physiological status data set, respectively;

[0028] The AI ​​preprocessing strategy includes filtering, image preprocessing, graph neural networks, machine algorithms, and AI-driven normalization.

[0029] The standard infusion physiological state data set includes the local venous permeability index A of the i-th patient i , terminal venous return compressibility factor Ys of patient i i , the diffusion hysteresis constant P of the drug solution molecules of the i-th patient i and the current remaining infusion volume V for the i-th patient i ;

[0030] The standard nursing behavior dataset includes the normalized path complexity Lj between the i-th patient and the j-th nurse ij , the task operation time of the i-th patient and the j-th nurse Cz ij and the number of abnormal states of the i-th patient Yc i .

[0031] Preferably, said S3 includes S31 and S32;

[0032] S31. Construct an infusion time prediction model using a machine algorithm, extract a standard infusion physiological state data set and input it into the infusion time prediction model, calculate and output a predicted value Tinf of the infusion completion time for each internal medicine emergency patient, and predict the infusion completion time for each internal medicine emergency patient;

[0033] The infusion completion time prediction value Tinf is calculated and output by the following infusion time prediction model;

[0034] ;

[0035] Where, Tinf i Predicted value of the infusion completion time for the i-th patient, Represents the minimum constant, with a value of 1×10 -3 , e represents the exponential function.

[0036] Preferably, S32, after obtaining the predicted value Tinf of the infusion completion time for each patient, a preliminary evaluation mechanism is executed. The preliminary evaluation mechanism sets the infusion completion warning time threshold ΔTth by the user, and extracts the current system time tnow and the predicted value Tinf of the infusion completion time for each patient for preliminary comparative evaluation. The emergency infusion patients who are about to complete the infusion are summarized to construct an infusion completion collection Ct. When the infusion completion collection Ct exceeds the resource upper limit threshold Nth preset for hospital resources, the scheduling optimization mechanism is triggered.

[0037] The infusion completion collection Ct is constructed in the following manner;

[0038] ;

[0039] If the infusion completion set Ct is greater than the resource upper limit threshold Nth, the scheduling optimization mechanism is triggered;

[0040] If the infusion completion set Ct ≤ resource upper limit threshold Nth, it will not be triggered.

[0041] Preferably, said S4 includes S41 and S42;

[0042] S41. After the preliminary evaluation mechanism triggers the scheduling optimization mechanism, the predicted infusion completion time Tinf of the patient in the infusion completion set Ct and the standard nursing behavior data set are extracted, and the nursing scheduling interference function Hs is output to analyze the pressure level of the task to be completed;

[0043] The nursing scheduling interference function Hs is calculated and output by the following algorithm formula:

[0044] ;

[0045] Where Hs j represents the nursing scheduling interference function of the jth nurse, Represents the minimum constant, with a value of 1×10 -3 .

[0046] Preferably, S42, traverse the nursing scheduling interference function Hs of all nurses every 60 seconds, and extract the actual average response time of each operation task of each nurse from the nursing record, mark the critical moment of nurse task overload, use the AI ​​model random forest to back-test the nursing scheduling interference function Hs of the current nurse, and take the 85% quantile as the nursing threshold F2, and perform nursing evaluation on the nursing scheduling interference function Hs of all nurses obtained in real time and the nursing threshold F2, and trigger the task diversion strategy based on the evaluation results. The specific evaluation content is as follows;

[0047] When the nursing scheduling interference function Hs of the jth nurse j When the value is greater than the nursing threshold F2, it indicates that the nurses’ task load is overloaded, and the task diversion strategy is triggered;

[0048] When the nursing scheduling interference function Hs of the jth nurse j When the nursing threshold F2 is less than or equal to the nursing threshold, it indicates that the nurse's task load is normal and priority is given to the nurse;

[0049] The task diversion strategy traverses all nurses j to find the nursing scheduling interference function Hs that satisfies the jth nurse. j The task load of the nurse with the nursing threshold F2 is overloaded, and the minimum allocation function Min (L ij +Cz ij ), the current task is transferred to the nurse j with the shortest path and the fastest task operation time;

[0050] The minimization function Min(L ij +Cz ij ) represents the path complexity Lj between the i-th patient and the j-th nurse ij and the task operation time Cz of the i-th patient and the j-th nurse ij The minimum value is taken, that is, the nurse j with the shortest path and the fastest operation is assigned to the i-th patient in the infusion completion set Ct.

[0051] Preferably, said S5 includes S51 and S52;

[0052] S51. After the task diversion strategy is executed, the nursing scheduling interference function Hs of all nurses is recalculated, and the standard deviation is calculated based on the nursing scheduling interference function Hs of all nurses. The global pressure index Osys is output and the task load of all nurses is globally analyzed.

[0053] The global pressure index Osys is calculated and output by the following algorithm formula:

[0054] ;

[0055] Where n represents the total number of nurses involved in scheduling, Represents the interference function of nursing scheduling for all nurses.

[0056] Preferably, S52, collect nurse dispatch records for the past 30 days, mark normal distribution and abnormal distribution, perform statistical analysis on the global pressure index Osys under these states, set the critical point where abnormalities begin to occur frequently as the pressure threshold Oth, and perform a secondary comparative evaluation of the global pressure index Osys obtained in real time and the pressure threshold Oth to determine the overall dispatch load balance. The specific evaluation content is as follows;

[0057] When the global pressure index Osys is less than the pressure threshold Oth, it indicates that the current scheduling is valid and the current scheduling is maintained;

[0058] When the global pressure indicator Osys ≥ the pressure threshold Oth, it indicates that the global scheduling is unbalanced. At this time, based on the current scheduling, the task diversion strategy continues to be executed until the scheduling is effective and the execution is stopped.

