An elderly critical MDT dynamic grading and dispatching and collaborative decision support method and system

Abstract

This invention provides a method and system for dynamic hierarchical scheduling and collaborative decision support of multidisciplinary teams (MDTs) in elderly patients with acute and critical illnesses. The steps are as follows: Collect and preprocess multimodal data on the patient's physiological indicators, voice, video, frailty state, and environment; based on the preprocessed data, combined with the frailty index and sentiment analysis, determine the patient's risk level through an assessment algorithm integrating long short-term memory networks and a large medical model; according to the risk level and suspected emergency type, use dynamic weight matching to select the optimal MDT combination, and perform resource scheduling based on multi-objective optimization to generate consultation notifications; utilize consortium blockchain nodes and edge computing nodes deployed in various departments to achieve real-time data sharing and solution collaboration among multiple disciplines, and handle solution conflicts through RBR and CBR reconciliation mechanisms. This solves the problems of low efficiency in manual coordination, delayed information transmission, and inaccurate triage and resource scheduling in existing MDT models for elderly patients with acute and critical illnesses, leading to insufficient timeliness and accuracy in treatment.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and more specifically, to a method and system for dynamic hierarchical scheduling and collaborative decision support of multidisciplinary team (MDT) for elderly patients with acute and critical illnesses. Background Technology

[0002] With the rapid development of medical informatization and artificial intelligence technologies, the core role of the Intensive Care Unit (ICU) in the treatment of critically ill patients is becoming increasingly prominent. Elderly patients with acute and critical illnesses often have frequent fluctuations in physiological indicators, complex and changeable conditions, and are often accompanied by multiple underlying diseases, which places higher demands on the accuracy and timeliness of the Multidisciplinary Team (MDT).

[0003] Currently, the traditional MDT (Multidisciplinary Team) model still dominates the treatment of acute and critical illnesses in the elderly. It primarily relies on the clinical experience of medical staff for manual coordination, case information transmission, and treatment plan formulation. This not only places a tremendous workload on medical staff but is also prone to delays in treatment due to human error and untimely information transmission, failing to meet the needs of elderly patients for "rapid diagnosis and precise intervention." Meanwhile, while existing intelligent medical management systems have partially incorporated data entry and triage functions, they are mostly based on single-dimensional data collection and fixed departmental triage, still requiring extensive manual data entry. This results in low clinical efficiency, an inability to achieve real-time monitoring and risk warning of waiting elderly patients, and difficulty in adapting to the triage challenges arising from differences in departmental naming, overlapping business areas, and the complexity of physician subspecialties across different hospitals. Consequently, resource allocation efficiency is low, and the accuracy of hierarchical scheduling is insufficient. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for dynamic hierarchical scheduling and collaborative decision support of multidisciplinary teams (MDTs) for elderly patients with acute and critical illnesses. This aims to solve the problems of insufficient timeliness and accuracy of treatment caused by low efficiency of manual coordination, delayed information transmission, and inaccurate system triage and resource scheduling in the existing collaborative group model for multidisciplinary comprehensive treatment of elderly patients with acute and critical illnesses.

[0005] This invention is achieved through the following technical solution: A dynamic hierarchical scheduling and collaborative decision support method for MDT (Multidisciplinary Team) in elderly patients with acute and critical illness includes the following steps: Multimodal data of patients are collected and preprocessed; wherein, the multimodal data includes physiological index data, voice complaint data, video behavior data, frailty state data and environmental monitoring data; Based on the preprocessed multimodal data, combined with the patient's frailty index and emotional state analysis results, the risk level of the patient is output by an assessment algorithm that integrates long short-term memory network and medical big model. Based on the patient's risk level and suspected emergency type, a dynamic weighting algorithm is used to match the optimal MDT combination, and a multi-objective optimization algorithm is used for resource scheduling to generate a consultation notice and send it to the corresponding department. Based on consortium blockchain nodes and edge computing nodes deployed in various departments, real-time data sharing and collaborative treatment plans among doctors from multiple disciplines are realized, and conflicts in treatment plans are handled through the reconciliation mechanism of RBR and CBR.

[0006] Optionally, the specific process of collecting the patient's multimodal data and preprocessing the multimodal data is as follows: By deploying medical IoT sensors, voice acquisition terminals, video surveillance equipment, and environmental monitoring equipment in the wards, real-time data on patients' physiological indicators, voice complaints, video behavior and frailty status, as well as environmental monitoring data of the wards are collected. The collected multimodal data is transmitted in real time to the edge computing nodes of the corresponding departments. In the edge computing nodes, the isolated forest algorithm is used to detect and remove outliers from the physiological index data, and to perform filtering and noise reduction. The normalization method is used to unify the physiological indicators to standard dimensions. The voice complaint data is subjected to speech recognition, converted into text data, and emotional feature vectors are extracted. The video behavior data is subjected to keyframe extraction and behavioral feature analysis to identify the patient's behavioral state and posture change trajectory. Based on the preset frailty assessment scale, the frailty status data is quantitatively scored to generate an geriatric frailty index. The environmental monitoring data was time-stamped and the sensors were calibrated. The processed multimodal data was then integrated into a unified structured data table according to the patient identifier, thus completing the preprocessing of the multimodal data.

