Voice triage device

Abstract

The invention relates to the technical field of triage devices, in particular to a voice triage device, which comprises a voice acquisition module used for acquiring oral symptom information of a patient; converting the symptom information into audio data; the voice recognition module is in communication connection with the voice acquisition module; the symptom analysis module is in communication connection with the voice recognition module and generates symptom feature vectors based on the symptom keywords; the intelligent triage module is in communication connection with the symptom analysis module and receives the symptom feature vector; calculating the similarity between the symptom feature vector and each disease based on a preset disease symptom database; according to the similarity, generating a preliminary diagnosis result and a recommended doctor seeing department; the output module is in communication connection with the intelligent triage module and is used for receiving the preliminary diagnosis result and a recommended treatment department; the preliminary diagnosis result and the recommended diagnosis department are output in a text or voice form, and a comprehensive and efficient voice triage system is constructed by combining advanced technologies such as voice recognition, natural language processing, machine learning and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of triage devices, and particularly to a voice triage device. Background Art

[0002] With the continuous growth of medical service demands and the increasing pressure on the allocation of medical resources, how to improve the operational efficiency of hospitals, especially optimize the patient visit process, has become an important challenge in the current medical field. In this context, the importance of hospital triage systems has become increasingly prominent. Although the traditional manual triage method can meet the basic needs to a certain extent, its limitations have become more obvious in the face of the increasing number of patients and the increasingly complex medical needs.

[0003] In recent years, with the rapid development of artificial intelligence technology, intelligent triage systems have begun to emerge in the medical field. These systems usually adopt the text input method, requiring patients to input their symptom descriptions through a touch screen or keyboard. However, this method is difficult to use for the elderly, groups with low education levels, or patients with physical disabilities. At the same time, the text input method is also prone to information omission or inaccurate expression, affecting the quality of triage.

[0004] On the other hand, although some existing voice interaction systems have achieved good application effects in certain fields, there are still many deficiencies in the highly professional field of medical triage. First of all, these systems often lack an in-depth understanding of medical terminology and are difficult to accurately identify and analyze the key symptom information in the patient's description. Secondly, they usually use simple rule matching or keyword retrieval methods for triage and cannot effectively handle complex symptom combinations or ambiguous expressions. Moreover, these systems often operate in isolation and are difficult to be effectively integrated with the existing information systems in hospitals, resulting in the formation of information islands and affecting the efficiency of the overall medical service.

[0005] In addition, existing intelligent triage systems also have deficiencies in data security and privacy protection. As extremely sensitive personal privacy data, the security and confidentiality of medical information are of crucial importance. However, many systems do not consider this aspect comprehensively enough and there is a risk of data leakage.

[0006] In view of the above problems, there is an urgent need for an intelligent triage system that can make full use of voice interaction technology and at the same time has the ability to understand professional medical knowledge. This system should be able to accurately identify the patient's oral symptoms, intelligently analyze the symptom characteristics, give reasonable preliminary diagnoses and department recommendations, and be seamlessly integrated with the existing hospital systems. At the same time, it is also necessary to ensure the security of data and the protection of patient privacy. Summary of the Invention

[0007] The voice triage device of the present invention is precisely proposed to address the above technical problems. It aims to construct an intelligent, efficient, and secure voice triage system that provides patients with convenient and accurate preliminary diagnoses and medical guidance through voice interaction.

[0008] The present invention proposes a voice triage device, comprising:

[0009] A voice acquisition module, configured to:

[0010] Obtain the symptom information dictated by the patient;

[0011] Convert the symptom information into audio data;

[0012] A voice recognition module, communicatively connected to the voice acquisition module, configured to:

[0013] Receive the audio data;

[0014] Convert the audio data into text data;

[0015] A symptom analysis module, communicatively connected to the voice recognition module, configured to:

[0016] Receive the text data;

[0017] Extract symptom keywords from the text data;

[0018] Generate a symptom feature vector based on the symptom keywords;

[0019] An intelligent triage module, communicatively connected to the symptom analysis module, configured to:

[0020] Receive the symptom feature vector;

[0021] Based on a preset disease symptom database, calculate the similarity between the symptom feature vector and each disease;

[0022] Generate a preliminary diagnosis result and recommended department for medical treatment according to the similarity;

[0023] An output module, communicatively connected to the intelligent triage module, configured to:

[0024] Receive the preliminary diagnosis result and recommended department for medical treatment;

[0025] Output the preliminary diagnosis result and recommended department for medical treatment in text or voice form.

[0026] Preferably, it further includes a knowledge base module, and the knowledge base module includes:

[0027] A keyword database for storing keywords related to symptoms;

[0028] Symptom database, used to store symptom descriptions of various diseases;

[0029] Diagnosis database, used to store disease diagnosis information;

[0030] Department database, used to store professional information of each department in the hospital;

[0031] Expert database, used to store the professional expertise information of doctors;

[0032] Among them, the intelligent triage module is communicatively connected to the knowledge base module, and is used to obtain relevant information from the knowledge base module to support triage decisions.