[0059] An AI-based medical assistant system for emergency internal medicine patients, including an infusion sensing module, an AI pre-processing module, an infusion time prediction module, a scheduling intervention module, and a comprehensive assessment module;

[0060] The infusion sensing module collects infusion physiological status data in real time by setting up a sensor group, and sets up a wireless communication network to transmit the infusion physiological status data to the AI ​​scheduling server;

[0061] The AI ​​preprocessing module extracts nursing behavior data in real time by accessing the nursing task platform in the AI ​​scheduling server, and preprocesses the nursing behavior data and infusion physiological status data to obtain a standard nursing behavior data set and a standard infusion physiological status data set;

[0062] The infusion time prediction module constructs an infusion time prediction model, extracts a standard infusion physiological state data set and inputs it into the infusion time prediction model, calculates and outputs the infusion completion time prediction value Tinf, and sets a preliminary evaluation mechanism to trigger the scheduling optimization mechanism;

[0063] The scheduling intervention module calculates and outputs the nursing scheduling interference function Hs based on the infusion completion time prediction value Tinf and the standard nursing behavior data set after triggering the scheduling optimization mechanism, sets the nursing threshold F2 for nursing evaluation, and executes the task diversion strategy based on the evaluation results;

[0064] After the task diversion strategy is executed, the comprehensive evaluation module calculates and outputs the global pressure index Osys based on the nursing scheduling interference function Hs of all current nurses, and sets the pressure threshold Oth for secondary comparative evaluation to determine the scheduling load balance.

[0065] The present invention provides an artificial intelligence-based medical assistant system and method for emergency internal medicine patients. It has the following beneficial effects:

[0066] (1) The method sets up a variety of sensor groups including a wristband pulse sensor, an infrared thermal imager, a drug physical database API and an intelligent infusion pump sensor in step S1 to collect infusion physiological state data including micro-pressure fluctuation curves, blood flow area heat map sequences, pharmacological characteristic data sets and infusion volume; and performs standardization and normalization processing through AI preprocessing strategy in step S2 to form a standard infusion physiological state data set; the infusion time prediction model constructed in step S3 uses the local venous permeability index A, the terminal venous return compression factor Ys, the drug solution molecule diffusion hysteresis constant P and the current remaining infusion volume V, combined with nonlinear attenuation modeling to achieve high-precision prediction of the patient's Tinf infusion completion time, thereby identifying the patient group whose infusion is about to be completed in advance, realizing accurate early warning and advance planning of the fluid replacement task, and significantly reducing the incidence of adverse events such as infusion emptying, blood return, and venous blockage.

[0067] (2) In step S4 of this method, after the preliminary evaluation mechanism triggers the scheduling optimization mechanism, the nursing scheduling interference function Hs is calculated based on the Tinf prediction results of the patients in the infusion completion set Ct and the standard nursing behavior data set, as an evaluation parameter reflecting the current task pressure of the j-th nurse. The nursing scheduling interference function Hs is compared and evaluated with the dynamically set nursing threshold F2. When the nursing scheduling interference function Hs>nursing threshold F2, the task diversion strategy is triggered. By traversing the minimization allocation function Min (L ij +Cz ij ), the current task is reallocated to nurse j with the shortest path and the lowest time consumption, so as to dynamically balance nursing resources without interrupting the original task process, improve the timeliness of nurse task response and the smoothness of overall nursing operations, especially during peak infusion periods, showing significant scheduling flexibility and improvement in nursing efficiency.

[0068] (3) In step S5, after the task diversion strategy is executed, the method recalculates the nursing scheduling interference function Hs for each nurse and calculates the global pressure index Osys based on it, that is, the standard deviation of the task pressure of all nurses; then compares and evaluates this index with the pressure threshold Oth obtained through historical scheduling record training to form a complete load balancing feedback loop. If the global pressure index Osys is less than the pressure threshold Oth, it means that the current scheduling resource configuration is reasonable and the nursing task distribution is balanced; if the global pressure index Osys is greater than or equal to the pressure threshold Oth, the task diversion strategy will automatically continue to be executed until the load difference is reduced. Through this strategy, the global distribution status of nursing tasks can be continuously tracked and optimized, effectively preventing problems such as delayed operation, service interruption or decreased nursing satisfaction caused by the overload of individual nurses' tasks, and significantly improving the stability and sustainability of emergency nursing scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a schematic diagram of the steps of the artificial intelligence-based medical assistant method for emergency internal medicine patients of the present invention;

[0070] Figure 2 This is a flow chart of the artificial intelligence-based medical assistant system for emergency internal medicine patients of the present invention;

[0071] Figure 3 This is a curve diagram of the change of the global pressure index Osys of the present invention. DETAILED DESCRIPTION

[0072] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0073] Example 1, please refer to Figure 1 and Figure 3 The present invention provides an artificial intelligence-based medical assistant method for emergency internal medicine patients. To achieve the above objectives, the present invention is implemented through the following technical solutions: comprising the following steps:

[0074] S1. By setting up a sensor group, real-time infusion physiological status data is collected, and a wireless communication network is set up to transmit the infusion physiological status data to the AI ​​scheduling server;

[0075] S2. Connect the nursing task platform to the AI ​​scheduling server, extract nursing behavior data in real time, and pre-process the nursing behavior data and infusion physiological status data to obtain a standard nursing behavior data set and a standard infusion physiological status data set;

[0076] S3. Build an infusion time prediction model, extract a standard infusion physiological state data set and input it into the infusion time prediction model, calculate and output the infusion completion time prediction value Tinf, and set up a preliminary evaluation mechanism to trigger the scheduling optimization mechanism;

[0077] S4. After the scheduling optimization mechanism is triggered, the nursing scheduling interference function Hs is calculated based on the predicted value of the infusion completion time Tinf and the standard nursing behavior data set, and the nursing threshold F2 is set to perform nursing evaluation, and the task diversion strategy is executed based on the evaluation results;

[0078] S5. After the task diversion strategy is executed, the global pressure index Osys is calculated and output based on the nursing scheduling interference function Hs of all current nurses, and the pressure threshold Oth is set for secondary comparative evaluation to determine the scheduling load balance.