[0007] Optionally, the specific process of outputting the patient's corresponding risk level based on the preprocessed multimodal data, combined with the patient's frailty index and emotional state analysis results, and through an assessment algorithm that integrates long short-term memory networks and a large medical model, is as follows: The physiological index time series data from the preprocessed multimodal data are input into the long short-term memory network model to extract physiological time series feature vectors; The preprocessed speech and text data is input into a pre-trained medical model to extract key chief complaint feature vectors. Based on the preprocessed voice complaint data, the emotional state coefficient is extracted through the voice emotion analysis model. Based on the preprocessed frailty data, the frailty index of the elderly is calculated according to the preset frailty assessment scale. The physiological time sequence feature vector, key chief complaint feature vector, emotional state coefficient and geriatric frailty index are concatenated to form a multi-dimensional fused feature vector. The multi-dimensional fused feature vector is input into the Bayesian fusion model, and the risk score is output by weighting it using preset weight coefficients. The risk score is mapped to a corresponding risk level based on the threshold range in which it falls; the risk levels include high risk, medium risk, and low risk.

[0008] Optionally, the specific process of matching the optimal MDT combination using a dynamic weighting algorithm based on the patient's corresponding risk level and suspected emergency type, and performing resource scheduling based on a multi-objective optimization algorithm to generate a consultation notification and send it to the corresponding department is as follows: Obtain the current patient's multi-dimensional fused feature vector, risk level, and suspected emergency type; Based on preset scheduling rules, multiple dynamic weighting factors required for MDT scheduling are obtained; among them, the dynamic weighting factors include: the real-time workload factor of the target department, the matching degree factor between the candidate doctor's subspecialty and the suspected emergency type, the candidate doctor's historical consultation success rate factor, and the candidate doctor's historical response time compliance rate factor. Based on the real-time values ​​of each dynamic weight factor, the real-time weight values ​​of each factor are determined through a dynamic weight calculation model. The multi-dimensional fusion feature vector of the current patient, the suspected emergency type, and the real-time weight value are input into the dynamic weight KNN algorithm model. The dynamic weight KNN algorithm model calculates the weighted similarity between the current patient's features and the features of historical consultation cases based on the real-time weight value, and retrieves K historical cases with similarity greater than the preset similarity value from the historical case database. Extract successful MDT combination information corresponding to K historical cases as an initial candidate MDT resource pool for matching; A multi-objective optimization function is constructed with the optimization objectives of minimizing response time, maximizing system resource utilization, and maximizing the predicted consultation success rate. The candidate MDT resource pool, the response time threshold constraints corresponding to the patient risk level, and the real-time load status of each department are input into the multi-objective optimization scheduler based on the NSGA-Ⅲ algorithm. Under the premise of satisfying the response time threshold constraint, the multi-objective optimization scheduler solves the multi-objective optimization function, outputs the Pareto optimal solution set, and selects the final scheduling scheme from the optimal solution set. The final scheduling scheme determines the specific departments and doctors participating in the consultation. Based on the final scheduling plan, a consultation notification is generated and sent to the terminals of the corresponding departments and doctors; The system monitors the load index of each department in real time. When the load index of a department exceeds the preset warning threshold, it triggers a request for cross-departmental or cross-hospital resource coordination and scheduling.

[0009] Optionally, the specific process of achieving real-time data sharing and collaborative treatment plans among doctors from multiple disciplines based on consortium blockchain nodes and edge computing nodes deployed in various departments is as follows: A hybrid collaborative network is constructed based on consortium blockchain nodes deployed in emergency departments, intensive care units, geriatric departments, and specialized departments, and edge computing nodes at nurse stations in each department. Through edge computing nodes, the patient's current diagnosis and treatment data, vital signs and test results are synchronized in real time from the hospital information system and electronic medical record system and uploaded to the corresponding department's consortium blockchain node; Consortium blockchain nodes complete data on-chain notarization based on the Proof-of-Authority (PoA) consensus mechanism, and encrypt real-time data during transmission using the national cryptographic algorithm SM4. Doctors from various departments can log in to collaborative terminals deployed on edge computing nodes and edit treatment plans, add treatment opinions, or make annotations online based on real-time synchronized patient data. The collaborative terminals will push the doctors' editing operations to other collaborative terminals of the same department in real time.

[0010] Optionally, the specific process of handling conflicts in the treatment plan through the reconciliation mechanism of RBR and CBR is as follows: When at least two doctors are detected to have editing conflicts regarding the treatment plan for the same patient, the conflict resolution process is triggered. The RBR mechanism matches the modified plans submitted by each doctor with a pre-set clinical diagnosis and treatment rule base, and calculates the fit score of each plan with the current diagnosis and treatment guidelines. Using the CBR mechanism, K historical cases with similar characteristics to the current patient's condition are retrieved from the historical consultation case database, the corresponding successful treatment plans are extracted, and the similarity score between the current modified plan and the historical successful plan is calculated. Based on the professional title information of each submitting doctor, a corresponding professional title weight coefficient is assigned; The comprehensive reconciliation score for each conflict resolution plan is obtained by weighting the fit score, similarity score, and professional title weight coefficient. The solution with the highest overall reconciliation score is selected as the recommended reconciliation solution and pushed to all participating doctors' terminals for confirmation or further revision. If the recommended solution receives majority confirmation, it is updated as the current execution solution, and the reconciliation process is recorded in the consortium blockchain for evidence storage.

[0011] Optionally, based on real-time physiological data of patients, the Kalman filter algorithm is used to predict the trend of changes in patients' physiological indicators over a future period of time. When the prediction results indicate a risk of abnormal indicators, an MDT combination adjustment plan is generated by combining the current MDT collaboration score with the disease prediction results; wherein, the MDT combination adjustment plan includes adding a specialist or replacing an existing specialist. Based on the MDT combination adjustment plan, update the MDT combination and notify relevant doctors to participate in the consultation; The patient's risk level is updated based on the adjusted MDT collaboration data and the patient's real-time physiological indicators.