[0033] Preferably, the symptom analysis module is further used for:

[0034] Performing natural language processing on the text data to identify semantic information such as negative words and degree words;

[0035] Quantitatively evaluating the severity of symptoms based on the semantic information;

[0036] Integrating the quantitative evaluation results into the symptom feature vector.

[0037] Preferably, the intelligent triage module further includes:

[0038] A machine learning unit, used for:

[0039] Training a triage model based on historical triage data;

[0040] Using the triage model to classify new symptom feature vectors to generate triage results;

[0041] An optimization unit, used for:

[0042] Collecting feedback on actual diagnosis results;

[0043] Updating the triage model based on the feedback to improve the triage accuracy.

[0044] Preferably, it further includes a personalized recommendation module, and the personalized recommendation module is communicatively connected to the intelligent triage module, and is used for:

[0045] Receiving the personal information of the patient, including age, gender, past medical history, etc.;

[0046] Generating personalized medical treatment suggestions based on the personal information and the preliminary diagnosis results;

[0047] Transmitting the personalized medical treatment suggestions to the output module.

[0048] Preferably, it further includes a multi-dimensional evaluation module, which is communicatively connected to the intelligent triage module and is used for:

[0049] Evaluating the severity of symptoms;

[0050] Analyzing potential disease risks;

[0051] Considering the expertise and resource situation of each department;

[0052] Calculating the current medical treatment pressure and waiting time;

[0053] Based on the above factors, optimizing and adjusting the preliminary diagnosis result.

[0054] Preferably, the speech recognition module further includes:

[0055] A noise reduction unit, which is used for:

[0056] Performing noise reduction processing on the audio data to improve the speech recognition accuracy;

[0057] A multi-language support unit, which is used for:

[0058] Identifying the language type of the audio data;

[0059] Selecting the corresponding speech recognition model for conversion according to the identified language type.

[0060] Preferably, it further includes a data security module, which is communicatively connected to all other modules and is used for:

[0061] Encrypting the data in transmission;

[0062] Performing desensitization processing on the stored patient information;

[0063] Controlling data access permissions to prevent unauthorized access;

[0064] Recording the system operation logs for easy traceability and auditing.

[0065] Preferably, it further includes a remote collaboration module, which is communicatively connected to the intelligent triage module and is used for:

[0066] Establishing a video connection with remote experts in case of complex cases;

[0067] Transmitting the triage-related information to remote experts in real time;

[0068] Receiving the diagnostic opinions of remote experts and integrating them into the triage results.

[0069] Preferably, it further includes a system interface module, which is used for:

[0070] Perform data interaction with the hospital's existing HIS system;

[0071] Receive the hospital's scheduling information for optimizing visit recommendations;

[0072] Transmit the triage results back to the HIS system for patient visit tracking;

[0073] Provide an API interface to support integration with other medical devices or systems.

[0074] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0075] Through the innovative combination of advanced technologies such as speech recognition, natural language processing, and machine learning, the present invention constructs a comprehensive and efficient voice triage system. This system can not only accurately understand the patient's oral symptoms but also conduct in-depth symptom analysis and intelligent diagnostic reasoning. By introducing a knowledge base module, the system can continuously update and expand medical knowledge, continuously improving the accuracy of triage. At the same time, the modular design and standardized interface of the system enable it to be seamlessly integrated with the hospital's existing information systems, realizing the efficient circulation of data and the full utilization of resources.

[0076] From a macroscopic perspective, the voice triage device of the present invention can significantly improve the overall operation efficiency of the hospital. It can greatly reduce the waiting time of patients, optimize the allocation of medical resources, and improve the smoothness of the visit process. This can not only enhance the patient's medical experience but also help the hospital better cope with peak visiting hours and relieve the problem of tight medical resources.

[0077] From a microscopic perspective, the present invention has innovations and breakthroughs in multiple technical points. For example, in speech recognition, the system adopts advanced deep learning algorithms, which can accurately recognize various accents and expressions, greatly improving the recognition accuracy. In symptom analysis, the system can not only recognize specific symptoms but also understand the correlation between symptoms, thus conducting a more comprehensive condition assessment. In diagnostic reasoning, the system adopts a reasoning mechanism based on a knowledge graph, which can simulate the doctor's diagnostic thinking and give more reliable preliminary diagnostic results.

[0078] In addition, the present invention also pays special attention to the security and privacy protection of the system. By adopting advanced encryption technologies and access control mechanisms, the system can effectively protect the patient's privacy information and prevent data leakage. This not only meets the requirements of relevant laws and regulations but also enhances the patient's trust in the system, promoting the wide application of the system.