[0079] In this embodiment, the method deploys multiple sensor groups to collect infusion physiological status data, and transmits it to the AI ​​scheduling server in real time via LoRa and MQTT wireless communication modules, achieving high-frequency, low-latency data collection and upload. Subsequently, the method acquires nurse behavior data by integrating with the nursing task platform, and performs data normalization, standardization, and structuring based on the AI ​​preprocessing strategy to generate a standard nursing behavior data set and a standard infusion physiological status data set, thereby establishing a unified feature expression method. A nonlinear infusion time prediction model is constructed using the standard infusion physiological status data set, and an individualized infusion completion time prediction value Tinf is output. A preliminary evaluation mechanism is constructed based on the prediction result. By comparing it with the preset infusion completion warning time threshold ΔTth, a high-risk patient set Ct whose infusion is about to be completed is promptly identified. When Ct exceeds the preset load threshold Nth, the scheduling optimization mechanism is automatically triggered, and step S4 is entered. At this point, the nurse scheduling interference function Hs is calculated by combining the predicted infusion completion time Tinf with the standard nursing behavior dataset. Nursing resource overload is determined based on the nursing threshold F2 learned by AI. A task diversion strategy is triggered for nurses who exceed the nursing threshold F2. By optimizing the minimum path complexity Lj and the task operation time Cz, low-pressure transfer and redistribution of patient tasks are achieved. Finally, in step S5, the nursing scheduling interference function Hs for each nurse is re-aggregated, and its standard deviation is calculated and output as the global pressure index Osys. This is used to analyze the balance of the overall task allocation and compare it with the pressure threshold Oth obtained through historical data training. This forms a global-level dynamic scheduling optimization feedback mechanism to ensure that the system scheduling achieves the optimal balance between efficiency and fairness. In summary, the present invention realizes closed-loop control of the entire process from infusion status perception, prediction, evaluation to dynamic task optimization by introducing multi-dimensional perception and AI-driven decision-making mechanism. It not only improves the scheduling efficiency and response timeliness of internal medicine emergency nursing tasks, but also significantly reduces infusion anomalies and nursing delays caused by task stacking and uneven resource allocation, and improves the intelligence level of the overall nursing process and the stability and safety of medical services.

[0080] Example 2, please refer to Figure 1 ,Specifically: S1 includes S11 and S12;

[0081] S11. Install a sensor group on the internal medicine emergency infusion patient and the infusion device to collect the infusion physiological status data of the internal medicine emergency infusion patient in real time;

[0082] The sensor suite includes a wristband pulse sensor, an infrared thermal imager, a pharmaceutical physical database API, and a smart infusion pump sensor;

[0083] Infusion physiological status data include micro-pressure fluctuation curve, blood flow area heat map sequence, pharmacological characteristic data set, total set volume Vset and injected volume Vinjected;

[0084] The micro-pressure fluctuation curve is obtained by having internal medicine emergency infusion patients wear a wristband pulse sensor, which is attached near the patient's venipuncture area. The micro-pressure fluctuation of the local tissue is detected every 5 seconds during the infusion process.

[0085] The blood flow area thermal map sequence is acquired by infrared thermal imaging combined with near-infrared reflective blood flow imaging (NIRS).

[0086] The pharmacological characteristic dataset automatically captures the drug code in the doctor's order through the drug physical database API, and extracts the pharmacological characteristic dataset based on the drug code. The pharmacological characteristic dataset includes the molecular weight M of the drug used by the i-th patient, the dynamic viscosity u of the solvent used by the i-th patient, and the molecular formula of the drug used by the i-th patient;

[0087] The set volume Vset and the injected volume Vinjected are collected and obtained through the intelligent infusion pump sensor;

[0088] S12. By connecting the network communication modules of all sensor groups to the intranet LoRa / MQTT gateway, the real-time collected infusion physiological status data will be wirelessly transmitted to the constructed AI scheduling server.

[0089] In this embodiment, the method constructs a complete set of infusion physiological status data perception and transmission mechanisms through step S1, wherein S11 realizes multi-dimensional collection of individual patient physiological status and pharmacological information by deploying multiple heterogeneous sensor groups on internal medicine emergency infusion patients and infusion equipment. Specifically, the wristband pulse sensor is attached near the venous puncture area and collects micro-pressure fluctuation curves at a high frequency of 5 seconds per time, effectively reflecting the changes in local venous permeability and elasticity during the infusion process; the infrared thermal imager combines with near-infrared reflective blood flow imaging (NIRS) technology to collect blood flow area heat map sequences, realizing the visualization of blood flow dynamics and thermal diffusion state in the infusion area; the drug physical database API is connected to the hospital pharmacy system to extract the molecular weight M, solvent dynamic viscosity u and drug molecular formula composition of each patient's corresponding drug in real time, forming a high-dimensional pharmacological feature data set; in addition, the intelligent infusion pump sensor obtains the set volume Vset and injection volume Vinjected in real time, providing key parameter support for subsequent infusion volume modeling. Subsequently, in S12, the pre-installed communication modules in each sensor group are used to wirelessly transmit the multi-dimensional infusion physiological status data to the constructed AI dispatch server in real time using the low-power wide-area network LoRa and the lightweight message transmission protocol MQTT. This design not only ensures the continuity and robustness of data transmission in the complex environment of the emergency department, but also effectively reduces data latency and system load through the edge-deployed network structure.

[0090] Example 3, please refer to Figure 1 , specifically: S2 includes S21 and S22;

[0091] S21. Set up a medical care platform API in the AI ​​scheduling server, integrate the AI ​​scheduling server with the nursing task platform, and extract nursing behavior data in real time;

[0092] Nursing behavior data includes the path complexity Lj between the i-th patient and the j-th nurse ij , the task operation time of the i-th patient and the j-th nurse Cz ij and the number of abnormal states of the i-th patient Yc i ;

[0093] The path complexity Lj between the i-th patient and the j-th nurse ij By using RFID tags worn by nurses and UWB indoor positioning, combined with the path node graph algorithm, the shortest time path between the nurse and the patient is calculated;

[0094] The time Cz required for the task operation of the i-th patient and the j-th nurse ij Obtained by automatically recording the difference between the task start time and the completion feedback time for regression analysis;

[0095] Number of abnormal states of patient i Yci The number of times the infusion blockage, backflow and alarm conditions are detected by the infusion pump is obtained;

[0096] S22. Construct an AI preprocessing strategy in the AI ​​scheduling server, and preprocess the acquired nursing behavior data and infusion physiological status data based on the AI ​​preprocessing strategy to obtain a standard nursing behavior data set and a standard infusion physiological status data set, respectively;

[0097] AI preprocessing strategies include filtering, image preprocessing, graph neural networks, machine algorithms, and AI-driven normalization.

