Intelligent triage method and system for outpatient and emergency treatment

Abstract

The invention provides an intelligent triage method and system for outpatient and emergency treatment, and relates to the technical field of medical information. The method comprises the steps that physiological data, symptom description text data, facial image data and voice audio data of a patient are collected through a multi-mode sensor; analyzing the image and audio data by using an emotion calculation engine to generate emotion state data; processing the physiological and text data through an analysis model to generate illness state emergency degree data; fusing emotion and illness state data to form a preliminary triage result; in combination with the real-time medical resource state data, a triage decision-making scheme including department allocation, processing priority and personalized pacifying strategies is generated through a dynamic decision-making device; finally, guiding and pacifying are executed through the execution terminal and the interaction robot, automation, intelligence and humanization of the triage process are achieved, the triage efficiency and accuracy are remarkably improved, medical resource configuration is optimized, and care for psychological needs of patients is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information technology, in particular to a method and system for intelligent triage in outpatient and emergency departments. BACKGROUND

[0002] In the current outpatient and emergency treatment process, the triage link is highly dependent on manual completion, and patients need to experience multiple queuing, inquiry and guidance, with low overall efficiency. Especially during peak hours, manual triage is difficult to quickly respond to diverse patient needs, resulting in prolonged waiting time and frequent congestion at key nodes. In addition, there are significant differences in the expression ability, emotional state and understanding level of the patient group, further increasing the difficulty and error of information transmission in the triage process, not only affecting the treatment experience, but also causing excessive consumption of medical human resources in repetitive and explanatory work.

[0003] Therefore, there is an urgent need for a triage method that can be efficient, accurate and have emotional perception ability to systematically improve the response speed, resource utilization rate and patient satisfaction of emergency services. SUMMARY

[0004] To solve the above problems existing in the prior art, the first aspect of the present application provides a method for intelligent triage in outpatient and emergency departments, comprising: S1: based on the interaction with the patient, collecting the physiological data, symptom description text data, facial image data and voice audio data of the patient through a data acquisition system configured with multi-modal sensors; S2: based on the facial image data and voice audio data, generating emotion state data for quantifying the current emotional state of the patient through an emotion computing engine; S3: based on the physiological data and symptom description text data, generating disease severity data representing the criticality of the patient's disease through an analysis model; S4: fusing the emotion state data and disease severity data, generating a preliminary triage result containing physiological and psychological analysis dimensions through a decision fusion module; S5: based on the preliminary triage result and medical resource state data obtained in real time from a hospital information system, generating a triage decision scheme containing department assignment, processing priority and personalized pacification strategy through a dynamic decision maker; S6: based on the triage decision scheme, executing department guidance instructions and starting personalized pacification strategies through an execution terminal and an interactive robot.

[0005] In some implementations, S1 includes: S101: based on the initial state of the patient after admission, automatically collecting physiological data including body temperature, heart rate and blood oxygen saturation through a vital sign monitoring device; S102: A human-computer interaction interface based on a triage desk or self-service terminal, receiving text data describing symptoms, including chief complaints and medical history, input by the patient or accompanying person; S103: Based on non-contact contact with the patient's face, facial image data including facial expressions and facial blood flow dynamics are acquired through a visual sensor; S104: Based on the question-and-answer interaction process with the patient, voice audio data containing tone and speech rate characteristics is collected through a microphone.

[0006] In some implementations, S2 includes: S201: Based on facial image data, feature extraction is performed using a convolutional neural network model to obtain image feature vectors representing facial expression features; S202: Based on speech audio data, noise reduction and feature analysis are performed through acoustic models to obtain acoustic feature vectors that characterize the emotional features of speech. S203: By fusing image feature vectors and acoustic feature vectors, a unified multimodal emotion feature representation is obtained through a multimodal fusion algorithm; S204: Based on multimodal emotion feature representation, through a pre-trained emotion classification model, it outputs emotion state data to quantify the patient's current emotional state.

[0007] In some implementations, S3 includes: S301: Based on physiological data, standardized physiological parameters that meet the model input requirements are obtained through data standardization processing; S302: Based on symptom description text data, keyword extraction and semantic encoding are performed using natural language processing technology to obtain structured symptom vectors; S303: Integrates standardized physiological parameters and structured symptom vectors to form a comprehensive feature input; S304: Based on comprehensive feature input, the system calculates and outputs data on the urgency of the patient's condition, representing the severity of the patient's illness, through an analytical model.

[0008] In some implementations, S5 generates personalized appeasement strategies, including: S501: Based on the emotional state data in the preliminary triage results, the basic soothing strategy type is obtained by querying the preset strategy mapping table; S502: Based on the real-time availability of medical staff and the working status of equipment in the medical resource status data, the execution subject and intensity of the basic reassurance strategy type are adaptively adjusted to generate personalized reassurance strategies. The personalized reassurance strategies include reassurance action instructions executed by the interactive robot or audiovisual content instructions played by the multimedia system.

[0009] In some implementations, the adaptive adjustments in S502 include: S5021: Based on medical resource status data, determine whether the current workload of medical staff exceeds a preset threshold; S5022: If the current workload of medical staff exceeds the preset threshold, select a personalized reassurance strategy with automated equipment as the main implementer. S5023: If the current workload of medical staff is lower than the preset threshold, add an auxiliary plan to the personalized comfort strategy that suggests that medical staff provide artificial psychological intervention.

[0010] In some implementations, S5 includes: S511: Based on the urgency data of the initial triage results, generate an initial treatment priority sequence for patients using a sorting algorithm; S512: Based on medical resource status data, the corresponding medical department and medical equipment resources are allocated to patients in the initial processing priority sequence through a resource matching algorithm; S513: Integrate the initial processing priority sequence, allocated departmental and medical equipment resources, and personalized reassurance strategies to form a triage decision plan.