[0079] Generally speaking, the voice triage device of the present invention innovatively solves multiple key problems existing in the prior art, achieving an organic unity of the convenience of voice interaction, the intelligence of the triage process, the accuracy of diagnosis results, the seamless integration of the system, and the security of data protection. It can not only significantly improve the operation efficiency and service quality of hospitals, but also make important contributions to promoting the development of smart healthcare. With the continuous optimization and improvement of the system, it is expected to become an important driving force for the transformation of future medical service models, bringing more convenient, efficient, and user-friendly medical experiences to patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a logic block diagram of the voice triage device of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0081] Refer to Figure 1 , first of all, the present invention provides a voice triage device, which can provide patients with rapid and accurate preliminary diagnoses and department recommendations through intelligent speech recognition and analysis technologies. The technical solutions of the present invention will be described in detail below.

[0082] The voice triage device of the present invention includes a voice acquisition module 1, a speech recognition module 2, a symptom analysis module 3, an intelligent triage module 4, and an output module 5. These modules work together to jointly realize the whole process from the description of patient symptoms to the output of preliminary diagnosis results.

[0083] Specifically, the voice acquisition module 1 is used to obtain the symptom information described orally by the patient and convert it into audio data. In practical applications, this module can be a high-sensitivity microphone device, which can effectively capture the patient's voice and has a certain noise reduction function to improve the accuracy of subsequent speech recognition. Preferably, the microphone device adopted in the present invention has a sensitivity of up to -38dB and a signal-to-noise ratio of not less than 60dB. Such parameter settings can clearly capture the patient's voice even in a noisy hospital environment.

[0084] The speech recognition module 2 is communicatively connected to the voice acquisition module 1 and is used to receive the audio data and convert it into text data. The present invention uses advanced deep learning algorithms to achieve speech recognition. Specifically, a speech recognition model based on the Transformer architecture can be used. The mathematical expression of this model is as follows:

[0085]

[0086] where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key vector. Through this self-attention mechanism, the model can effectively capture the long-distance dependence relationships in the speech and improve the recognition accuracy.

[0087] In one embodiment of the present invention, the character error rate (CER) of the speech recognition module 2 can be controlled below 5%. This means that when transcribing 100 characters, on average only 5 characters may be incorrect. This level of accuracy is sufficient to ensure the reliability of subsequent symptom analysis.

[0088] The symptom analysis module 3 is communicatively connected to the speech recognition module 2 and is used to receive text data, extract symptom keywords therefrom, and generate symptom feature vectors. This module employs natural language processing techniques, particularly the named entity recognition (NER) algorithm to identify symptom descriptions in the text. Specifically, the present invention uses a model based on BiLSTM-CRF (bidirectional long short-term memory network - conditional random field), and its mathematical expression is as follows:

[0089]

[0090] Where X is the input sequence, Y is the predicted label sequence, and φ is the feature function. This model can effectively capture context information and improve the accuracy of symptom recognition.

[0091] In a preferred embodiment of the present invention, the symptom analysis module 3 also quantifies the severity of the identified symptoms. For example, a quantification scale of 0 - 10 can be set, where 0 indicates no symptoms and 10 indicates the most severe. The system will give a quantification score based on the patient's description, such as words like mild or severe, combined with the characteristics of the symptoms themselves. This score will be an important dimension of the symptom feature vector.

[0092] The intelligent triage module 4 is communicatively connected to the symptom analysis module 3 and is used to receive the symptom feature vectors, calculate the similarity based on a preset disease symptom database, and generate a preliminary diagnosis result and recommended departments for medical treatment. This module employs a matching algorithm based on cosine similarity, and its mathematical expression is:

[0093]

[0094] Where A is the symptom feature vector of the patient, and B is the feature vector of each disease in the database. By calculating the cosine similarity, the system can find the most similar disease and thus give a preliminary diagnosis recommendation.

[0095] In practical applications, the present invention sets a similarity threshold, for example, 0.7. When the highest similarity exceeds 0.7, the system will give a clear preliminary diagnosis suggestion; when the highest similarity is between 0.5 and 0.7, the system will give a possible diagnosis direction but recommends further examination; when the highest similarity is below 0.5, the system will prompt that the symptom description may be insufficient and recommends that the patient supplement more information.

[0096] The output module 5 is communicatively connected to the intelligent triage module 4, and is used to receive the preliminary diagnosis result and the recommended department for treatment, and output it in text or voice form. When outputting in voice, the present invention adopts a voice synthesis technology based on Wave Net, which can generate natural and fluent voices, improving the user experience.

[0097] The voice triage device of the present invention further includes a knowledge base module 6. This module is the core data support of the entire system and includes a keyword database, a symptom database, a diagnosis database, a department database, and an expert database. The design of these databases fully considers the professionalism and complexity of the medical field.

[0098] For example, in the symptom database, each symptom not only includes its description but also possible variant expressions. For example, for the symptom of headache, the database will include synonymous expressions such as "headache" and "head pain" to improve the accuracy of matching. At the same time, the symptom will be associated with the diseases that may occur and the probability of occurrence will be marked, which provides an important basis for subsequent intelligent triage.