[0012] Based on the same inventive concept, this invention also provides a dynamic hierarchical scheduling and collaborative decision support system for elderly critical care MDTs, used to implement the aforementioned dynamic hierarchical scheduling and collaborative decision support method for elderly critical care MDTs, including: The data acquisition and preprocessing module is used to collect patients' physiological index data, voice complaint data, video behavior data, frailty state data and environmental monitoring data, and to perform real-time transmission, anomaly detection, feature extraction and structured integration of the multimodal data. The risk assessment and classification module, connected to the data acquisition and preprocessing module, is used to receive preprocessed multimodal data, combine the results of geriatric frailty index and emotional state analysis, and output the patient's corresponding risk level through an assessment algorithm that integrates long short-term memory network and medical big model. The dynamic scheduling module, connected to the risk assessment and grading module, is used to match the optimal MDT combination using the dynamic weighted KNN algorithm based on the risk level and suspected emergency type, and to perform resource scheduling based on the NSGA-Ⅲ multi-objective optimization algorithm, generate consultation notices and send them to the corresponding departments. The collaborative decision-making module is connected to the dynamic scheduling module and the data acquisition and preprocessing module, respectively. It includes consortium blockchain nodes and edge computing nodes deployed in various departments. It is used to realize real-time data sharing and collaborative treatment plans among doctors from multiple disciplines, and to handle plan conflicts through the RBR and CBR reconciliation mechanism. The dynamic adjustment module is connected to the collaborative decision-making module and the risk assessment and grading module, respectively. It is used to predict the trend of changes in the patient's physiological indicators based on the Kalman filter algorithm, generate an adjustment plan in combination with the current MDT collaborative score, and update the MDT combination and risk level to form a closed-loop feedback.

[0013] Based on the same inventive concept, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described dynamic hierarchical scheduling and collaborative decision support method for MDT in elderly patients with acute and critical illness.

[0014] Based on the same inventive concept, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for dynamic hierarchical scheduling and collaborative decision support of MDT for acute and critical illness in the elderly.

[0015] The technical solution of the present invention has at least the following advantages and beneficial effects: By integrating multimodal data such as physiological indicators, voice complaints, video behavior, frailty status, and environmental monitoring, and combining them with the analysis of the frailty index and emotional state, a comprehensive and dynamic assessment of the patient's condition can be achieved. The intelligent algorithm that integrates long short-term memory networks and medical big data models can deeply mine the temporal characteristics of data and their correlation with medical knowledge, thereby improving the accuracy of risk classification and early warning capabilities.

[0016] Based on risk level and suspected acute illness type, the system uses a dynamic weighting algorithm to match the optimal MDT combination and performs resource scheduling based on multi-objective optimization. This effectively overcomes the scheduling inaccuracy caused by differences in department naming and business overlap in the traditional manual triage model. The system automatically generates and sends consultation notices, which greatly reduces the coordination burden of medical staff and shortens the treatment response time. It is especially suitable for elderly patients with complex conditions and multiple coexisting diseases who are critically ill.

[0017] By deploying consortium blockchain nodes and edge computing nodes, real-time and reliable data sharing and collaborative decision-making among multiple disciplines can be achieved while ensuring data security and privacy. Edge computing supports low-latency processing of bedside data, and the consortium blockchain ensures that medical records are traceable and tamper-proof, thereby breaking down departmental information silos and improving the informatization level of MDT collaboration and the continuity of treatment processes.

[0018] Introducing a reconciliation mechanism based on rule-based reasoning (RBR) and case-based reasoning (CBR) can automatically handle potential conflicts of opinion among multidisciplinary experts in treatment plans. It integrates clinical guidelines and historical similar cases to generate consistent recommended plans, assisting doctors in making more balanced and scientific group decisions and reducing diagnostic and treatment biases caused by subjective differences. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the dynamic hierarchical scheduling and collaborative decision support method for MDT in elderly patients with acute and critical illnesses according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the dynamic hierarchical scheduling and collaborative decision support system for MDT in elderly patients with acute and critical illnesses according to an embodiment of the present invention. Detailed Implementation

[0020] The following is a detailed description of the embodiments, in conjunction with the accompanying drawings.

[0021] Reference Figure 1 A dynamic hierarchical scheduling and collaborative decision support method for MDT (Multidisciplinary Team) in elderly patients with acute and critical illnesses includes the following steps: Step 1: Collect multimodal data from the patient and preprocess the multimodal data; wherein, the multimodal data includes physiological index data, voice complaint data, video behavior data, frailty state data and environmental monitoring data.

[0022] In some embodiments, the specific process of collecting the patient's multimodal data and preprocessing the multimodal data is as follows: By deploying medical IoT sensors, speech acquisition terminals, video surveillance equipment, and environmental monitoring equipment in patient wards, real-time data on patients' physiological indicators, voice complaints, video behavior, and frailty status, as well as environmental monitoring data, are collected. Medical IoT sensors can be multi-parameter monitors (such as ECG, blood oxygen, blood pressure, and body temperature sensors) supporting 5G transmission deployed in emergency rooms, ICUs, and geriatric wards, collecting 12 core physiological indicators at a frequency of 1 second per acquisition. Speech acquisition terminals can be intelligent voice terminals (with built-in noise-canceling microphones) that support dialect adaptation, collecting real-time audio of patients' complaints and supporting Chinese, English, and common dialects (such as Cantonese and Sichuan-Chongqing dialects). Video surveillance equipment can be low-power high-definition cameras (resolution ≥1080P, frame rate ≥30fps) to capture patients' facial micro-expressions and body movements (such as curling up, trembling, and groaning). Medical staff can input data on-site or patients can self-assess the 10 indicators of the Frailty Scale (FRAIL) via tablets or handheld terminals. Environmental monitoring equipment can include temperature and humidity sensors, noise sensors, and light sensors to collect real-time data on the ward environment.