[0011] In some implementations, S6 is followed by: S7: Based on positioning beacons deployed within the hospital and smart tags worn by patients, it continuously tracks patient locations through a real-time positioning system and generates patient flow data; S8: Based on patient flow data, monitor the movement and dwell status of patients between predefined treatment nodes; S9: In response to monitoring abnormal patient status or waiting time exceeding expectations, generate an updated triage decision plan by triggering a reassessment mechanism.

[0012] In some implementations, S9 includes: S901: Based on patient flow data, if it is detected that the duration of a patient's stay at a certain node exceeds a preset time limit associated with the patient's priority, or if an alarm is received indicating abnormal vital signs of the patient, a reassessment trigger signal is generated. S902: Based on the reassessment trigger signal, it automatically calls the data acquisition system and emotion computing engine to re-acquire the patient's current data and generate updated emotion state data and disease urgency data; S903: Based on updated emotional state data and disease urgency data, an updated triage decision plan is generated again through the decision fusion module and dynamic decision maker.

[0013] Secondly, the present invention provides an intelligent triage system for outpatient and emergency departments, the system employing the method provided in any of the above embodiments, the system comprising: The data acquisition module is used to collect patients' physiological data, symptom description text data, facial image data, and voice and audio data based on interactions with patients and through a data acquisition system configured with multimodal sensors. The emotion state analysis module is used to generate emotion state data to quantify the patient's current emotional state based on facial image data and voice audio data through an emotion computing engine. The urgency analysis module is used to generate urgency data that characterizes the severity of a patient's condition based on physiological data and symptom description text data through an analysis model. The data fusion and triage module is used to integrate emotional state data and disease urgency data. Through the decision fusion module, it generates preliminary triage results that include physiological and psychological analysis dimensions. The dynamic decision-making and resource scheduling module is used to generate a triage decision plan based on the preliminary triage results and the medical resource status data obtained in real time from the hospital information system, through a dynamic decision-maker. This plan includes the allocation of the treatment department, the processing priority, and the personalized comfort strategy. The decision execution and patient interaction module is used to execute departmental guidance instructions and initiate personalized reassurance strategies based on the triage decision plan, through the execution terminal and interactive robot.

[0014] Compared with existing technologies, the advantages of this invention are as follows: By systematically integrating multimodal data acquisition and intelligent analysis through an intelligent triage method for outpatient and emergency departments, the efficiency and accuracy of triage are effectively improved. First, a data acquisition system equipped with multimodal sensors collects patients' physiological data, symptom description text data, facial image data, and voice / audio data, achieving a comprehensive perception of the patient's state and providing a rich, multidimensional data foundation for subsequent analysis. Next, an emotion computing engine generates emotion state data based on facial image data and voice / audio data, objectively quantifying the patient's emotions and providing a basis for subsequent psychological-based triage decisions. The analysis model generates urgency data based on physiological data and symptom description text data, accurately reflecting the patient's physiological severity. The decision fusion module integrates the emotion state data and urgency data to generate preliminary triage results, which simultaneously encompass both physiological and psychological states, making them more comprehensive and humane. The dynamic decision-maker further combines medical resource status data from the hospital information system to generate triage decision schemes that include departmental allocation, processing priorities, and personalized reassurance strategies, achieving simultaneous optimization of resource allocation and personalized services. Finally, by implementing departmental guidance and reassurance strategies through the terminal and interactive robot, the triage decision-making process was ensured to be effectively implemented.

[0015] This method, through the collaborative operation of multiple modules, not only significantly shortens triage time and reduces reliance on manual labor, but also enhances the ability to respond to patients' emotional needs while improving triage accuracy, thus promoting the overall upgrade of emergency services towards intelligence and humanization. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 The diagram shown is a flowchart of an intelligent triage method for outpatient and emergency departments according to an embodiment of the present invention.

[0018] Figure 2 The diagram shown is a structural schematic of an intelligent triage system for outpatient and emergency departments provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] The specific embodiments of the present invention will be described below.

[0021] Example 1 like Figure 1 As shown, this invention proposes an intelligent triage method for outpatient and emergency departments, comprising: S1: Based on the interaction with the patient, a data acquisition system with multimodal sensors is configured to collect the patient's physiological data, symptom description text data, facial image data, and voice audio data; S2: Based on facial image data and voice audio data, an emotion computing engine is used to generate emotion state data to quantify the patient's current emotional state. S3: Based on physiological data and symptom description text data, through analysis models, generate data on the urgency of the patient's condition, which characterizes the severity of the patient's condition. S4: By integrating emotional state data and disease urgency data, the decision fusion module generates preliminary triage results that include physiological and psychological analysis dimensions. S5: Based on the preliminary triage results and the medical resource status data obtained in real time from the hospital information system, a triage decision plan is generated through a dynamic decision-maker, which includes the allocation of the treatment department, the processing priority, and the personalized comfort strategy. S6: Based on the triage decision-making scheme, the system executes departmental guidance instructions and initiates personalized reassurance strategies through the execution terminal and interactive robot.

[0022] First, the data acquisition system initiates an interaction process with the patient. This system integrates multiple types of sensors to comprehensively acquire information reflecting the patient's condition. These sensors include physiological parameter monitoring devices for collecting indicators such as body temperature, heart rate, and blood oxygen saturation; an interactive terminal for recording the patient's chief complaint and medical history; visual sensors for capturing facial expressions and micro-blood flow changes; and a microphone array for recording the patient's voice. This multimodal collaborative data acquisition lays a solid data foundation for subsequent intelligent analysis, avoiding the bias and errors that may arise from a single data source, thus ensuring the comprehensiveness and accuracy of subsequent triage judgments from the outset.