[0098] The standard infusion physiological state data set includes the local venous permeability index A of the i-th patient i , terminal venous return compressibility factor Ys of patient i i , the diffusion hysteresis constant P of the drug solution molecules of the i-th patient i and the current remaining infusion volume V for the i-th patient i ;

[0099] The standard nursing behavior dataset includes the normalized path complexity Lj between the i-th patient and the j-th nurse ij , the task operation time of the i-th patient and the j-th nurse Cz ij and the number of abnormal states of the i-th patient Yc i

[0100] The local venous permeability index A of the i-th patient i The AI ​​analysis strategy is used to filter the micro-pressure fluctuation curve to remove high-frequency jitter. The filter waveform is decomposed by wavelet to extract the low-frequency rebound reaction curve. The molecular recovery time, pressure rise slope and relative amplitude are extracted based on the low-frequency rebound reaction curve. After the dimension is eliminated by AI-driven normalization processing, a nonlinear polynomial fitting model is constructed using a machine algorithm. The extracted molecular recovery time, pressure rise slope and relative amplitude are weightedly fitted to obtain the local venous permeability index A of the i-th patient. i The larger the value, the better the vascular elasticity and the lower the infusion resistance. The weights in the formula are obtained through AI analysis of the marked real vascular rebound levels for training;

[0101] The terminal venous return compressibility factor Ys of the i-th patient i Through image preprocessing in the AI ​​analysis strategy, sparse principal component analysis (SPCA) is used to perform background subtraction on the blood flow region heat map sequence, extracting the regional temperature gradient ▽T, temperature texture change bHT, and infrared region conduction delay time △tther. After eliminating the dimension through AI-driven normalization processing, a Bayesian network is used to establish a relationship between the temperature change pattern and the degree of blockage, and the output is a probabilistic blockage factor.

[0102] The diffusion hysteresis constant P of the drug solution molecules of the i-th patient i By extracting the acquired pharmacological feature data set, the graph neural network is used to extract the features of the drug molecular formula of the i-th patient's medication, and the drug diffusion coefficient D and the molecular structure polarity index C of the i-th patient's medication are predicted. Combined with the pharmacological feature data set, AI-driven normalization processing is used to eliminate the dimension, and then a machine algorithm is used to construct a ratio calculation formula for calculation. The specific calculation formula is: ;

[0103] The current remaining infusion volume V for patient i i By using machine algorithms to construct a difference model between the total set volume Vset and the injected volume Vinjected, the difference calculation output is obtained;

[0104] AI-driven normalization processing constructs an Autoencoder-style normalization network, inputs nursing behavior data and infusion physiological status data into the Autoencoder-style normalization network, performs preliminary normalization and standardization processing, and then performs secondary normalization and standardization processing after outputting the standard nursing behavior data set and the standard infusion physiological status data set to eliminate the dimensional influence of all parameters.

[0105] In this embodiment, the method, in S21, establishes a medical platform API within the AI ​​scheduling server to extract three core data types closely related to patient interactions from the nursing task platform in real time: path complexity, task operation time, and patient abnormality status records. Path complexity Lj is obtained using RFID tags worn by nurses and a UWB indoor high-precision positioning system, combined with a path node graph algorithm, quantifying the path cost between nurses and patients. Task operation time Cz is extracted using automated time recording and regression analysis to determine the objective duration of each nursing task. Abnormality status Yc is detected by the infusion pump in real time, implementing a high-sensitivity early warning mechanism. This process effectively establishes a digital model between "nursing resource utilization efficiency" and "task risk factors." Next, in S22, the present invention utilizes an AI preprocessing strategy to fuse and normalize the collected nursing behavior data with the infusion physiological status data. Specifically, this AI preprocessing strategy integrates multiple intelligent computing modules, including filtering, wavelet decomposition, image preprocessing, graph neural networks, and autoencoder-style normalization networks, to extract, clean, and normalize multidimensional data features. In terms of physiological status data processing, the micro-pressure fluctuation curve is filtered and wavelet decomposed to extract key parameters, followed by nonlinear fitting to generate the local venous permeability index A. Sparse principal component analysis (SPCA) and Bayesian networks are used to process blood flow heat maps, extract regional thermal responses, and calculate the terminal venous return compressibility factor (Ys). A graph neural network is used to analyze the molecular structure and pharmacological properties of the drug and calculate the diffusion hysteresis constant (P) of the drug solution. A difference model is constructed by combining the set volume and the infusion volume to determine the current remaining infusion volume (V). Regarding nursing behavior data, AI-driven normalization is used to unify the scale and output a standardized nursing behavior dataset. In summary, step S2, by constructing a multi-source heterogeneous data fusion and standardization mechanism centered on AI analysis, not only achieves a structured representation of traditional nursing behavior and patient physiological data, but also provides a unified, computable, and reliable input data foundation for subsequent predictive models and scheduling optimization algorithms. This significantly improves the system's intelligence and overall operational efficiency, while also enhancing the dynamic perception and precise response capabilities for nursing resource allocation and infusion status warnings.

[0106] Example 4, please refer to Figure 1 ,Specifically: S3 includes S31 and S32;

[0107] S31. Construct an infusion time prediction model using a machine algorithm, extract a standard infusion physiological state data set and input it into the infusion time prediction model, calculate and output a predicted value Tinf of the infusion completion time for each internal medicine emergency patient, and predict the infusion completion time for each internal medicine emergency patient;

[0108] The predicted value of infusion completion time Tinf is calculated and output by the following infusion time prediction model;

[0109] ;

[0110] Where, Tinf i Predicted value of the infusion completion time for the i-th patient, Represents the minimum constant, with a value of 1×10 -3 , e represents the exponential function;

[0111] in: Represents the venous condition correction coefficient, which is a coefficient that simulates the influence of the patient's individual venous condition on the actual flow rate. The larger the value, the smoother the infusion and the shorter the time. The mathematical meaning is: if the terminal venous return compression factor Ys of the i-th patient i If it approaches 0, it means the passage is very smooth; if it approaches 1, it means there is distal resistance. Physical meaning: this ratio controls the "difficulty of entry of unit liquid" and is used to correct the ideal flow rate;

[0112] Represents the composite index of the interaction between the drug solution and blood vessels. This index simulates the decrease in flow rate caused by the movement, diffusion, and hysteresis of the drug solution in the microcirculatory system. The smaller the index, the more severe the "flow rate limitation";

[0113] The denominator of the formula's calculation logic represents the "correction of the amount of liquid injected per unit time": traditionally, only the drip rate is considered, but here the vascular state and the characteristics of the drug solution are taken into account. The exponential term is a nonlinear control term, introducing a dynamic simulation of diffusion hysteresis, such as a fast drip rate in the early part of the drug solution and a slower drip rate in the latter part.