[0023] The collected multimodal data is transmitted in real time to the edge computing nodes of the corresponding departments. At these nodes, the isolated forest algorithm is used to detect and remove outliers from the physiological indicator data, followed by filtering and noise reduction. Normalization is then used to standardize the physiological indicators to the correct dimensions. Voice complaints are processed through speech recognition, converted to text, and emotional feature vectors are extracted. Keyframe extraction and behavioral feature analysis are performed on video behavior data to identify the patient's behavioral state and posture changes. Based on a pre-set frailty assessment scale, frailty status data is quantified and scored to generate an geriatric frailty index. All data acquisition devices are connected via 5GCPE (Customer Premise Equipment) or the hospital intranet, with data transmitted in real time to the edge computing nodes of the corresponding departments. The transmission protocol uses MQTT (Message Queuing Telemetry Transport) + SSL (Secure Sockets Layer) encryption. Physiological data is uploaded at a frequency of 1 second, and audio and video data are streamed after H.265 encoding and compression, with end-to-end latency controlled within 50ms. It can integrate Isolation Forest and Attention mechanisms to handle outliers and missing values. Isolation Forest detects outliers in physiological indicators, while Attention mechanisms locate missing key features (such as heart rate and blood oxygen). If the missing value rate is ≤10%, it uses Generative Adversarial Network (GAN) to complete the data based on time-series data of patients of the same age and with the same underlying diseases; if the missing value rate is >10%, it triggers a real-time data entry reminder from the healthcare provider. The outlier scoring formula for Isolation Forest is shown below:

[0024] in, This indicates outlier values ​​in the physiological indicator data to be tested; The data are the physiological indicators to be tested; For the sample size, The average path length; This represents the average path length under a standard normal distribution. Let be Euler's constant; when When the corresponding physiological indicator data is abnormal, it is determined that the data is abnormal. The loss function of the GAN missing value completion generator is shown in the following formula:

[0025] in, This represents the generator's total loss value. The smaller this value, the higher the quality of the completed data generated by the generator. For mathematical expectation; This represents the latent spatial noise vector, which serves as the input seed for the generator and is used to initiate the data generation process. The probability distribution representing the potential spatial noise vector can be either a standard normal distribution or a uniform distribution. For the discriminator function; The data used to complete the generator represents the generator's performance under given conditions. and latent space noise vector In this case, the physiological indicator data generated to fill in missing values; To harmonic hyperparameters, a scalar greater than 0 is used to balance the relative importance between the adversarial loss term and the mean squared error loss term in the loss function; Mean squared error loss is used to measure the data completed by the generator. With real and complete data The differences between them.

[0026] The environmental monitoring data was time-stamped and the sensors were calibrated. The processed multimodal data was then integrated into a unified structured data table according to the patient identifier, thus completing the preprocessing of the multimodal data.

[0027] Step 2: Based on the preprocessed multimodal data, combined with the patient's frailty index and emotional state analysis results, the risk level of the patient is output through an assessment algorithm that integrates long short-term memory networks and a medical big data model.

[0028] In some embodiments, the specific process of outputting the patient's risk level based on preprocessed multimodal data, combined with the patient's frailty index and emotional state analysis results, and through an assessment algorithm that integrates long short-term memory networks and a large medical model, is as follows: The physiological indicator time-series data from the preprocessed multimodal data are input into a long short-term memory network model to extract physiological time-series feature vectors. The preprocessed speech text data is input into a pre-trained medical large-scale model to extract key chief complaint feature vectors. Based on the preprocessed speech chief complaint data, an emotional state coefficient is extracted using a speech emotion analysis model. Based on the preprocessed frailty state data, an aging frailty index is calculated according to a preset frailty assessment scale. The physiological time-series feature vector, key chief complaint feature vector, emotional state coefficient, and aging frailty index are concatenated to form a multi-dimensional fusion feature vector. The multi-dimensional fusion feature vector is input into a Bayesian fusion model, and a risk score is output by weighting the data using preset weight coefficients. The risk score is mapped to the corresponding risk level according to the threshold range in which it falls. The risk levels include high risk, medium risk, and low risk.

[0029] The frailty index is quantitatively calculated based on 10 preset frailty assessment indicators (such as weight loss, weakened grip strength, etc.), using the following formula:

[0030] in, The frailty index represents the range of values ​​for the elderly. ; This indicates the number of frailty assessment indicators that the patient meets; This indicates the preset total number of frailty assessment indicators (fixed at 10 items, corresponding to the Frailty Scale for the Aged).

[0031] The risk score is output by weighted fusion of multi-dimensional features using a Bayesian fusion model, as shown in the following formula:

[0032] in, This is the risk score, and its value range is... ; The weighted output of the physiological temporal feature vector extracted by the Long Short-Term Memory (LSTM) network (after normalization, with values ​​ranging from...). ); The weighted output of the key chief complaint feature vector extracted for pre-trained large medical models (such as Med-LLaMA) (after normalization, with values ​​ranging from...) ); The emotional state coefficient is extracted by the speech emotion analysis model (the value range is...). 1 indicates extreme pain. , , and The weight coefficients for each item are determined through grid search optimization to balance the contribution of each feature to risk assessment (which can be taken as...). , , , ).