[0023] The collected data is synchronously transmitted to the system's backend processing unit. The emotion computing engine then processes the facial image and audio data. Utilizing advanced computer vision and speech signal processing algorithms, the engine extracts key facial feature points from the images and analyzes acoustic features such as pitch, speech rate, and loudness from the audio. Through in-depth analysis and fusion of these features, the engine generates objective, quantifiable emotional state data that accurately describes the patient's current emotional level, such as anxiety, distress, or calmness. This step transforms subjective emotional feelings into objective data, introducing a crucial psychological dimension to triage decisions.

[0024] In parallel, the analysis model processes physiological data and symptom description text data. The physiological data undergoes standardization, cleaning, and formatting, transforming it into standardized parameters recognizable by the model. The symptom description text data is then deeply analyzed using natural language processing techniques to extract key medical terms, symptom descriptors, and their contextual relationships, transforming them into structured numerical vectors. The analysis model is typically a machine learning or deep learning model trained on a large amount of medical data. It receives the fused standardized physiological parameters and structured symptom vectors and, through complex internal calculations, outputs a severity-based assessment of the patient's condition. This data serves as the core medical basis for triage decisions.

[0025] Subsequently, the decision fusion module begins operation. This module receives emotional state data from the emotion computing engine and urgency data from the analysis model. Its function is not a simple aggregation, but rather it employs specific fusion algorithms (such as weighted averaging, feature concatenation, or rule-based reasoning) to integrate and assess the patient's physiological urgency and psychological stress state, generating a preliminary triage result. This result thus includes analysis from both physiological and psychological dimensions, making the assessment of the patient's condition more comprehensive and humane, providing a technological path to achieving compassionate healthcare that "treats both the disease and the person."

[0026] The initial triage results are fed into the dynamic decision-making system. A key function of this system is its real-time access to the hospital's information system, continuously acquiring the latest medical resource status data. This data includes the current number of patients waiting in each department, the real-time workload of doctors and nurses, and the availability of critical examination equipment. The dynamic decision-making system uses the initial triage results as its basic input, overlaid with current resource constraints, and performs comprehensive calculations using built-in sorting and resource matching algorithms. Ultimately, it outputs an executable triage decision plan. This plan not only includes recommendations for departmental allocation and priority ranking, but also innovatively incorporates personalized reassurance strategies tailored to the patient's emotional state. This step represents a leap from static analysis to dynamic scheduling, ensuring that triage decisions are not only scientific but also practical, maximizing the utilization of limited medical resources.

[0027] Ultimately, the triage decision-making plan is distributed to the execution terminal and the interactive robot. The execution terminal typically includes a display screen and printer, used to show guidance information to patients or staff and print documents. The interactive robot then executes specific guidance actions according to the instructions in the plan, such as leading patients to designated departments through voice and screen instructions, or initiating personalized reassurance strategies, such as playing soothing music or engaging in comforting dialogue. Through automated execution, the burden of manual guidance for medical staff is greatly reduced, ensuring the efficient and accurate implementation of triage decisions and completing a closed loop from intelligent decision-making to physical execution. The entire method, through the close connection of six steps, forms a complete intelligent triage closed loop, effectively solving the problems of low efficiency, high resource consumption, and lack of emotional care in manual triage mentioned in the background technology, and significantly improving the overall efficiency of outpatient and emergency services and patient experience.

[0028] In some implementations, S1 includes: S101: Based on the patient's initial state upon arrival at the hospital, physiological data including body temperature, heart rate, and blood oxygen saturation are automatically collected through vital sign monitoring equipment; S102: A human-computer interaction interface based on a triage desk or self-service terminal, receiving text data describing symptoms, including chief complaints and medical history, input by the patient or accompanying person; S103: Based on non-contact contact with the patient's face, facial image data including facial expressions and facial blood flow dynamics are acquired through a visual sensor; S104: Based on the question-and-answer interaction process with the patient, voice audio data containing tone and speech rate characteristics is collected through a microphone.

[0029] Specifically, the data collection process begins after the patient arrives at the hospital, and the vital signs monitoring equipment automatically starts its testing program. These devices are typically non-contact or minimally invasive, such as infrared thermometers and finger-clip pulse oximeters. They can automatically collect multiple key physiological data points, including body temperature, heart rate, and blood oxygen saturation, in a very short time. This automated process reduces manual intervention and potential errors, while improving data collection efficiency and saving valuable time for subsequent rapid triage, making it particularly suitable for rapid screening of large populations.

[0030] While or after collecting physiological data, the system guides patients or their companions to input information through a human-computer interaction interface at the triage desk or self-service terminal. The interface features clear and easy-to-understand prompts, allowing users to input their chief complaint, present illness, and past medical history via touchscreen or keyboard, forming structured symptom description text data. This method transforms the patient's subjective description into digital text that can be directly processed by the machine, providing a crucial source of unstructured data for subsequent natural language processing and analysis models. It is a key step in compensating for the inadequacy of purely physiological data and gaining a deeper understanding of the patient's condition.

[0031] Data acquisition by the visual sensor is performed in a non-contact manner. This sensor is typically a high-definition camera cleverly integrated into the triage desk or interactive robot, capturing facial image data of the patient during normal operation. This image data not only contains rich facial expression information but also captures subtle dynamic changes in facial blood flow through specific imaging techniques. This non-contact acquisition method avoids adding extra burden or discomfort to the patient, ensuring the naturalness and authenticity of the data acquisition. It provides high-quality, interference-free raw image input for the emotion computing engine, forming the basis for accurate assessment of emotional states.

[0032] The microphone array operates continuously throughout the question-and-answer interaction. The system may interact with the patient by asking questions, and the microphones are responsible for capturing the patient's responses, forming audio data. This audio data not only contains semantic information but, more importantly, its acoustic features, such as pitch, speech rate, and the degree of vocal tremor. These paralinguistic features are crucial inputs for affective computing. By simultaneously acquiring speech and facial data, the affective computing engine can obtain mutually corroborating multimodal information, making the generated affective state data more robust and reliable, effectively avoiding misjudgments caused by distortion of a single modality signal.