[0099] In the department database, in addition to the basic department information, it also includes information such as the specialized diseases of each department, the main equipment, and the representative experts. This information helps the system to more comprehensively consider the needs of patients when recommending departments for treatment.

[0100] The knowledge base module 6 is communicatively connected to the intelligent triage module 4 to provide data support for triage decisions. In actual operation, the knowledge base will be continuously updated and optimized. For example, a weekly update frequency can be set to update the data according to the latest medical research results and clinical practice experience. This dynamic update mechanism ensures that the system can always provide the latest and most accurate triage suggestions.

[0101] The symptom analysis module 3 of the present invention also has more in-depth natural language processing capabilities. Specifically, this module can identify semantic information such as negative words and degree words, quantitatively evaluate the severity of symptoms, and integrate the evaluation results into the symptom feature vector.

[0102] When identifying negative words, the system will pay special attention to words such as "not", "have not", "not at all", etc. For example, when the patient says "I don't have a fever", the system will correctly understand that this is a negation of the fever symptom and will not misjudge it as having a fever symptom. When identifying degree words, the system will focus on expressions such as "slight", "severe", "intense", etc. For example, although both slight headache and intense headache are headache symptoms, the difference in severity will directly affect the subsequent diagnostic suggestions.

[0103] In terms of quantitative assessment, the present invention adopts a scoring system from 0 to 10. 0 represents no symptoms at all, and 10 represents extremely severe symptoms. The system will give a quantitative score based on the identified degree words and in combination with the characteristics of the symptoms themselves. For example:

[0104] A slight headache may be rated 2 - 3 points;

[0105] A moderate headache may be rated 4 - 6 points;

[0106] An intense headache may be rated 7 - 9 points;

[0107] An unbearable headache may be rated 10 points

[0108] The result of this quantitative assessment will be an important dimension of the symptom feature vector. For example, the feature vector of a headache symptom may look like:

[0109] Symptom vector = [headache, 7, duration, 24, location, forehead]

[0110] Among them, 7 is the quantitative score of the headache severity, 24 is the duration (hours), and forehead is the pain location.

[0111] Through this meticulous semantic analysis and quantitative assessment, the voice triage device of the present invention can more accurately understand the patient's symptom description, thereby providing a more accurate preliminary diagnosis and department recommendation. This not only improves the accuracy of triage but also provides more valuable reference information for doctors to assist them in making better diagnostic decisions.

[0112] The application of artificial intelligence technology in the medical field is a challenging topic. The voice triage device of the present invention realizes an intelligent, efficient, and user-friendly triage system by combining advanced speech recognition technology, natural language processing technology, and medical knowledge bases. This system can not only improve the triage efficiency of the hospital, reduce the workload of medical staff, but also provide a more convenient and accurate medical experience for patients. In the future, with the continuous progress of technology and the continuous update of medical knowledge, the voice triage device of the present invention still has great room for optimization and expansion and is expected to become an important part of the intelligent medical field.

[0113] On the basis described above, the voice triage device of the present invention also has more advanced functions and optimized features, which will be described in detail below.

[0114] The intelligent triage module 4 of the present invention further includes a machine learning unit and an optimization unit. The introduction of these two units enables the system to have the ability of self-learning and continuous optimization, greatly improving the accuracy and adaptability of triage.

[0115] The machine learning unit is mainly responsible for training a triage model based on historical triage data and using the model to classify new symptom feature vectors to generate triage results. In a preferred embodiment of the present invention, a classification model based on a deep neural network is adopted. The mathematical expression of this model can be simplified as:

[0116] y = f(Wx + b),

[0117] where x is the input symptom feature vector, W is the weight matrix, b is the bias vector, f is the activation function, and y is the output classification result. In practical applications, the present invention uses a multi-layer perceptron (MLP) structure, usually including 3 - 5 hidden layers, and the number of neurons in each layer is adjusted according to specific circumstances, generally between 100 - 500. The activation function selects ReLU (Rectified Linear Unit), and its mathematical expression is:

[0118] f(x) = max(0, x),

[0119] This structure can effectively capture the complex non-linear relationships between symptom features and improve the classification accuracy. The main function of the optimization unit is to collect feedback of actual diagnosis results and update the triage model based on these feedbacks to continuously improve the triage accuracy. The present invention adopts an online learning method, that is, whenever a new feedback is received, the model is immediately updated slightly. The update method adopts the stochastic gradient descent method, and its mathematical expression is

[0120]

[0121] where W t is the current weight, η is the learning rate, is the gradient of the loss function.

[0122] In an embodiment of the present invention, the learning rate η is initially set to 0.01, and an adaptive learning rate strategy is adopted, that is, when the directions of consecutive updates are the same, the learning rate is slightly increased, and vice versa. This strategy can accelerate the convergence speed while ensuring the stability of the model.