[0033] Based on risk score The threshold range is mapped to the corresponding risk level: when At that time, the risk level was determined to be high-risk; when At that time, the risk level was determined to be medium risk; when At that time, the risk level was determined to be low.

[0034] Step 3: Based on the patient's risk level and suspected emergency type, use a dynamic weighting algorithm to match the optimal MDT combination, and use a multi-objective optimization algorithm to schedule resources, generate a consultation notice and send it to the corresponding department.

[0035] In some embodiments, the specific process of matching the optimal MDT combination using a dynamic weighting algorithm based on the patient's corresponding risk level and suspected emergency type, and performing resource scheduling based on a multi-objective optimization algorithm to generate a consultation notification and send it to the corresponding department is as follows: The system acquires the current patient's multi-dimensional fusion feature vector, risk level, and suspected emergency type; based on preset scheduling rules, it acquires multiple dynamic weighting factors required for MDT scheduling; these dynamic weighting factors include: the target department's real-time workload factor, the matching degree factor between the candidate doctor's subspecialty and the suspected emergency type, the candidate doctor's historical consultation success rate factor, and the candidate doctor's historical response time achievement rate factor; based on the real-time values ​​of each dynamic weighting factor, the system determines the real-time weight value of each factor through a dynamic weighting calculation model; the calculation formula for the real-time weight is shown below:

[0036] in, Indicates in At that moment, the The real-time weight values ​​obtained from the dynamic weight factor calculation are used for subsequent weighted similarity calculations; For the first The initial weights of the dynamic weighting factors; Indicates in At that moment, the The normalized real-time value of the dynamic weight factor, with a value range of [value range missing]. This reflects the actual state of the factor at the current moment (e.g., the workload of the department, the matching degree of doctors, etc.). For the loop variable in the summation symbol, and The meaning is the same, representing the first The dynamic weighting factor is included in the summation of the denominator. Iterate from 1 to 4, covering all four weight factors; These correspond to the target department's real-time workload factor, subspecialty matching factor, historical consultation success rate factor, and historical response time compliance rate factor, respectively.

[0037] The multi-dimensional fused feature vector of the current patient, the suspected emergency type, and the real-time weight value are input into the dynamic weight KNN algorithm model. The dynamic weight KNN algorithm model calculates the weighted similarity between the features of the current patient and the features of historical consultation cases based on the real-time weight value, and retrieves K historical cases from the historical case database with a similarity greater than a preset similarity value; the weighted similarity calculation formula is as follows:

[0038] in, Weighted similarity; This is a multi-dimensional fusion feature vector for the current patient; The feature vector of historical cases; For the first Cosine similarity along the feature dimension; Successful MDT (Multidisciplinary Team) combinations corresponding to K historical cases are extracted to form an initial candidate MDT resource pool. A multi-objective optimization function is constructed with the optimization objectives of shortest response time, highest system resource utilization, and highest estimated consultation success rate. The candidate MDT resource pool, response time threshold constraints corresponding to patient risk levels, and real-time load status of each department are input into a multi-objective optimization scheduler based on the NSGA-III algorithm. The multi-objective optimization scheduler solves the multi-objective optimization function while satisfying the response time threshold constraints, outputs a Pareto optimal solution set, and selects the final scheduling scheme from the optimal solution set. The final scheduling scheme determines the specific departments and doctors participating in the consultation. The multi-objective optimization function is shown in the following equation:

[0039] in, The first optimization objective function aims to minimize the total system response time, which is the sum of the total time from issuing a consultation notification to all scheduled doctors confirming their participation. Indicates the first The estimated response time for each doctor; This indicates the total number of doctors who were ultimately dispatched. The index variable is used for summation, representing the index of the first index. The doctor who was dispatched ( From 1 to ); The second optimization objective function aims to minimize the total system resource load rate, which is the sum of the ratios of the real-time workload to the capacity of all departments where the scheduled doctors are located, in order to achieve balanced resource utilization and avoid overload in some departments. Indicates the first The real-time workload of a doctor's target department can be understood as the number of emergency consultations currently being handled by that department, the doctor's workload index, or the length of the queue of cases to be processed. Indicates the first The load capacity of a doctor's target department represents the maximum workload that the department can handle per unit of time, such as the total number of doctors in the department and the maximum number of concurrent tasks. (Ratio) This is the real-time workload rate of the department; The third optimization objective function aims to maximize the predicted overall consultation success rate, which is the sum of probabilities predicted for the selected MDT combination to ensure successful consultation for the current patient based on historical data and other factors; representing the selection of the third The contribution of a doctor's participation in a consultation to the estimated success rate can be calculated based on factors such as the matching degree between the doctor's subspecialty and the patient's suspected emergency type, the doctor's historical consultation success rate, and professional title.

[0040] According to the final scheduling plan, a consultation notification is generated and sent to the terminals of the corresponding departments and doctors; the load index of each department is monitored in real time, and when the load index of a certain department exceeds the preset warning threshold, a cross-department or cross-hospital resource collaborative scheduling request is triggered.

[0041] Step 4: Based on the consortium blockchain nodes and edge computing nodes deployed in various departments, realize real-time data sharing and treatment plan collaboration among doctors from multiple disciplines, and handle conflicts in treatment plans through the reconciliation mechanism of RBR (Rule-Based Reasoning) and CBR (Case-Based Reasoning).