[0033] In some implementations, S2 includes: S201: Based on facial image data, feature extraction is performed using a convolutional neural network model to obtain image feature vectors representing facial expression features; S202: Based on speech audio data, noise reduction and feature analysis are performed through acoustic models to obtain acoustic feature vectors that characterize the emotional features of speech. S203: By fusing image feature vectors and acoustic feature vectors, a unified multimodal emotion feature representation is obtained through a multimodal fusion algorithm; S204: Based on multimodal emotion feature representation, through a pre-trained emotion classification model, it outputs emotion state data to quantify the patient's current emotional state.

[0034] First, convolutional neural network (CNN) models were applied to process facial image data. This model, a deep learning method, uses a multi-layered structure to automatically learn and extract abstract features from raw pixels, ranging from low-level to high-level, such as the shape of the eyes and mouth, and the degree of eyebrow distortion, ultimately outputting a highly generalized and information-rich image feature vector. Using CNNs avoids the tedious and potentially inaccurate manual feature design of traditional methods, achieving automated and high-precision feature extraction, laying a solid foundation for accurate facial expression recognition.

[0035] Acoustic models, on the other hand, are specifically designed for processing speech audio data. Their processing typically begins with noise reduction and preprocessing to eliminate ambient noise and interference from recording equipment, thereby improving the signal-to-noise ratio. Subsequently, the model extracts a series of acoustic features from the clean audio signal, such as Mel-frequency cepstral coefficients, fundamental frequency profile, and energy envelope. These features collectively constitute the acoustic feature vector. This vector quantifies various emotion-related characteristics in speech, transforming the previously abstract sensory impression of "sounding anxious" into a concrete, computable mathematical representation, providing standardized input for subsequent emotion classification.

[0036] The multimodal fusion algorithm is one of the core innovations of this step. This algorithm receives feature vectors from both visual and auditory channels: image feature vectors and acoustic feature vectors. Instead of simply concatenating the two vectors, it employs a more advanced fusion strategy, such as attention-based fusion. This strategy dynamically evaluates the importance weights of the two modalities at different times, achieving finer feature integration and ultimately generating a unified multimodal emotional feature representation. This fusion method effectively utilizes the complementarity between different modal information; for example, facial expressions may mask emotions, but speech can reveal them, resulting in a more comprehensive and robust emotional representation than any single modality.

[0037] The pre-trained emotion classification model receives the fused multimodal emotion feature representation. This model is typically trained on a large number of labeled multimodal emotion datasets, learning the complex mapping relationships between emotion features and specific emotion categories (such as happiness, sadness, anger, fear, neutrality, etc.) or dimensions (such as valence, arousal). After calculating this feature representation, the model outputs the final emotion state data, which is usually a probability distribution or a set of continuous dimensional scores, thus achieving a refined and quantitative description of the patient's current emotional state. This provides objective and reliable psychological state parameters for the subsequent decision fusion module, enabling the triage system to "read between the lines" like an experienced nurse.

[0038] In some implementations, S3 includes: S301: Based on physiological data, standardized physiological parameters that meet the model input requirements are obtained through data standardization processing; S302: Based on symptom description text data, keyword extraction and semantic encoding are performed using natural language processing technology to obtain structured symptom vectors; S303: Integrates standardized physiological parameters and structured symptom vectors to form a comprehensive feature input; S304: Based on comprehensive feature input, the system calculates and outputs data on the urgency of the patient's condition, representing the severity of the patient's illness, through an analytical model.

[0039] Specifically, the generation of data on the urgency of a patient's condition relies on in-depth analysis of physiological and textual data. First, the collected raw physiological data needs to undergo data standardization. Since monitoring devices from different sources may have different data formats, units, and dimensions, this step uses a series of algorithms (such as normalization and Z-score standardization) to transform these heterogeneous data into standardized physiological parameters of a uniform scale that meet the input requirements of the analytical model. Standardization eliminates dimensional differences between data points, prevents certain features with large values ​​from dominating model training, ensures the fairness and accuracy of the analysis, and provides a stable and reliable input for the model.

[0040] Natural Language Processing (NLP) technology is specifically designed for processing textual data describing symptoms. The process includes keyword extraction, entity recognition, and semantic encoding. The system identifies key medical terms (such as "chest pain," "difficulty breathing," and "nausea"), body parts, and adverbs of severity from the text, understanding their modifiers and logical relationships. Subsequently, through word embeddings or more advanced sentence encoding models, these discrete textual symbols are transformed into a continuous, dense, structured symptom vector. This vector contains semantic information about the symptoms, enabling the computer to "understand" the patient's description of their ailment, serving as a crucial bridge for converting unstructured text into computable data.

[0041] Subsequently, the standardized physiological parameters and structured symptom vectors need to be fused to form a comprehensive feature input. This fusion operation is typically performed at the feature level, concatenating the parameter vector representing the physiological state and the semantic vector representing the symptom description into a longer, more comprehensive feature vector. This comprehensive feature input thus includes both objective physiological indicators and subjective symptom description information, complementing each other and providing the analysis model with a panoramic view of the patient's condition. This avoids the limitations of making judgments based on a single information source and greatly enhances the comprehensiveness of the disease assessment.

[0042] The analytical model receives this comprehensive feature input and performs calculations. This model is typically a machine learning-based classifier or regressor that internally learns from a large amount of historical case data, establishing a complex mapping relationship between various combinations of physiological symptoms and the severity of the illness. After calculating the input features, the model outputs a severity level data point. This data can be a specific classification (such as the level in the five-level triage system of emergency departments) or a continuous score representing the probability of critical illness. This output provides the core medical judgment basis for triage decisions and is the core technology for quickly identifying critically ill patients.