[0114] Specific example: Assume V i =100, A i =0.9, Ys i =0.3, P i =2.0;

[0115] Although it is set to drip out every hour, due to blood vessels and liquid medicine, the system is estimated to take 74 minutes.

[0116] S32. After obtaining the predicted infusion completion time Tinf for each patient, a preliminary evaluation mechanism is executed. The preliminary evaluation mechanism uses the infusion completion warning time threshold △Tth set by the user, and extracts the current system time tnow and the predicted infusion completion time Tinf for each patient for preliminary comparative evaluation. The infusion completion set Ct is then compiled for the emergency infusion patients who are about to complete the infusion. When the infusion completion set Ct exceeds the resource upper limit threshold Nth preset for the hospital resources, the scheduling optimization mechanism is triggered.

[0117] The infusion completion collection Ct is constructed in the following way;

[0118] ;

[0119] If the infusion completion set Ct is greater than the resource upper limit threshold Nth, the scheduling optimization mechanism is triggered;

[0120] If the infusion completion set Ct ≤ resource upper limit threshold Nth, it will not be triggered;

[0121] The significance of the infusion completion collection Ct is to traverse all patients receiving infusion and calculate the difference between the time when the patient completes the infusion and the current system time tnow, which is the remaining infusion time of the infusion patient. The infusion completion warning time threshold △Tth is set by the user, such as 5 minutes or 3 minutes. If the remaining infusion time of the infusion patient is lower than this time range, it means that the patient's infusion is about to be completed. At the same time, the infusion completion collection Ct and the resource upper limit threshold Nth are evaluated to analyze how many patients are about to complete the infusion at the current moment and trigger the scheduling optimization mechanism.

[0122] In this embodiment, the method establishes an AI-driven intelligent infusion task identification and scheduling pre-perception capability by constructing an infusion time prediction and early warning evaluation mechanism. Specifically, in S31, by extracting multiple key parameters from the standard infusion physiological status data set, including the patient's current remaining infusion volume V, local venous permeability index A, terminal venous return compression factor Ys, and drug molecule diffusion hysteresis constant P, and inputting them into the infusion time prediction model constructed by the machine algorithm, the infusion completion time Tinf of each patient is accurately predicted. The mathematical model adopted is jointly constructed based on the venous correction coefficient and the diffusion hysteresis factor in the form of a nonlinear exponential, which fully simulates the combined influence of vascular status and drug properties on flow rate changes. It is significantly better than the traditional static estimation method based on flow rate or drip rate, and can dynamically feedback the real physiological effects of drug deceleration in the late process, vascular blockage, etc., thereby achieving high-precision infusion time prediction. Further in S32, based on the difference between each patient's predicted infusion completion time Tinf and the current system time tnow, an infusion completion collection Ct is constructed, and a double comparison and evaluation is performed with the infusion completion warning time threshold △Tth and the resource upper limit threshold Nth set by the hospital. When the number of patients in the infusion completion collection Ct exceeds Nth, it means that a large number of infusions are about to end in a short period of time. The system immediately triggers the subsequent scheduling optimization mechanism to achieve early identification and response to peak loads. The core value of this mechanism is to transform passive waiting nursing tasks into active perception scheduling guidance, greatly reducing nursing resource conflicts caused by the concentrated arrival of tasks.

[0123] Example 5, please refer to Figure 1, specifically: S4 includes S41 and S42;

[0124] S41. After the preliminary evaluation mechanism triggers the scheduling optimization mechanism, the predicted infusion completion time Tinf of the patient in the infusion completion set Ct and the standard nursing behavior data set are extracted, and the nursing scheduling interference function Hs is output to analyze the pressure level of the task to be completed;

[0125] The nursing scheduling interference function Hs is calculated and output by the following algorithm formula;

[0126] ;

[0127] Where Hs j represents the nursing scheduling interference function of the jth nurse, Represents the minimum constant, with a value of 1×10 -3 ;

[0128] The urgency factor of the task for the i-th patient is represented. A larger value indicates that the task for this patient is more urgent and important. It is derived from a comprehensive analysis of the patient's expected end of infusion and abnormal status.

[0129] The denominator represents the “unit cost” for the nurse to complete this task, which is the cost of the path cost and the average processing time;

[0130] The ratio calculation is to analyze the weight of urgent tasks under unit action cost. Urgent tasks, short paths and fast operations have large ratios, which means strong pressure.

[0131] S42. Traverse the nursing scheduling interference function Hs of all nurses every 60 seconds, and extract the actual average response time of each nurse for each operation task from the nursing records, mark the critical moment when the nurse's task is overloaded, use the AI ​​model random forest to back-calculate the nursing scheduling interference function Hs of the current nurse, and take the 85% quantile as the nursing threshold F2. Perform a nursing evaluation on the nursing scheduling interference function Hs of all nurses obtained in real time and the nursing threshold F2, and trigger the task diversion strategy based on the evaluation results. The specific evaluation content is as follows;

[0132] When the nursing scheduling interference function Hs of the jth nurse j When the value is greater than the nursing threshold F2, it indicates that the nurses’ task load is overloaded, and the task diversion strategy is triggered;

[0133] When the nursing scheduling interference function Hs of the jth nurse j When the nursing threshold F2 is less than or equal to the nursing threshold, it indicates that the nurse's task load is normal and priority is given to the nurse;

[0134] The task diversion strategy traverses all nurses j to find the nursing scheduling interference function Hs that satisfies the jth nurse. jThe task load of the nurse with the nursing threshold F2 is overloaded, and the minimum allocation function Min (L ij +Cz ij ), the current task is transferred to the nurse j with the shortest path and the fastest task operation time;

[0135] The minimization function Min(L ij +Cz ij ) represents the path complexity Lj between the i-th patient and the j-th nurse ij and the task operation time Cz of the i-th patient and the j-th nurse ij The minimum value is taken, that is, the nurse j with the shortest path and the fastest operation is assigned to the i-th patient in the infusion completion set Ct.