[0123] Through the collaborative work of the machine learning unit and the optimization unit, the voice triage device of the present invention can continuously learn from actual diagnosis results and continuously improve the triage accuracy rate. Preferably, the triage accuracy rate of the system can reach about 70% in the initial stage, and can be increased to more than 85% after 3 to 6 months of continuous optimization.

[0124] The present invention further includes a personalized recommendation module 7. This module is communicatively connected to the intelligent triage module 4 and can generate more personalized medical treatment suggestions based on the patient's personal information and preliminary diagnosis results.

[0125] The personalized recommendation module 7 first receives the patient's personal information, including but not limited to age, gender, past medical history, etc. This information is usually obtained through patient registration or docking with the hospital HIS system. In a preferred embodiment of the present invention, the personal information may further include the patient's living habits, occupational characteristics, etc., and this information can be obtained through a simple questionnaire survey.

[0126] Next, the personalized recommendation module 7 combines this personal information and the preliminary diagnosis results and uses a rule-based expert system to generate personalized medical treatment suggestions. This expert system contains a large number of if-then rules, such as:

[0127] If the patient's age > 60 years AND the preliminary diagnosis result contains 'heart disease' THEN it is recommended to give priority to seeing a cardiologist AND it is recommended to perform an electrocardiogram examination

[0128] If the patient's gender = 'female' AND the age is between 20 - 45 AND the preliminary diagnosis result contains 'abdominal pain' THEN it is recommended to consider a gynecological examination

[0129] These rules are formulated by medical experts based on clinical experience, can take into account the individual differences of patients, and provide more targeted medical treatment suggestions.

[0130] In practical applications, the personalized recommendation module 7 of the present invention will also consider the actual situation of the hospital, such as the expert schedule of each department, equipment situation, etc., so as to give the optimal medical treatment time and specific doctor recommendation. For example, if the system recommends that the patient go to the cardiologist for treatment, it will further query the expert outpatient schedule of the cardiologist and recommend the most suitable medical treatment time and expert.

[0131] Finally, the personalized recommendation module 7 will transmit the generated personalized medical treatment suggestions to the output module 5 and present them to the patient in the form of text or voice. This personalized recommendation greatly improves the accuracy of triage and the patient's medical treatment experience.

[0132] The present invention further includes a multi-dimensional evaluation module 8. This module is communicatively connected to the intelligent triage module 4 and can evaluate and optimize the preliminary diagnosis results from multiple perspectives.

[0133] The multi-dimensional evaluation module 8 mainly considers the following aspects:

[0134] 1. Severity of symptoms: Based on the quantitative evaluation results of the symptom analysis module 3, judge the urgency of the symptoms. For example, if the severity score of headache exceeds 8 points, the system will increase the urgency of seeking medical treatment.

[0135] 2. Potential disease risks: Based on the patient's personal information and symptom characteristics, evaluate the possible potential risks. For example, if a male patient over 50 years old has chest tightness symptoms, the system will consider the risk of cardiovascular disease and recommend seeking medical treatment in a timely manner.

[0136] 3. Expertise and resource status of each department: Consider the professional expertise and current resource status of each department in the hospital. For example, if a certain department has recently introduced new diagnostic equipment, the system will appropriately increase the recommended priority of this department.

[0137] 4. Current medical treatment pressure and waiting time: Obtain the number of patients seeking medical treatment and the average waiting time of each department in real time, and try to balance the medical treatment pressure of each department on the premise of ensuring medical quality.

[0138] The multi-dimensional evaluation module 8 adopts a weighted scoring method to comprehensively consider these factors. Its mathematical model can be expressed as:

[0139]

[0140] where Score i is the final score of the i-th department, w j is the weight of the j-th factor, and f j (i) is the scoring function of the i-th department on the j-th factor.

[0141] In a preferred embodiment of the present invention, the initial weights of each factor are set as follows:

[0142] Severity of symptoms: 0.3;

[0143] Potential disease risks: 0.3;

[0144] Degree of match of department expertise: 0.2;

[0145] Department resource status: 0.1;

[0146] Current medical treatment pressure: 0.1;

[0147] These weights can be dynamically adjusted according to the actual situation and management requirements of the hospital. The system will select the department with the highest score as the final recommended result.

[0148] By leveraging the functions of the multi-dimensional evaluation module 8, the voice triage device of the present invention can comprehensively consider the actual situation of patients and the operation status of the hospital when giving diagnostic suggestions, thereby providing more reasonable and effective triage results. This not only improves the medical treatment efficiency of patients but also helps the hospital better manage medical resources and enhance the overall service quality.

[0149] The voice recognition module 2 of the present invention further includes a noise reduction unit 21 and a multi-language support unit 22. The addition of these two units significantly improves the adaptability and user-friendliness of the system.