[0043] In some implementations, S5 generates personalized appeasement strategies, including: S501: Based on the emotional state data in the preliminary triage results, the basic soothing strategy type is obtained by querying the preset strategy mapping table; S502: Based on the real-time availability of medical staff and the working status of equipment in the medical resource status data, the execution subject and intensity of the basic reassurance strategy type are adaptively adjusted to generate personalized reassurance strategies. The personalized reassurance strategies include reassurance action instructions executed by the interactive robot or audiovisual content instructions played by the multimedia system.

[0044] Specifically, the generation of personalized reassurance strategies is a crucial component of the dynamic decision-making process, reflecting the system's response to the patient's psychological needs. This process begins with the analysis of emotional state data from the initial triage results. Based on the quantified results of this data, the system queries a pre-defined strategy mapping table. This mapping table defines the correspondence between different emotional states (such as high anxiety, moderate fear, and calmness) and basic reassurance strategy types (such as verbal reassurance, distraction, and in-depth explanation). By looking up the table, the system can quickly determine the basic strategy framework for dealing with the current patient's emotions. This ensures the logical consistency and coherence of strategy generation, providing a foundational template for subsequent personalized adjustments.

[0045] Basic reassurance strategies still need to be refined based on actual resource conditions to become feasible solutions. The dynamic decision-maker obtains real-time data on the status of medical resources from the hospital information system, with the real-time availability of medical staff and the operational status of equipment being key parameters. Based on these parameters, the decision-maker adaptively adjusts the implementing entity and intensity of the basic reassurance strategy. For example, if emotional state data indicates that the patient needs in-depth psychological intervention, but current psychological counselor resources are scarce, the system may adjust its strategy, first having an interactive robot perform basic reassurance and marking the patient as requiring further human follow-up. This adjustment ensures that the reassurance strategy not only meets the patient's needs technically but is also operationally feasible.

[0046] The resulting personalized reassurance strategies are then concretized into actionable instructions. These instructions fall into two main categories: one is reassurance actions performed by the interactive robot, such as playing pre-set soothing audio content, making specific physical movements (like nodding), or guiding patients to perform deep breathing exercises; the other is audiovisual content played by the multimedia system, such as playing educational videos or soothing music that specific patients might find interesting on screens in the waiting area. By using automated equipment to execute these strategies, timely and humanistic care can be provided when medical staff are unable to attend to patients, filling the emotional gap in the medical process with technology, improving the overall patient experience, and ensuring that medical resources are focused more on core diagnostic and treatment activities.

[0047] In some implementations, the adaptive adjustments in S502 include: S5021: Based on medical resource status data, determine whether the current workload of medical staff exceeds a preset threshold; S5022: If the current workload of medical staff exceeds the preset threshold, select a personalized reassurance strategy with automated equipment as the main implementer. S5023: If the current workload of medical staff is lower than the preset threshold, add an auxiliary plan to the personalized comfort strategy that suggests that medical staff provide artificial psychological intervention.

[0048] Specifically, adaptive adjustment is the core decision-making step in generating personalized reassurance strategies. Its purpose is to ensure that the strategies not only technically adapt to the patient's emotional state but also operationally align with the hospital's real-time resource load. This adjustment process begins with in-depth analysis of medical resource status data, from which the system extracts key indicators: the workload of medical staff. The system compares the current actual workload value with a preset threshold. This threshold is a reasonable boundary value set based on a combination of factors, including historical operational data, departmental capacity models, and nursing standards, to scientifically define whether resources are in a "busy" or "relatively idle" state. This comparison provides an objective and quantitative basis for subsequent strategy branch selection, ensuring the consistency and reliability of the system's decisions.

[0049] If the assessment indicates that the workload of current medical staff exceeds a preset threshold, it signifies a high level of strain on human resources. In this case, priority should be given to utilizing automated resources to perform reassurance tasks and alleviate manpower pressure. Therefore, when selecting the execution entity, the system tends to choose personalized reassurance strategies implemented by automated devices. These solutions rely entirely on automated facilities such as interactive robots and multimedia playback systems, for example, robots providing voice-based reassurance or playing preset soothing videos or music. This selection ensures that even with extremely limited manpower, patients' basic emotional needs can still be responded to promptly and effectively, preventing potential emotional deterioration due to waiting for human intervention, and demonstrating the crucial role of technology in optimizing resource allocation.

[0050] Conversely, if the assessment indicates that the current workload of medical staff is below a preset threshold, it suggests that the system has the manpower to provide deeper and more humane services. In this case, when generating personalized reassurance strategies, the system will not merely rely on automated solutions, but will add an auxiliary plan suggesting human psychological intervention by medical staff. This auxiliary plan might be a reminder sent to the nurse's workstation, suggesting that the nurse communicate with the patient face-to-face, explain, or comfort them at an appropriate time. This does not replace automated strategies, but rather enhances and complements them, organically combining cold, automated services with warm, human care. This adaptive adjustment greatly improves the flexibility and humanization of reassurance strategies, enabling the system to intelligently allocate services of different qualities based on resource availability. Ultimately, this achieves the goal of maximizing the satisfaction of patients' psychological needs while ensuring operational efficiency, ensuring that care measures are truly implemented.

[0051] In some implementations, S5 includes: S511: Based on the urgency data of the initial triage results, generate an initial treatment priority sequence for patients using a sorting algorithm; S512: Based on medical resource status data, the corresponding medical department and medical equipment resources are allocated to patients in the initial processing priority sequence through a resource matching algorithm; S513: Integrate the initial processing priority sequence, allocated departmental and medical equipment resources, and personalized reassurance strategies to form a triage decision plan.