[0042] In some embodiments, the specific process of realizing real-time data sharing and collaborative treatment plans among doctors from multiple disciplines based on consortium blockchain nodes and edge computing nodes deployed in various departments is as follows: A hybrid collaborative network is constructed based on consortium blockchain nodes deployed in emergency departments, intensive care units, geriatric departments, and specialized departments, and edge computing nodes at nurse stations in each department. Through edge computing nodes, the patient's current diagnosis and treatment data, vital signs and test results are synchronized in real time from the hospital information system and electronic medical record system and uploaded to the corresponding department's consortium blockchain node; Consortium blockchain nodes complete data on-chain notarization based on the Proof-of-Authority (PoA) consensus mechanism, and encrypt real-time data during transmission using the national cryptographic algorithm SM4. Doctors from various departments can log in to collaborative terminals deployed on edge computing nodes and edit treatment plans, add treatment opinions, or make annotations online based on real-time synchronized patient data. The collaborative terminals will push the doctors' editing operations to other collaborative terminals of the same department in real time.

[0043] In some embodiments, the specific process of handling conflicts in treatment plans through the reconciliation mechanism of RBR and CBR is as follows: When at least two doctors are detected to have editing conflicts regarding the treatment plan for the same patient, the conflict resolution process is triggered. The RBR mechanism matches the modified plans submitted by each doctor with a pre-set clinical diagnosis and treatment rule base, and calculates the fit score of each plan with the current diagnosis and treatment guidelines. Using the CBR mechanism, K historical cases with similar characteristics to the current patient's condition are retrieved from the historical consultation case database, the corresponding successful treatment plans are extracted, and the similarity score between the current modified plan and the historical successful plan is calculated. Based on the professional title information of each submitting doctor, a corresponding professional title weight coefficient is assigned; Based on the compatibility score, similarity score, and professional title weighting coefficient, a weighted calculation is performed to obtain the comprehensive reconciliation score for each conflict resolution plan; the formula for calculating the comprehensive reconciliation score is as follows:

[0044] in, A comprehensive reconciliation score is given for conflict resolution solutions; the higher the score, the more recommended the solution is. This indicates the degree of matching between the treatment plan and the pre-set clinical practice guidelines (such as the "Guidelines for the Diagnosis and Treatment of Critical and Acute Illnesses in the Elderly 2024"), with a value range of [value missing]. 1 indicates a perfect match; This indicates the degree of similarity between the proposed treatment plan and historical successful treatment cases. It is calculated by retrieving similar historical cases using Case-Based Reasoning (CBR), and the value ranges from [value range missing]. 1 indicates high similarity; This indicates the weighting value assigned based on the professional title of the doctor submitting the plan, reflecting their clinical authority; , and These are the weighting coefficients for each score item (optionally, , , ).

[0045] The solution with the highest overall reconciliation score is selected as the recommended reconciliation solution and pushed to all participating doctors' terminals for confirmation or further revision. If the recommended solution receives majority confirmation, it is updated as the current execution solution, and the reconciliation process is recorded in the consortium blockchain for evidence storage.

[0046] In some embodiments, the Kalman filter algorithm is used to predict the trend of changes in the patient's physiological indicators over a future period of time based on the patient's real-time physiological indicator data. When the prediction results indicate a risk of abnormal indicators, an MDT combination adjustment plan is generated by combining the current MDT collaboration score with the disease prediction results; wherein, the MDT combination adjustment plan includes adding a specialist or replacing an existing specialist. Based on the MDT combination adjustment plan, update the MDT combination and notify relevant doctors to participate in the consultation; The patient's risk level is updated based on the adjusted MDT collaboration data and the patient's real-time physiological indicators.

[0047] Based on the same inventive concept, and corresponding to any of the above embodiments, refer to... Figure 2This invention provides a dynamic hierarchical scheduling and collaborative decision support system for MDT (Multidisciplinary Team) in elderly patients with acute and critical illness, used to implement the aforementioned method for dynamic hierarchical scheduling and collaborative decision support for MDT in elderly patients with acute and critical illness, comprising: The data acquisition and preprocessing module is used to collect patients' physiological index data, voice complaint data, video behavior data, frailty state data and environmental monitoring data, and to perform real-time transmission, anomaly detection, feature extraction and structured integration of the multimodal data. The risk assessment and classification module, connected to the data acquisition and preprocessing module, is used to receive preprocessed multimodal data, combine the results of geriatric frailty index and emotional state analysis, and output the patient's corresponding risk level through an assessment algorithm that integrates long short-term memory network and medical big model. The dynamic scheduling module, connected to the risk assessment and grading module, is used to match the optimal MDT combination using the dynamic weighted KNN algorithm based on the risk level and suspected emergency type, and to perform resource scheduling based on the NSGA-Ⅲ multi-objective optimization algorithm, generate consultation notices and send them to the corresponding departments. The collaborative decision-making module is connected to the dynamic scheduling module and the data acquisition and preprocessing module, respectively. It includes consortium blockchain nodes and edge computing nodes deployed in various departments. It is used to realize real-time data sharing and collaborative treatment plans among doctors from multiple disciplines, and to handle plan conflicts through the RBR and CBR reconciliation mechanism. The dynamic adjustment module is connected to the collaborative decision-making module and the risk assessment and grading module, respectively. It is used to predict the trend of changes in the patient's physiological indicators based on the Kalman filter algorithm, generate an adjustment plan in combination with the current MDT collaborative score, and update the MDT combination and risk level to form a closed-loop feedback.

[0048] Based on the same inventive concept, corresponding to any of the above embodiments, the present invention provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the dynamic hierarchical scheduling and collaborative decision support method for MDT in elderly patients with acute and critical illnesses as described in the embodiment.

[0049] Alternatively, the aforementioned electronic device may be a server.

[0050] In addition, this embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the dynamic hierarchical scheduling and collaborative decision support method for elderly critical care MDTs in this embodiment.