[0136] In this embodiment, the method constructs a nursing scheduling interference function Hs by extracting the predicted infusion completion time Tinf for all patients nearing completion of an infusion from the infusion completion collection Ct and their corresponding standard nursing behavior datasets in S41. This function calculates the ratio of each patient's task urgency factor (including infusion completion prediction, abnormal status, etc.) to the nurse's current path complexity and task processing time, thereby quantitatively assessing the task pressure experienced by the nurse per unit of work cost. The constructed nursing scheduling interference function Hs not only considers the urgency of the task itself but also incorporates differences in nurses' task execution paths and efficiency, making the assessment results more realistic. In S42, the nursing scheduling interference function Hs is periodically traversed every 60 seconds for all nurses. Combined with an AI model, such as a random forest, task response time is back-tested from historical nursing records, and the 85th percentile is dynamically extracted as the nursing threshold F2. By comparing the real-time nursing scheduling interference function Hs with the nursing threshold F2, it is possible to determine whether the current nurse is overloaded. If the nursing scheduling interference function Hs exceeds the nursing threshold F2, the nurse's task overload is detected. This indicates that the nurse's task pressure is too high, and the system immediately initiates the task diversion strategy. The task diversion logic traverses all current nurses and reselects the nurse with the shortest path and fastest task processing speed to take over the task based on the weighted value of the minimized path complexity Lj and the task operation time Cz, ensuring that the task flows to the optimal execution path.

[0137] Example 6, please refer to Figure 1 and Figure 3 ,Specifically: S5 includes S51 and S52;

[0138] S51. After the task diversion strategy is executed, the nursing scheduling interference function Hs of all nurses is recalculated, and the standard deviation is calculated based on the nursing scheduling interference function Hs of all nurses. The global pressure index Osys is output and the task load of all nurses is globally analyzed.

[0139] The global pressure index Osys is calculated and output by the following algorithm formula;

[0140] ;

[0141] Where n represents the total number of nurses involved in scheduling, represents the interference function of all nurses’ nursing scheduling, i.e., the average value of all nurses’ task pressures;

[0142] The overall calculation logic of the formula is the square of a standard deviation, which is used to describe the degree of dispersion of the distribution of task loads of all nurses after the task diversion strategy is implemented.

[0143] S52. Collect nurse dispatch records from the past 30 days, mark normal and abnormal distributions, such as whether delays or nurse complaints occur, perform statistical analysis on the global pressure indicator Osys under these conditions, set the critical point where abnormalities begin to occur frequently as the pressure threshold Oth, and perform a secondary comparative evaluation of the global pressure indicator Osys obtained in real time with the pressure threshold Oth to determine the overall dispatch load balance. The specific evaluation content is as follows;

[0144] When the global pressure index Osys is less than the pressure threshold Oth, it indicates that the current scheduling is valid and the current scheduling is maintained;

[0145] When the global pressure indicator Osys ≥ the pressure threshold Oth, it indicates that the global scheduling is unbalanced. At this time, based on the current schedule, the task diversion strategy continues to be executed until the schedule is valid and then stops execution;

[0146] Specific example: Nurse A's nursing scheduling interference function Hs=2.1, B's nursing scheduling interference function Hs=0.9, and C's nursing scheduling interference function Hs=1.6;

[0147] =2.1+0.9+1.6 / 3=1.53;

[0148] Osys=1 / 3[(2.1-1.53) 2 +(0.9-1.53) 2 +(1.6-1.53) 2 ]=0.242;

[0149] If Oth=0.2, the global pressure index Osys ≥ the pressure threshold Oth, indicating that the tasks are unbalanced and the task diversion strategy continues to be executed.

[0150] In this embodiment, the method realizes macro analysis and closed-loop control of the overall nursing resource scheduling status by constructing a global pressure assessment mechanism after completing the task diversion strategy. In the specific implementation, S51 first recalculates the latest nursing scheduling interference function Hs for all nurses, performs standard deviation analysis based on these values, and outputs the global pressure index Osys. This indicator reflects the degree of dispersion of the distribution of the current nurse task pressure. The larger the standard deviation, the more unbalanced the task distribution, and the more likely it is that local nursing resources will be overloaded or the scheduling will be unbalanced. The global pressure index Osys, as a quantitative reflection of the health of the scheduling, provides a decision-making basis for subsequent scheduling strategies. In S52, in order to scientifically set the judgment criteria, a historical data comparison mechanism is introduced to collect and analyze the normal and abnormal distribution in the nurse scheduling records of the past 30 days, such as delays, complaints, etc. The critical point of the frequent abnormality of the global pressure index Osys is determined through data backtracking, and this point is set as the pressure threshold Oth. The system performs a secondary comparison and evaluation of the current global pressure indicator Osys against the pressure threshold Oth in real time. If the global pressure indicator Osys is less than the pressure threshold Oth, it indicates that the current task scheduling has reached a reasonable balance, and the system can stop diversion to maintain the status quo. If the global pressure indicator Osys is greater than or equal to the pressure threshold Oth, it indicates that a significant imbalance still exists, and task diversion should continue until the overall system load stabilizes and falls back to a safe range. The implementation of this strategy establishes a complete closed loop from "local task pressure identification" to "overall scheduling status" and then to "continuous iteration of dynamic strategies." It not only ensures local optimization in a short period of time, but also ensures continuous scheduling balance in the long term. By accurately comparing the global pressure indicator Osys with the pressure threshold Oth, it can dynamically respond to sudden nursing peaks, effectively reduce nurse load differences, minimize operational errors, and reduce nursing delays, thereby significantly improving the intelligent management level and overall operational efficiency of emergency nursing services.