[0052] Specifically, generating a triage decision-making scheme is a comprehensive process involving dynamic optimization of multiple factors. Its core objective is to accurately and efficiently match limited medical resources with differentiated patient needs. This process begins by initially ranking patients based on the urgency data included in the preliminary triage results. The ranking algorithm receives this data as input, which is typically a quantified score or a defined level. Based on the fundamental principle of "prioritizing critical cases," the algorithm generates an initial treatment priority sequence according to the urgency of the condition, from highest to lowest. This sequence forms the basic framework for resource allocation, ensuring that critically ill patients with unstable vital signs are automatically identified and prioritized by the system. This algorithmic approach safeguards the core bottom line of medical safety and avoids errors that might occur due to negligence or pressure during manual ranking.

[0053] However, simple prioritization does not consider the real-time availability of medical resources. An ideal solution must combine patient demand with resource supply. Therefore, a resource matching algorithm is introduced, which reads medical resource status data in real time. This data provides a snapshot of the current operation of each subsystem in the hospital, including which clinics still have capacity, which doctors will soon be available, and which examination equipment is currently available. Using an initial priority sequence as a baseline, the algorithm attempts to match each patient in the sequence with the most suitable department and medical equipment resources. The matching process is not a simple assignment; it comprehensively considers multiple constraints such as departmental specialization matching, path distance, and estimated resource idle time, aiming to reduce overall patient wait time and improve resource utilization. This step transforms the static priority sequence into a dynamic, executable resource allocation plan.

[0054] Ultimately, the dynamic decision-maker needs to integrate and encapsulate the aforementioned results. It integrates three parts of information: the initial processing priority sequence, the allocation of departments and medical equipment resources calculated through resource matching algorithms, and the personalized reassurance strategies tailored for each patient, forming a complete and executable triage decision plan. This plan is a structured set of instructions that clearly indicates "which patient" belongs to "what priority level," should be guided to "which department," can use "what equipment," and should be provided with "what kind of reassurance" during this process. The formation of this plan marks the final completion from data analysis to decision output. It provides clear, explicit, and comprehensive action guidelines for subsequent execution terminals and interactive robots, ensuring the continuity, efficiency, and humanization of the entire triage process.

[0055] In some implementations, S6 is followed by: S7: Based on positioning beacons deployed within the hospital and smart tags worn by patients, it continuously tracks patient locations through a real-time positioning system and generates patient flow data; S8: Based on patient flow data, monitor the movement and dwell status of patients between predefined treatment nodes; S9: In response to monitoring abnormal patient status or waiting time exceeding expectations, generate an updated triage decision plan by triggering a reassessment mechanism.

[0056] Specifically, the execution of the triage decision-making plan is not the end of the service. The system also constructs a closed-loop mechanism for continuous monitoring and dynamic feedback to address uncertainties in the treatment process. This mechanism begins with the continuous tracking of patient location, achieved through the collaboration of location beacons deployed within the hospital and smart tags worn by patients. Location beacons form a wireless network coverage along key paths and nodes within the hospital, and the smart tags worn by patients periodically emit signals, which are received by nearby beacons. By processing these signals, the real-time positioning system can continuously track patient location and generate patient flow data reflecting patient movement trajectories and dwell times. This system enables digital and visual monitoring of patient movement within the hospital, providing a data foundation for process optimization and anomaly early warning.

[0057] Based on continuously generated patient flow data, the system can monitor the movement and dwell time of patients between predefined treatment nodes. These nodes include triage areas, waiting areas, examination rooms, and treatment rooms. The system incorporates rule-based monitoring logic; for example, it records the timestamp of a patient entering a certain area and calculates the duration of their stay there. By comparing the actual dwell time with the expected time range based on their triage level, the system can automatically monitor the progress of the entire treatment process and identify bottlenecks. This monitoring extends system management from single-point triage to full-process management, deepening the focus from "where the patient is" to "whether the patient is receiving treatment smoothly."

[0058] When the monitoring logic detects anomalies, the system can proactively respond. Anomalies primarily include abnormal patient conditions or waiting times exceeding expectations, such as a patient spending excessive time in a non-rest area, or receiving abnormal alarms from integrated vital sign monitoring devices. Once these conditions are triggered, the system proactively intervenes by initiating a reassessment mechanism. This mechanism automatically initiates a new round of data collection and analysis to verify the patient's current status and generate an updated triage decision. This closed-loop feedback mechanism significantly enhances the system's robustness and proactivity, ensuring that triage decisions adapt to dynamic changes in patient conditions and hospital conditions, promptly correcting potential deviations, adding another strong technical layer of protection for patient safety, and achieving truly intelligent continuous care.

[0059] In some implementations, S9 includes: S901: Based on patient flow data, if it is detected that the duration of a patient's stay at a certain node exceeds a preset time limit associated with the patient's priority, or if an alarm is received indicating abnormal vital signs of the patient, a reassessment trigger signal is generated. S902: Based on the reassessment trigger signal, it automatically calls the data acquisition system and emotion computing engine to re-acquire the patient's current data and generate updated emotion state data and disease urgency data; S903: Based on updated emotional state data and disease urgency data, an updated triage decision plan is generated again through the decision fusion module and dynamic decision maker.

[0060] Specifically, the reassessment mechanism is a crucial emergency response process for ensuring patient safety and handling emergencies, and its activation relies on clearly defined triggering conditions. The generation logic of the reassessment trigger signal is based on real-time analysis of patient flow data and integrated alarms. The system continuously monitors the duration of a patient's stay at a specific treatment node and compares this duration with preset time limits associated with that patient's priority; for example, the waiting time for critically ill patients will be much shorter than that of ordinary patients. Once the system detects that a patient's stay exceeds their personalized preset time limit, it preliminarily determines that the process may be blocked or the patient's condition may have changed. Simultaneously, the system also listens for real-time alarm signals from vital sign monitoring devices. Any signal indicating an abnormal vital sign will be considered a highest-priority trigger event. Both of these situations (timeout or abnormality) will immediately generate a reassessment trigger signal, which is the switch to initiate the entire reassessment process.