[0150] The noise reduction unit 21 is mainly used for noise reduction processing of audio data to improve the accuracy of voice recognition. In a noisy environment such as a hospital, effective noise reduction processing is particularly important. The present invention adopts a noise reduction algorithm based on deep learning, and its core is a denoising autoencoder (DAE). Its mathematical model can be expressed as:

[0151]

[0152] where x is the original audio signal, n is the noise, g is the encoding function, f is the decoding function, is the reconstructed clear signal.

[0153] In practical applications, the present invention uses a multi-layer convolutional neural network (CNN) to implement the encoding and decoding functions. This structure can effectively capture the time-frequency characteristics of audio signals and achieve high-quality noise reduction effects. Preferably, the noise reduction processing of this system can reduce the background noise by 15 - 20 decibels without losing effective information, significantly improving the accuracy of subsequent voice recognition.

[0154] The introduction of the multi-language support unit 22 enables the voice triage device of the present invention to recognize and process inputs in multiple languages. This is particularly useful in multi-lingual regions or when dealing with foreign patients. This unit first identifies the language type of the input audio and then selects the corresponding voice recognition model for conversion.

[0155] In a preferred embodiment of the present invention, the multi-language support unit 22 adopts the i-vector-based language recognition technology. Its core idea is to map the voice signal to a low-dimensional space and then perform language classification in this space. Its mathematical model can be simplified as:

[0156] M = m + Tw,

[0157] Among them, M is the supervector of the speech signal, m is the supervector of the Universal Background Model (UBM), T is the Total Variability Matrix, and w is the i-vector.

[0158] The system will pre-train i-vector models for various languages. When new audio is input, by calculating its i-vector and comparing it with the language models, the language type to which it belongs can be determined. Preferably, the number of languages supported by this system is not less than 10, including but not limited to Chinese (Mandarin and major dialects), English, Japanese, Korean, French, German, etc.

[0159] After determining the language type, the system will automatically switch to the corresponding speech recognition model. These models are all based on deep learning technology and optimized according to the characteristics of each language. For example, for tonal languages (such as Chinese), the model will pay special attention to the pitch changes; for inflectional languages (such as Russian), the model will focus more on the recognition of word form changes.

[0160] Through the collaborative work of the noise reduction unit 21 and the multilingual support unit 22, the voice triage device of the present invention can accurately identify the symptom descriptions of patients from different language backgrounds in various complex environments. This greatly expands the application scope of the system, enabling it to play an important role in the international medical environment.

[0161] Generally speaking, the present invention constructs an intelligent, flexible and efficient voice triage system by introducing machine learning, personalized recommendation, multi-dimensional evaluation and advanced speech processing technology. This system can not only provide accurate preliminary diagnosis and department recommendation, but also self-optimize and make personalized adjustments according to the actual situation. Its application will greatly improve the triage efficiency of the hospital, improve the medical experience of patients, and make important contributions to promoting the development of smart healthcare. On the basis of the above technical solutions, the voice triage device of the present invention also includes some key functional modules to further improve the security, collaboration ability and integration of the system. The specific implementation and functions of these modules will be described in detail below.

[0162] The present invention also includes a data security module. This module is communicatively connected to all other modules of the system and is responsible for ensuring the security of data and the protection of patient privacy during the entire triage process. In the context of increasing attention to data security and privacy protection today, the importance of the data security module is self-evident.

[0163] The primary task of the data security module is to encrypt the data in transit. The present invention adopts the Advanced Encryption Standard (AES) algorithm, specifically using the AES-GCM (Galois / Counter Mode) mode with a 256-bit key. This encryption method can not only ensure the confidentiality of data, but also guarantee the integrity and authenticity of data. Its encryption process can be simplified and expressed by the following mathematical formula:

[0164] C = E K (P, N, A),

[0165] where C is the ciphertext, E is the encryption function, K is the key, P is the plaintext, N is the nonce (random number), and A is the additional authentication data. In a preferred embodiment of the present invention, the system updates the nonce N every 15 minutes. In this way, even if an attacker obtains some encrypted data, they cannot infer the content of other data, greatly improving the security of the system. For the stored patient information, the data security module will perform desensitization processing. Specifically, the system adopts a reversible desensitization technology based on the hash function. Its mathematical model can be expressed as:

[0166] M = H(D || K),

[0167] where M is the desensitized data, H is the hash function, D is the original data, K is the key, and ∥ represents the concatenation operation.

[0168] The advantage of this method is that without knowing the key K, even if the desensitized data M is obtained, the original data D cannot be inferred. At the same time, for authorized operations, as long as the correct key K is provided, the original data can be quickly restored. In this way, it not only protects the privacy of patients, but also does not affect necessary medical operations.

[0169] The data security module is also responsible for controlling data access permissions. The present invention adopts the Role-Based Access Control (RBAC) model. In this model, the system predefines multiple roles, such as doctors, nurses, administrators, etc. Each role has specific data access permissions. When a user logs in to the system, they will be assigned the corresponding role and thus obtain the corresponding data access permissions. This method not only ensures the security of data, but also improves the efficiency of permission management.