[0151] Example 7, please refer to Figure 1 and Figure 2 , an AI-based medical assistant system for internal medicine emergency patients, including an infusion perception module, an AI preprocessing module, an infusion time prediction module, a scheduling intervention module, and a comprehensive evaluation module;

[0152] The infusion sensing module collects infusion physiological status data in real time by setting up a sensor group, and sets up a wireless communication network to transmit the infusion physiological status data to the AI ​​scheduling server;

[0153] The AI ​​preprocessing module extracts nursing behavior data in real time by connecting to the nursing task platform in the AI ​​scheduling server, and preprocesses the nursing behavior data and infusion physiological status data to obtain standard nursing behavior data sets and standard infusion physiological status data sets;

[0154] The infusion time prediction module constructs an infusion time prediction model, extracts the standard infusion physiological state data set and inputs it into the infusion time prediction model, calculates and outputs the infusion completion time prediction value Tinf, and sets a preliminary evaluation mechanism to trigger the scheduling optimization mechanism;

[0155] After triggering the scheduling optimization mechanism, the scheduling intervention module calculates and outputs the nursing scheduling interference function Hs based on the predicted value of infusion completion time Tinf and the standard nursing behavior dataset, sets the nursing threshold F2 for nursing evaluation, and executes the task diversion strategy based on the evaluation results;

[0156] After the task diversion strategy is executed, the comprehensive evaluation module calculates and outputs the global pressure index Osys based on the nursing scheduling interference function Hs of all current nurses, and sets the pressure threshold Oth for secondary comparative evaluation to judge the scheduling load balance.

[0157] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. An artificial intelligence-based medical assistant method for emergency internal medicine patients, characterized by: The following steps are involved: S1. By setting up a sensor group, real-time infusion physiological status data is collected, and a wireless communication network is set up to transmit the infusion physiological status data to the AI ​​scheduling server; S2. Connect the nursing task platform to the AI ​​scheduling server, extract nursing behavior data in real time, and pre-process the nursing behavior data and infusion physiological status data to obtain a standard nursing behavior data set and a standard infusion physiological status data set; S3. Build an infusion time prediction model, extract a standard infusion physiological state data set and input it into the infusion time prediction model, calculate and output the infusion completion time prediction value Tinf, and set up a preliminary evaluation mechanism to trigger the scheduling optimization mechanism; S4. After the scheduling optimization mechanism is triggered, the nursing scheduling interference function Hs is calculated based on the predicted value of the infusion completion time Tinf and the standard nursing behavior data set, and the nursing threshold F2 is set to perform nursing evaluation, and the task diversion strategy is executed based on the evaluation results; S5. After the task diversion strategy is executed, the global pressure index Osys is calculated and output based on the nursing scheduling interference function Hs of all current nurses, and the pressure threshold Oth is set for secondary comparative evaluation to determine the scheduling load balance.

2. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 1, characterized in that: Said S1 includes S11 and S12; S11. Install a sensor group on the internal medicine emergency infusion patient and the infusion device to collect the infusion physiological status data of the internal medicine emergency infusion patient in real time; The sensor group includes a wristband pulse sensor, an infrared thermal imager, a drug physical database API, and a smart infusion pump sensor; The infusion physiological state data includes micro-pressure fluctuation curve, blood flow area heat map sequence, pharmacological characteristic data set, total set volume Vset and injection volume Vinjected; The micro-pressure fluctuation curve is obtained by having an internal medicine emergency infusion patient wear a wristband pulse sensor, attached to the patient's venipuncture area, and detecting the micro-pressure fluctuation of the local tissue every 5 seconds during the infusion process; The blood flow area heat map sequence is collected and acquired by an infrared thermal imager fused with near-infrared reflective blood flow imaging NIRS; The pharmacological characteristic dataset automatically captures the drug code in the doctor's order through the drug physical database API, and extracts the pharmacological characteristic dataset based on the drug code, wherein the pharmacological characteristic dataset includes the molecular weight M of the drug used by the i-th patient, the dynamic viscosity u of the solvent used by the i-th patient, and the molecular formula of the drug used by the i-th patient; The set volume Vset and the injected volume Vinjected are collected and acquired by the intelligent infusion pump sensor; S12. By connecting the network communication modules of all sensor groups to the intranet LoRa / MQTT gateway, the real-time collected infusion physiological status data will be wirelessly transmitted to the constructed AI scheduling server.

3. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 2, characterized in that: Said S2 includes S21 and S22; S21. Set up a medical care platform API in the AI ​​scheduling server, integrate the AI ​​scheduling server with the nursing task platform, and extract nursing behavior data in real time; The nursing behavior data includes the path complexity Lj between the i-th patient and the j-th nurse ij , the task operation time of the i-th patient and the j-th nurse Cz ij and the number of abnormal states of the i-th patient Yc i ; The path complexity Lj between the i-th patient and the j-th nurse ij By using RFID tags worn by nurses and UWB indoor positioning, combined with the path node graph algorithm, the shortest time path between the nurse and the patient is calculated; The task operation time of the i-th patient and the j-th nurse is Cz ij Obtained by automatically recording the difference between the task start time and the completion feedback time for regression analysis; The number of abnormal states of the i-th patient Yc i The number of times the infusion blockage, backflow and alarm conditions are detected by the infusion pump is obtained; S22. Construct an AI preprocessing strategy in the AI ​​scheduling server, and preprocess the acquired nursing behavior data and infusion physiological status data based on the AI ​​preprocessing strategy to obtain a standard nursing behavior data set and a standard infusion physiological status data set, respectively; The AI ​​preprocessing strategy includes filtering, image preprocessing, graph neural networks, machine algorithms, and AI-driven normalization. The standard infusion physiological state data set includes the local venous permeability index A of the i-th patient i , terminal venous return compressibility factor Ys of patient i i , the diffusion hysteresis constant P of the drug solution molecules of the i-th patient i and the current remaining infusion volume V for the i-th patient i ; The standard nursing behavior dataset includes the normalized path complexity Lj between the i-th patient and the j-th nurse ij , the task operation time of the i-th patient and the j-th nurse Cz ij and the number of abnormal states of the i-th patient Yc i .

4. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 3, characterized in that: Said S3 includes S31 and S32; S31. Construct an infusion time prediction model using a machine algorithm, extract a standard infusion physiological state data set and input it into the infusion time prediction model, calculate and output a predicted value Tinf of the infusion completion time for each internal medicine emergency patient, and predict the infusion completion time for each internal medicine emergency patient; The infusion completion time prediction value Tinf is calculated and output by the following infusion time prediction model; ； Where, Tinf i Predicted value of the infusion completion time for the i-th patient, Represents the minimum constant, with a value of 1×10 -3 , e represents the exponential function.

5. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 4, characterized in that: S32. After obtaining the predicted infusion completion time Tinf for each patient, a preliminary evaluation mechanism is executed. The preliminary evaluation mechanism uses the user-set infusion completion warning time threshold ΔTth, extracts the current system time tnow and each patient's predicted infusion completion time Tinf for preliminary comparative evaluation, and aggregates the infusion completion set Ct for emergency infusion patients who are about to complete the infusion. When the infusion completion set Ct exceeds the resource upper limit threshold Nth preset for hospital resources, the scheduling optimization mechanism is triggered. The infusion completion collection Ct is constructed in the following manner; ； If the infusion completion set Ct is greater than the resource upper limit threshold Nth, the scheduling optimization mechanism is triggered; If the infusion completion set Ct ≤ resource upper limit threshold Nth, it will not be triggered.

6. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 5, characterized in that: Said S4 includes S41 and S42; S41. After the preliminary evaluation mechanism triggers the scheduling optimization mechanism, the predicted infusion completion time Tinf of the patient in the infusion completion set Ct and the standard nursing behavior data set are extracted, and the nursing scheduling interference function Hs is output to analyze the pressure level of the task to be completed; The nursing scheduling interference function Hs is calculated and output by the following algorithm formula: ； Where Hs j represents the nursing scheduling interference function of the jth nurse, Represents the minimum constant, with a value of 1×10 -3 .

7. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 6, characterized in that: S42. Traverse the nursing scheduling interference function Hs of all nurses every 60 seconds, and extract the actual average response time of each nurse for each operation task from the nursing records, mark the critical moment when the nurse's task is overloaded, use the AI ​​model random forest to back-calculate the nursing scheduling interference function Hs of the current nurse, and take the 85% quantile as the nursing threshold F2. Perform a nursing evaluation on the nursing scheduling interference function Hs of all nurses obtained in real time and the nursing threshold F2, and trigger the task diversion strategy based on the evaluation results. The specific evaluation content is as follows; When the nursing scheduling interference function Hs of the jth nurse j When the value is greater than the nursing threshold F2, it indicates that the nurses’ task load is overloaded, and the task diversion strategy is triggered; When the nursing scheduling interference function Hs of the jth nurse j When the nursing threshold F2 is less than or equal to the nursing threshold, it indicates that the nurse's task load is normal and priority is given to the nurse; The task diversion strategy traverses all nurses j to find the nursing scheduling interference function Hs that satisfies the jth nurse. j The task load of the nurse with the nursing threshold F2 is overloaded, and the minimum allocation function Min (L ij +Cz ij ), the current task is transferred to the nurse j with the shortest path and the fastest task operation time; The minimization function Min(L ij +Cz ij ) represents the path complexity Lj between the i-th patient and the j-th nurse ij and the task operation time Cz of the i-th patient and the j-th nurse ij The minimum value is taken, that is, the nurse j with the shortest path and the fastest operation is assigned to the i-th patient in the infusion completion set Ct.

8. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 6, characterized in that: Said S5 includes S51 and S52; S51. After the task diversion strategy is executed, the nursing scheduling interference function Hs of all nurses is recalculated, and the standard deviation is calculated based on the nursing scheduling interference function Hs of all nurses. The global pressure index Osys is output and the task load of all nurses is globally analyzed. The global pressure index Osys is calculated and output by the following algorithm formula: ； Where n represents the total number of nurses involved in scheduling, Represents the interference function of nursing scheduling for all nurses.

9. The artificial intelligence-based medical assistant method for emergency internal medicine patients according to claim 8, characterized in that: S52. Collect nurse dispatch records from the past 30 days, mark normal and abnormal distributions, perform statistical analysis on the global pressure indicator Osys under these conditions, set the critical point where abnormalities begin to occur frequently as the pressure threshold Oth, and perform a secondary comparative evaluation of the global pressure indicator Osys obtained in real time with the pressure threshold Oth to determine the overall dispatch load balance. The specific evaluation content is as follows; When the global pressure index Osys is less than the pressure threshold Oth, it indicates that the current scheduling is valid and the current scheduling is maintained; When the global pressure indicator Osys ≥ the pressure threshold Oth, it indicates that the global scheduling is unbalanced. At this time, based on the current scheduling, the task diversion strategy continues to be executed until the scheduling is effective and the execution is stopped.

10. An artificial intelligence-based medical assistant system for internal medicine emergency patients, applied to the artificial intelligence-based medical assistant method for internal medicine emergency patients according to any one of claims 1 to 9, characterized in that: It includes infusion perception module, AI preprocessing module, infusion time prediction module, scheduling intervention module and comprehensive evaluation module; The infusion sensing module collects infusion physiological status data in real time by setting up a sensor group, and sets up a wireless communication network to transmit the infusion physiological status data to the AI ​​scheduling server; The AI ​​preprocessing module extracts nursing behavior data in real time by accessing the nursing task platform in the AI ​​scheduling server, and preprocesses the nursing behavior data and infusion physiological status data to obtain a standard nursing behavior data set and a standard infusion physiological status data set; The infusion time prediction module constructs an infusion time prediction model, extracts a standard infusion physiological state data set and inputs it into the infusion time prediction model, calculates and outputs the infusion completion time prediction value Tinf, and sets a preliminary evaluation mechanism to trigger the scheduling optimization mechanism; The scheduling intervention module calculates and outputs the nursing scheduling interference function Hs based on the infusion completion time prediction value Tinf and the standard nursing behavior data set after triggering the scheduling optimization mechanism, sets the nursing threshold F2 for nursing evaluation, and executes the task diversion strategy based on the evaluation results; After the task diversion strategy is executed, the comprehensive evaluation module calculates and outputs the global pressure index Osys based on the nursing scheduling interference function Hs of all current nurses, and sets the pressure threshold Oth for secondary comparative evaluation to determine the scheduling load balance.

Images