[0061] Once the reassessment trigger signal is generated, the system automatically invokes the data acquisition system and the affective computing engine to perform a snapshot-like reassessment of the patient's current state. This process is similar to initial triage, but it is a targeted reassessment for a specific patient. The data acquisition system will attempt to collect the patient's physiological data, facial image data, and voice audio data again, while the affective computing engine will calculate updated affective state data based on the new data. The analysis model will also work in parallel, generating updated urgency data based on the latest physiological data and potentially updated symptom descriptions (such as those supplemented by accompanying persons). This re-collection and calculation process ensures that decisions are based on the patient's latest and most accurate state information, avoiding the risk of making incorrect decisions based on outdated information, and demonstrating the system's high level of responsibility towards patients.

[0062] After acquiring updated emotional state and urgency data, the decision-making process restarts. This updated data is input into the decision fusion module, which, like during the initial triage, integrates physiological and psychological dimensions to generate an updated preliminary triage result. This updated result is then sent to the dynamic decision-maker, which combines the latest medical resource status data to perform a new round of decision calculations, ultimately outputting an updated triage decision plan. This new plan may adjust the patient's treatment priority, reassign the relevant department, or change the reassurance strategy. Through this complete, automated reassessment chain, the system achieves rapid detection, timely response, and dynamic correction of abnormal situations, upgrading traditional static one-time triage to dynamic and continuous triage, greatly improving the responsiveness and safety of medical services.

[0063] Example 2 like Figure 2As shown, in a second aspect, the present invention provides an intelligent triage system for outpatient and emergency departments. The system employs the method provided in any of the above embodiments, and the system includes: The data acquisition module is used to collect patients' physiological data, symptom description text data, facial image data, and voice and audio data based on interactions with patients and through a data acquisition system configured with multimodal sensors. The emotion state analysis module is used to generate emotion state data to quantify the patient's current emotional state based on facial image data and voice audio data through an emotion computing engine. The urgency analysis module is used to generate urgency data that characterizes the severity of a patient's condition based on physiological data and symptom description text data through an analysis model. The data fusion and triage module is used to integrate emotional state data and disease urgency data. Through the decision fusion module, it generates preliminary triage results that include physiological and psychological analysis dimensions. The dynamic decision-making and resource scheduling module is used to generate a triage decision plan based on the preliminary triage results and the medical resource status data obtained in real time from the hospital information system, through a dynamic decision-maker. This plan includes the allocation of the treatment department, the processing priority, and the personalized comfort strategy. The decision execution and patient interaction module is used to execute departmental guidance instructions and initiate personalized reassurance strategies based on the triage decision plan, through the execution terminal and interactive robot.

[0064] Specifically, the hardware and software components of an intelligent triage system for outpatient and emergency departments are specifically configured to execute the methods provided in any of the above embodiments, thus forming an organically unified whole. The system includes a data acquisition module that integrates multimodal sensors, such as vital sign monitoring devices, interactive input terminals, and camera / audio pickup devices. Its core function is to efficiently and synchronously collect multi-dimensional raw data during patient interaction, including physiological data, symptom description text data, facial image data, and voice / audio data, providing a comprehensive and reliable data source for the entire intelligent triage process. The emotion state analysis module, as the system's "emotion perception center," has a built-in emotion computing engine. This engine receives facial image data and voice / audio data from the data acquisition module and, through a series of complex feature extraction, fusion, and classification calculations, ultimately outputs quantified and objective emotion state data, thereby transforming elusive emotions into calculable analytical factors.

[0065] The urgency analysis module acts as the "medical analysis center," receiving physiological data and symptom description text data. This module utilizes analytical models and natural language processing technology to standardize physiological parameters, semantically analyze and vectorize textual symptoms, and ultimately calculates and generates urgency data representing the severity of the patient's physiological condition, providing a core medical basis for triage. The data fusion and triage module is the system's first decision fusion point. It receives output data from the emotion state analysis module and the urgency analysis module, employing a specific fusion algorithm to integrate and evaluate psychological and physiological state indicators, generating a preliminary triage result that simultaneously incorporates physiological and psychological analysis dimensions, making the patient assessment more comprehensive and multi-dimensional.

[0066] The dynamic decision-making and resource scheduling module serves as the system's "command and dispatch center." It introduces a key external variable: real-time medical resource status data acquired from the hospital information system. Taking preliminary triage results as input, and under real-time resource constraints, this module performs comprehensive calculations through a dynamic decision-maker, ultimately outputting an executable triage decision plan that includes departmental allocation, processing priority, and personalized reassurance strategies. This achieves real-time linkage and optimization between triage decision-making and resource scheduling. The decision execution and patient interaction module acts as the system's "hands and feet." It receives the triage decision plan and executes departmental guidance instructions and initiates personalized reassurance strategies through physical devices such as execution terminals (e.g., screens, printers) and interactive robots. This completes the final link from intelligent decision-making to physical execution, ultimately achieving the comprehensive goals of improving triage efficiency, optimizing resource utilization, and enhancing humanistic care.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent triage in outpatient and emergency departments, characterized in that, include: S1: Based on the interaction with the patient, a data acquisition system with multimodal sensors is configured to collect the patient's physiological data, symptom description text data, facial image data, and voice audio data; S2: Based on facial image data and voice audio data, an emotion computing engine is used to generate emotion state data to quantify the patient's current emotional state. S3: Based on physiological data and symptom description text data, through analysis models, generate data on the urgency of the patient's condition, which characterizes the severity of the patient's condition. S4: By integrating emotional state data and disease urgency data, the decision fusion module generates preliminary triage results that include physiological and psychological analysis dimensions. S5: Based on the preliminary triage results and the medical resource status data obtained in real time from the hospital information system, a triage decision plan is generated through a dynamic decision-maker, which includes the allocation of the treatment department, the processing priority, and the personalized comfort strategy. S6: Based on the triage decision-making scheme, the system executes departmental guidance instructions and initiates personalized reassurance strategies through the execution terminal and interactive robot.