[0051] It is understood that the processor in the embodiments of the present invention may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0052] The method steps in the embodiments of the present invention can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0053] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted through a storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive (SSD)).

Claims

1. A dynamic hierarchical scheduling and collaborative decision support method for MDT (Multidisciplinary Team) in elderly patients with acute and critical illnesses, characterized in that, Includes the following steps: Multimodal data of patients are collected and preprocessed; wherein, the multimodal data includes physiological index data, voice complaint data, video behavior data, frailty state data and environmental monitoring data; Based on the preprocessed multimodal data, combined with the patient's frailty index and emotional state analysis results, the risk level of the patient is output by an assessment algorithm that integrates long short-term memory network and medical big model. Based on the patient's risk level and suspected emergency type, a dynamic weighting algorithm is used to match the optimal MDT combination, and a multi-objective optimization algorithm is used for resource scheduling to generate a consultation notice and send it to the corresponding department. Based on consortium blockchain nodes and edge computing nodes deployed in various departments, real-time data sharing and collaborative treatment plans among doctors from multiple disciplines are realized, and conflicts in treatment plans are handled through the reconciliation mechanism of RBR and CBR.

2. The method for dynamic hierarchical scheduling and collaborative decision support of MDT for elderly patients with acute and critical illnesses as described in claim 1, characterized in that... The specific process of collecting multimodal data from patients and preprocessing the multimodal data is as follows: By deploying medical IoT sensors, voice acquisition terminals, video surveillance equipment, and environmental monitoring equipment in the wards, real-time data on patients' physiological indicators, voice complaints, video behavior and frailty status, as well as environmental monitoring data of the wards are collected. The collected multimodal data is transmitted in real time to the edge computing nodes of the corresponding departments. In the edge computing nodes, the isolated forest algorithm is used to detect and remove outliers from the physiological index data, and to perform filtering and noise reduction. The normalization method is used to unify the physiological indicators to standard dimensions. The voice complaint data is subjected to speech recognition, converted into text data, and emotional feature vectors are extracted. The video behavior data is subjected to keyframe extraction and behavioral feature analysis to identify the patient's behavioral state and posture change trajectory. Based on the preset frailty assessment scale, the frailty status data is quantitatively scored to generate an geriatric frailty index. The environmental monitoring data was time-stamped and the sensors were calibrated. The processed multimodal data was then integrated into a unified structured data table according to the patient identifier, thus completing the preprocessing of the multimodal data.

3. The method for dynamic hierarchical scheduling and collaborative decision support of MDT for acute and critical illness in the elderly as described in claim 1, characterized in that, The specific process by which the patient's risk level is output based on the preprocessed multimodal data, combined with the patient's frailty index and emotional state analysis results, and through an assessment algorithm that integrates long short-term memory networks and a large medical model, is as follows: The physiological index time series data from the preprocessed multimodal data are input into the long short-term memory network model to extract physiological time series feature vectors; The preprocessed speech and text data is input into a pre-trained medical model to extract key chief complaint feature vectors. Based on the preprocessed voice complaint data, the emotional state coefficient is extracted through the voice emotion analysis model. Based on the preprocessed frailty data, the frailty index of the elderly is calculated according to the preset frailty assessment scale. The physiological time sequence feature vector, key chief complaint feature vector, emotional state coefficient and geriatric frailty index are concatenated to form a multi-dimensional fused feature vector. The multi-dimensional fused feature vector is input into the Bayesian fusion model, and the risk score is output by weighting it using preset weight coefficients. The risk score is mapped to a corresponding risk level based on the threshold range in which it falls; the risk levels include high risk, medium risk, and low risk.

4. The method for dynamic hierarchical scheduling and collaborative decision support of MDT for acute and critical illness in the elderly as described in claim 3, characterized in that, The specific process of matching the optimal MDT combination using a dynamic weighting algorithm based on the patient's corresponding risk level and suspected emergency type, and generating a consultation notification and sending it to the corresponding department based on a multi-objective optimization algorithm for resource scheduling is as follows: Obtain the current patient's multi-dimensional fused feature vector, risk level, and suspected emergency type; Based on preset scheduling rules, multiple dynamic weighting factors required for MDT scheduling are obtained; among them, the dynamic weighting factors include: the real-time workload factor of the target department, the matching degree factor between the candidate doctor's subspecialty and the suspected emergency type, the candidate doctor's historical consultation success rate factor, and the candidate doctor's historical response time compliance rate factor. Based on the real-time values ​​of each dynamic weight factor, the real-time weight values ​​of each factor are determined through a dynamic weight calculation model. The multi-dimensional fusion feature vector of the current patient, the suspected emergency type, and the real-time weight value are input into the dynamic weight KNN algorithm model. The dynamic weight KNN algorithm model calculates the weighted similarity between the current patient's features and the features of historical consultation cases based on the real-time weight value, and retrieves K historical cases with similarity greater than the preset similarity value from the historical case database. Extract successful MDT combination information corresponding to K historical cases as an initial candidate MDT resource pool for matching; A multi-objective optimization function is constructed with the optimization objectives of minimizing response time, maximizing system resource utilization, and maximizing the predicted consultation success rate. The candidate MDT resource pool, the response time threshold constraints corresponding to the patient risk level, and the real-time load status of each department are input into the multi-objective optimization scheduler based on the NSGA-Ⅲ algorithm. Under the premise of satisfying the response time threshold constraint, the multi-objective optimization scheduler solves the multi-objective optimization function, outputs the Pareto optimal solution set, and selects the final scheduling scheme from the optimal solution set. The final scheduling scheme determines the specific departments and doctors participating in the consultation. Based on the final scheduling plan, a consultation notification is generated and sent to the terminals of the corresponding departments and doctors; The system monitors the load index of each department in real time. When the load index of a department exceeds the preset warning threshold, it triggers a request for cross-departmental or cross-hospital resource coordination and scheduling.