[0170] Finally, the data security module will also record all operation logs of the system for easy traceability and auditing. The log information includes but is not limited to operation time, operator, operation type, operation object, etc. These logs are stored using blockchain technology to ensure that the log information cannot be tampered with, providing a reliable basis for subsequent security audits.

[0171] Through the multiple protection measures of the data security module, the voice triage device of the present invention can protect the privacy and data security of patients to the greatest extent while providing efficient services, meeting the high standards of modern medical systems.

[0172] The present invention further comprises a remote collaboration module 10. This module is in communication with the intelligent triage module 4 and is intended to handle some complex or difficult cases and improve the accuracy of triage through the participation of remote experts.

[0173] The core function of the remote collaboration module 10 is to establish a video connection with a remote expert in the case of complex cases. The present invention adopts a video communication solution based on WebRTC (Web Real-Time Communication) technology. This technology can achieve low-latency, high-quality point-to-point video communication, which is very suitable for the needs of medical scenarios.

[0174] When establishing a video connection, the system will use STUN (Session Traversal Utilities for NAT) and TURN (Traversal Using Relays around NAT) servers to solve possible NAT penetration problems and ensure that the connection can be established smoothly in various network environments. After the connection is established, the audio and video data will be encrypted and transmitted using SRTP (Secure Real-time Transport Protocol) to ensure the security of communication.

[0175] The remote collaboration module 10 is not just a simple video communication, it can also transmit triage-related information to remote experts in real time. This information includes the patient's symptom description, preliminary diagnosis results, related examination data, etc. The information is displayed using split-screen technology, and experts can view this key information while watching the video to make more accurate judgments.

[0176] In a preferred embodiment of the present invention, the remote collaboration module 10 also supports a real-time collaborative annotation function. The remote expert can directly annotate the patient's examination images (such as CT, MRI, etc.), and these annotations will be displayed synchronously on the screen of the on-site doctor in real time. This intuitive communication method greatly improves the efficiency and accuracy of remote consultation.

[0177] The diagnostic opinions given by the remote experts will be recorded by the system in real time, and key information will be extracted through natural language processing technology and integrated into the final triage results. This process can be represented by the following pseudo code:

[0178] ```

[0179] function integrateExpertOpinion(initialDiagnosis, expertOpinion):

[0180] keyPoints = NLP.extractKeyPoints(expertOpinion)

[0181] for point in keyPoints:

[0182] if point.confidence > initialDiagnosis.confidence:

[0183] initialDiagnosis.update(point)

[0184] return initialDiagnosis

[0185] ```

[0186] Through the introduction of the remote collaboration module 10, the voice triage device of the present invention breaks through the geographical restrictions, can quickly mobilize high-quality medical resources when dealing with complex cases, and greatly improves the accuracy of triage and the quality of medical services.

[0187] The present invention further includes a system interface module. The main function of this module is to achieve seamless docking with the existing hospital systems and provide integration interfaces for other medical devices or systems.

[0188] The first problem to be solved by the system interface module is the data interaction problem with the hospital HIS (Hospital Information System) system. The present invention adopts an interface design based on the HL7 (Health Level Seven) standard, which is a widely used standard in medical information systems. Specifically, the system uses the HL7 FHIR (Fast Healthcare Interoperability Resources) specification, which is the latest version of the HL7 standard and supports more flexible data exchange.

[0189] When interacting with the HIS system, the system interface module maps the internal data model to FHIR resources. For example, patient information is mapped to the FHIR Patient resource, and the diagnosis result is mapped to the Condition resource. This mapping can be represented by the following pseudocode:

[0190] ```

[0191] function mapToFHIR(internalData,resourceType):

[0192] fhirResource = FHIRFactory.create(resourceType)

[0193] for field in internalData:

[0194] if field in fhirResource.mapping:

[0195] fhirResource[fhirResource.mapping[field]] = internalData[field]

[0196] return fhirResource

[0197] ```

[0198] The system interface module is also responsible for receiving the hospital's scheduling information for optimizing appointment recommendations. This information typically includes the attending doctors in each department, their working hours, available appointment slots, etc. The system will regularly (e.g., every hour) obtain this information from the HIS system and update it in the local database to ensure the timeliness and accuracy of the recommendations.

[0199] When transmitting the triage results back to the HIS system, the system interface module generates a standard HL7 message. This message contains the patient's basic information, preliminary diagnosis results, recommended departments for treatment, etc. After receiving this message, the HIS system can automatically create corresponding medical records, facilitating the patient's subsequent medical treatment.

[0200] In addition to the connection with the HIS system, the system interface module also provides a set of RESTful APIs to support the integration with other medical devices or systems. This set of APIs uses the OAuth 2.0 protocol for authentication and authorization to ensure the security of access. The main functions of the APIs include, but are not limited to: obtaining triage results; submitting patient symptom information; querying department information; and making appointment reservations.