2. The intelligent triage method for outpatient and emergency departments according to claim 1, characterized in that, S1 includes: S101: Based on the patient's initial state upon arrival at the hospital, physiological data including body temperature, heart rate, and blood oxygen saturation are automatically collected through vital sign monitoring equipment; S102: A human-computer interaction interface based on a triage desk or self-service terminal, receiving text data describing symptoms, including chief complaints and medical history, input by the patient or accompanying person; S103: Based on non-contact contact with the patient's face, facial image data including facial expressions and facial blood flow dynamics are acquired through a visual sensor; S104: Based on the question-and-answer interaction process with the patient, voice audio data containing tone and speech rate characteristics is collected through a microphone.

3. The intelligent triage method for outpatient and emergency departments according to claim 1, characterized in that, S2 includes: S201: Based on facial image data, feature extraction is performed using a convolutional neural network model to obtain image feature vectors representing facial expression features; S202: Based on speech audio data, noise reduction and feature analysis are performed through acoustic models to obtain acoustic feature vectors that characterize the emotional features of speech. S203: By fusing image feature vectors and acoustic feature vectors, a unified multimodal emotion feature representation is obtained through a multimodal fusion algorithm; S204: Based on multimodal emotion feature representation, through a pre-trained emotion classification model, it outputs emotion state data to quantify the patient's current emotional state.

4. The intelligent triage method for outpatient and emergency departments according to claim 1, characterized in that, S3 includes: S301: Based on physiological data, standardized physiological parameters that meet the model input requirements are obtained through data standardization processing; S302: Based on symptom description text data, keyword extraction and semantic encoding are performed using natural language processing technology to obtain structured symptom vectors; S303: Integrates standardized physiological parameters and structured symptom vectors to form a comprehensive feature input; S304: Based on comprehensive feature input, the system calculates and outputs data on the urgency of the patient's condition, representing the severity of the patient's illness, through an analytical model.

5. The intelligent triage method for outpatient and emergency departments according to claim 1, characterized in that, Personalized appeasement strategies generated in S5 include: S501: Based on the emotional state data in the preliminary triage results, the basic soothing strategy type is obtained by querying the preset strategy mapping table; S502: Based on the real-time availability of medical staff and the working status of equipment in the medical resource status data, the execution subject and intensity of the basic reassurance strategy type are adaptively adjusted to generate personalized reassurance strategies. The personalized reassurance strategies include reassurance action instructions executed by the interactive robot or audiovisual content instructions played by the multimedia system.

6. The intelligent triage method for outpatient and emergency departments according to claim 5, characterized in that, The adaptive adjustments in S502 include: S5021: Based on medical resource status data, determine whether the current workload of medical staff exceeds a preset threshold; S5022: If the current workload of medical staff exceeds the preset threshold, select a personalized reassurance strategy with automated equipment as the main implementer. S5023: If the current workload of medical staff is lower than the preset threshold, add an auxiliary plan to the personalized comfort strategy that suggests that medical staff provide artificial psychological intervention.

7. The intelligent triage method for outpatient and emergency departments according to claim 1, characterized in that, S5 include: S511: Based on the urgency data of the initial triage results, generate an initial treatment priority sequence for patients using a sorting algorithm; S512: Based on medical resource status data, the corresponding medical department and medical equipment resources are allocated to patients in the initial processing priority sequence through a resource matching algorithm; S513: Integrate the initial processing priority sequence, allocated departmental and medical equipment resources, and personalized reassurance strategies to form a triage decision plan.

8. The intelligent triage method for outpatient and emergency departments according to claim 1, characterized in that, Following S6, it also includes: S7: Based on positioning beacons deployed within the hospital and smart tags worn by patients, it continuously tracks patient locations through a real-time positioning system and generates patient flow data; S8: Based on patient flow data, monitor the movement and dwell status of patients between predefined treatment nodes; S9: In response to monitoring abnormal patient status or waiting time exceeding expectations, an updated triage decision plan is generated by triggering a reassessment mechanism.

9. The intelligent triage method for outpatient and emergency departments according to claim 8, characterized in that, S9 includes: S901: Based on patient flow data, if the time a patient spends at a certain node exceeds a preset time limit associated with the patient's priority, or if an alarm is received indicating abnormal vital signs of the patient, a reassessment trigger signal is generated. S902: Based on the reassessment trigger signal, it automatically calls the data acquisition system and the emotion computing engine to re-acquire the patient's current data and generate updated emotion state data and disease urgency data; S903: Based on updated emotional state data and disease urgency data, an updated triage decision plan is generated again through the decision fusion module and dynamic decision maker.

10. An intelligent triage system for outpatient and emergency departments, characterized in that, The system employs the method according to any one of claims 1 to 9, the system comprising: The data acquisition module is used to collect patients' physiological data, symptom description text data, facial image data, and voice and audio data based on interactions with patients and through a data acquisition system configured with multimodal sensors. The emotion state analysis module is used to generate emotion state data to quantify the patient's current emotional state based on facial image data and voice audio data through an emotion computing engine. The urgency analysis module is used to generate urgency data that characterizes the severity of a patient's condition based on physiological data and symptom description text data through an analysis model. The data fusion and triage module is used to integrate emotional state data and disease urgency data. Through the decision fusion module, it generates preliminary triage results that include physiological and psychological analysis dimensions. The dynamic decision-making and resource scheduling module is used to generate a triage decision plan based on the preliminary triage results and the medical resource status data obtained in real time from the hospital information system, through a dynamic decision-maker. This plan includes the allocation of the treatment department, the processing priority, and the personalized comfort strategy. The decision execution and patient interaction module is used to execute departmental guidance instructions and initiate personalized reassurance strategies based on the triage decision plan, through the execution terminal and interactive robot.

Images