5. The method for dynamic hierarchical scheduling and collaborative decision support of MDT for acute and critical illness in the elderly as described in claim 1, characterized in that, The specific process of achieving real-time data sharing and collaborative treatment plans among doctors from multiple disciplines based on consortium blockchain nodes and edge computing nodes deployed in various departments is as follows: A hybrid collaborative network is constructed based on consortium blockchain nodes deployed in emergency departments, intensive care units, geriatric departments, and specialized departments, and edge computing nodes at nurse stations in each department. Through edge computing nodes, the patient's current diagnosis and treatment data, vital signs and test results are synchronized in real time from the hospital information system and electronic medical record system and uploaded to the corresponding department's consortium blockchain node; Consortium blockchain nodes complete data on-chain notarization based on the Proof-of-Authority (PoA) consensus mechanism, and encrypt real-time data during transmission using the national cryptographic algorithm SM4. Doctors from various departments can log in to collaborative terminals deployed on edge computing nodes and edit treatment plans, add treatment opinions, or make annotations online based on real-time synchronized patient data. The collaborative terminals will push the doctors' editing operations to other collaborative terminals associated with the same patient in real time.

6. The method for dynamic hierarchical scheduling and collaborative decision support of MDT for elderly patients with acute and critical illnesses as described in claim 1, characterized in that, The specific process of resolving conflicts in treatment plans through the reconciliation mechanism of RBR and CBR is as follows: When at least two doctors are detected to have editing conflicts regarding the treatment plan for the same patient, the conflict resolution process is triggered. The RBR mechanism matches the modified plans submitted by each doctor with a pre-set clinical diagnosis and treatment rule base, and calculates the fit score of each plan with the current diagnosis and treatment guidelines. Using the CBR mechanism, K historical cases with similar characteristics to the current patient's condition are retrieved from the historical consultation case database, the corresponding successful treatment plans are extracted, and the similarity score between the current modified plan and the historical successful plan is calculated. Based on the professional title information of each submitting doctor, a corresponding professional title weight coefficient is assigned; The comprehensive reconciliation score for each conflict resolution plan is obtained by weighting the fit score, similarity score, and professional title weight coefficient. The solution with the highest overall reconciliation score is selected as the recommended reconciliation solution and pushed to all participating doctors' terminals for confirmation or further revision. If the recommended solution receives majority confirmation, it is updated as the current execution solution, and the reconciliation process is recorded in the consortium blockchain for evidence storage.

7. The method for dynamic hierarchical scheduling and collaborative decision support of MDT for elderly patients with acute and critical illnesses as described in claim 1, characterized in that, Based on real-time physiological data of patients, the Kalman filter algorithm is used to predict the trend of changes in patients' physiological indicators in the future period. When the prediction results indicate a risk of abnormal indicators, an MDT combination adjustment plan is generated by combining the current MDT collaboration score with the disease prediction results; wherein, the MDT combination adjustment plan includes adding a specialist or replacing an existing specialist. Based on the MDT combination adjustment plan, update the MDT combination and notify relevant doctors to participate in the consultation; The patient's risk level is updated based on the adjusted MDT collaboration data and the patient's real-time physiological indicators.

8. A dynamic hierarchical scheduling and collaborative decision support system for MDT (Multidisciplinary Team) in elderly patients with acute and critical illness, used to implement the dynamic hierarchical scheduling and collaborative decision support method for MDT in elderly patients with acute and critical illness as described in any one of claims 1-7, characterized in that, include: The data acquisition and preprocessing module is used to collect patients' physiological index data, voice complaint data, video behavior data, frailty state data and environmental monitoring data, and to perform real-time transmission, anomaly detection, feature extraction and structured integration of the multimodal data. The risk assessment and grading module, connected to the data acquisition and preprocessing module, is used to receive preprocessed multimodal data, combine the results of geriatric frailty index and emotional state analysis, and output the patient's corresponding risk level through an assessment algorithm that integrates long short-term memory network and medical big model. The dynamic scheduling module, connected to the risk assessment and grading module, is used to match the optimal MDT combination using the dynamic weighted KNN algorithm based on the risk level and suspected emergency type, and to perform resource scheduling based on the NSGA-Ⅲ multi-objective optimization algorithm, generate consultation notices and send them to the corresponding departments. The collaborative decision-making module is connected to the dynamic scheduling module and the data acquisition and preprocessing module, respectively. It includes consortium blockchain nodes and edge computing nodes deployed in various departments. It is used to realize real-time data sharing and collaborative treatment plans among doctors from multiple disciplines, and to handle plan conflicts through the RBR and CBR reconciliation mechanism. The dynamic adjustment module is connected to the collaborative decision-making module and the risk assessment and grading module, respectively. It is used to predict the trend of changes in the patient's physiological indicators based on the Kalman filter algorithm, generate an adjustment plan in combination with the current MDT collaborative score, and update the MDT combination and risk level to form a closed-loop feedback.

9. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program and the processor runs the computer program to enable the electronic device to perform the dynamic hierarchical scheduling and collaborative decision support method for acute and critical care MDT in the elderly as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the dynamic hierarchical scheduling and collaborative decision support method for MDT in elderly patients with acute and critical illnesses as described in any one of claims 1-7.

Images