[0201] In a preferred embodiment of the present invention, the system interface module also supports the DICOM (Digital Imaging and Communications in Medicine) standard, which enables the voice triage device to directly exchange data with medical imaging devices (such as CT, MRI, etc.). This provides the possibility for further expanding the system functions in the future, such as incorporating imaging features into the triage considerations.

[0202] Through the design of the system interface module, the voice triage device of the present invention can be seamlessly docked with the existing information systems and various medical devices in the hospital, realizing the efficient circulation of data and the full utilization of resources. This not only improves the efficiency of the entire medical process but also provides a good foundation for future function expansion and system upgrade.

[0203] Generally speaking, by introducing a data security module, a remote collaboration module, and a system interface module, the present invention constructs a secure, collaborative, and open voice triage system. This system can not only protect patient privacy but also rely on the strength of remote experts in complex situations. At the same time, it has good integration and scalability. Its application will significantly improve the overall service level of the hospital and make an important contribution to promoting the development of intelligent healthcare.

[0204] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A voice triage device, characterized in that: include: Voice collection module, used for: Obtain symptom information verbally reported by patients; converting the symptom information into audio data; A speech recognition module is connected to the speech acquisition module for: receiving the audio data; Converting the audio data into text data; A symptom analysis module, in communication with the speech recognition module, is used to: receiving the text data; extracting symptom keywords from the text data; generating a symptom feature vector based on the symptom keywords; The intelligent triage module is connected to the symptom analysis module for: receiving the symptom feature vector; Based on a preset disease symptom database, calculating the similarity between the symptom feature vector and each disease; According to the similarity, a preliminary diagnosis result and a recommended consultation department are generated; An output module is connected to the intelligent triage module for: Receiving the preliminary diagnosis result and recommending the medical department; The preliminary diagnosis result and the recommended treatment department are output in text or voice form.

2. The voice triage device according to claim 1, characterized in that: Also included is a knowledge base module, the knowledge base module comprising: A keyword database for storing symptom-related keywords; Symptom database, used to store symptom descriptions of various diseases; Diagnosis database, used to store disease diagnosis information; Department database, used to store professional information of each department in the hospital; Expert database, used to store information on doctors’ professional expertise; The intelligent triage module is communicatively connected with the knowledge base module and is used to obtain relevant information from the knowledge base module to support triage decisions.

3. The voice triage device according to claim 1, characterized in that: The symptom analysis module is also used for: Performing natural language processing on the text data to identify semantic information such as negative words and degree words; Quantitatively assessing the severity of the symptoms based on the semantic information; The quantitative assessment results are integrated into the symptom feature vector.

4. The voice triage device according to claim 1, characterized in that: The intelligent triage module also includes: Machine Learning Unit for: Train triage models based on historical triage data; Using the triage model to classify the new symptom feature vector to generate a triage result; Optimization unit for: Collect feedback on actual diagnostic results; The triage model is updated based on the feedback to improve triage accuracy.

5. The voice triage device according to claim 1, characterized in that: It also includes a personalized recommendation module, which is communicatively connected with the intelligent triage module and is used for: Receive the patient's personal information, including age, gender, medical history, etc.; Generate personalized medical advice based on the personal information and the preliminary diagnosis result; The personalized medical consultation recommendation is transmitted to the output module.

6. The voice triage device according to claim 1, characterized in that: It also includes a multi-dimensional evaluation module, which is communicatively connected with the intelligent triage module and is used for: Assess the severity of symptoms; Analyze potential disease risks; Consider the expertise and resources of each department; Calculate current visit pressure and waiting time; Based on the above factors, the preliminary diagnostic results are optimized and adjusted.

7. The voice triage device according to claim 1, characterized in that: The speech recognition module also includes: Noise reduction unit for: Performing noise reduction processing on the audio data to improve the accuracy of speech recognition; Multi-language support unit for: Identifying a language type of the audio data; Select the corresponding speech recognition model for conversion according to the recognized language type.

8. The voice triage device according to claim 1, characterized in that: It also includes a data security module, which is connected to all other modules for: Encrypt data in transit; Desensitize stored patient information; Control data access rights to prevent unauthorized access; Record system operation logs for easy traceability and auditing.

9. The voice triage device according to claim 1, characterized in that: It also includes a remote collaboration module, which is communicatively connected with the intelligent triage module and is used for: In case of complex cases, establish a video link with a remote specialist; Transmit triage-related information to remote specialists in real time; Receive diagnostic opinions from remote experts and integrate them into triage results.

10. The voice triage device according to claim 1, characterized in that: Also included is a system interface module, the system interface module is used to: Interact data with the hospital's existing HIS system; Receive hospital scheduling information to optimize medical recommendations; Send triage results back to the HIS system for patient follow-up; Provides API interface to support integration with other medical devices or systems.

Images