Intelligent diagnosis guidance method, system and computer-readable storage medium
Through the deep learning interaction model and knowledge graph engine, patient information is analyzed and standardized medical vectors are generated, which solves the problem of patients' difficulty in selecting departments and doctors in hospital guidance, improves registration accuracy and efficiency, and simplifies the doctor's diagnosis process.
Patent Information
- Application Number
- CN202110889888.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-04
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-08-04
AI Technical Summary
Due to the lack of medical foundation for patients' medical consultation, it is difficult for patients to accurately select departments or doctors, which makes registration time-consuming and labor-intensive and may delay the condition. It is difficult for existing intelligent robots to understand the input of natural texts and generate standardized medical texts.
Use deep learning interaction model to analyze patient's diagnosis information, generate standardized medical vectors, and search for related symptoms through the knowledge graph engine, update vector information, predict symptoms, match departments and doctors, generate main complaints and current medical history drafts, and modify them with knowledge graphs. , import electronic medical records.
It improves the convenience and accuracy of the diagnosis, saves doctors' consultation time, improves the efficiency of visits, and realizes the accurate conversion of the patient's natural text description into standardized medical information, enhancing doctors' understanding of the patient's situation.
Smart Images

Figure CN113689951B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of digital medical technology, and in particular to an intelligent diagnosis guidance method, system and computer-readable storage medium. Background Art
[0002] With the development of society and the increase in population, there are more and more patients, and the division of departments is becoming more and more detailed. The complexity and diversity of diseases have made it difficult for patients to register. Most patients do not know which department or doctor they should go to for their physical symptoms because they have no medical background. In addition, patients often register the wrong number, which not only increases the patient's consultation time, but also may delay the patient's condition.
[0003] At present, most hospitals have begun to use intelligent robots to guide patients. However, the text of current intelligent robots mostly adopts structured question-and-answer guidance, because the terms used are all standard medical terms (such as epistaxis, hiccups, etc.), which are simply incomprehensible to ordinary patients. If natural text input is used, many colloquial expressions cannot be effectively converted and captured; in addition, it is difficult for current intelligent robots to generate standardized medical text from existing information and import it into the electronic medical record system. Summary of the invention
[0004] In order to solve at least one of the above technical problems, the present invention proposes an intelligent guidance method, system and computer-readable storage medium, which can improve the convenience and accuracy of guidance and registration, and further improve the efficiency of medical treatment.
[0005] The first aspect of the present invention provides an intelligent diagnosis guidance method, the method comprising:
[0006] Receive the patient's medical information, analyze it through the deep learning interactive model, obtain the corresponding vector information, and perform medical terminology normalization on the vector information to generate a standardized medical vector;
[0007] According to the known symptoms in the standardized medical vector, searching in the knowledge graph engine for other symptoms that are highly correlated with the known symptoms, and generating relevant problem descriptions by combining the known symptoms and other symptoms;
[0008] receiving response information of the patient to the problem description, and then updating the standardized medical vector according to the response information;
[0009] Inputting the updated standardized medical vector into a disease prediction model, and using the disease prediction model to predict the disease corresponding to the patient;
[0010] The corresponding department is matched according to the corresponding symptoms of the patient, and a list of doctors sitting in the corresponding department is exported to provide it to the patient for selection and registration.
[0011] In this solution, after the registration is completed, the method further includes:
[0012] Determine the relationship between the various contents in the updated standardized medical vector in combination with the knowledge graph engine, select corresponding associated words for connection, and generate a draft of the chief complaint and the history of present illness;
[0013] Associating the chief complaint and the draft history of present illness with the patient's identification information and storing them in a database;
[0014] A query request from a doctor's workstation front end is received, and based on the query request, corresponding chief complaints and draft medical history are searched in the database for the doctor to diagnose and treat, wherein the query request at least includes the patient's identity information.
[0015] In this solution, the updated standardized medical vector is input into the disease prediction model, and the disease prediction model predicts the disease corresponding to the patient, specifically including:
[0016] Calculate the information quantity of the updated standardized medical vector;
[0017] Comparing the calculated amount of information with a preset threshold;
[0018] If the amount of information calculated exceeds the preset threshold, a disease prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease; if the amount of information calculated does not exceed the preset threshold, a disease group prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease group, wherein the disease prediction model and the disease group prediction model both belong to the disease prediction model.
[0019] In this solution, the patient's diagnosis information is received and analyzed by the deep learning interaction model to obtain the corresponding vector information, including:
[0020] Receiving the patient's voice information about seeking a diagnosis, and preprocessing the voice information about seeking a diagnosis to obtain each frame signal;
[0021] Extract signal feature parameters of each frame signal based on Mel frequency cepstral coefficients;
[0022] The extracted signal feature parameters are input into the deep learning voice interaction model to obtain the probability that each frame signal belongs to each state in the preset state set;
[0023] Based on each frame signal, the state with the highest probability is selected as the sound path of the frame signal, and the corresponding text information is extracted from the state with the highest probability;
[0024] The text information extracted from all frame signals is combined into vector information based on the word order sequence.
[0025] In this solution, the patient's diagnosis information is received and analyzed by the deep learning interaction model to obtain the corresponding vector information, which specifically includes:
[0026] The image collector collects the diagnostic image information of continuous frames in real time, and pre-processes each frame of the diagnostic image information;
[0027] Detect the patient's hand information from each frame of preprocessed diagnostic image information, determine the hand position information in the image, and segment the hand features;
[0028] Based on the segmented hand features from the diagnostic image information of continuous frames, the hand change features are tracked in time sequence;
[0029] Based on the hand features and hand change features obtained above, semantic recognition is performed through a deep learning sign language interaction model, and the corresponding vector information is output.
[0030] In this solution, the hand features segmented based on the diagnostic image information of continuous frames are tracked in time sequence to obtain the hand change features, specifically including:
[0031] It is preset that the diagnostic image information has M frames, and corresponding bone graphs are extracted from the M frames of diagnostic image information respectively;
[0032] Calculate the information volume I between two adjacent frames of skeleton images and the average value of all information volumes I The calculation formula of information volume I is: I = h(S t )+h(S t+1 )-h(S t , S t+1 ), where S represents the skeleton graph, h represents the entropy function, and entropy h(S t ) represents the activity level of the skeleton graph in the tth frame, and the entropy h(S t+1 ) represents the activity level of the skeleton graph in frame t+1, and the entropy h(S t , S t+1 ) represents the similarity between two adjacent skeleton images;
[0033] Mark the code value of the bone graph corresponding to each amount of information I. During the marking process, if The value of the skeleton diagram code corresponding to the mark is 1; otherwise, the value of the skeleton diagram code corresponding to the mark is 0;
[0034] Preset a fixed value d that can be divided by M, and a variable integer a, Calculate the sum of the code values of the corresponding skeleton graphs from the [d×(a-1)+1]th frame to the [d×a]th frame, and determine whether the sum of the code values is 0. If it is not 0, retain all the corresponding skeleton graphs with code values of 1 between the [d×(a-1)+1]th frame to the [d×a]th frame as the key frame skeleton graph; otherwise, retain the skeleton graph of the [d / 2]th frame as the key frame skeleton graph;
[0035] The variable integer a is increased from the initial value 1 to Take values in sequence and obtain all key frame skeleton graphs in the M frame skeleton graph;
[0036] Based on all the key frame skeleton graphs, the hand change features are tracked in time sequence.
[0037] The second aspect of the present invention further provides an intelligent medical guidance system, comprising a memory and a processor, wherein the memory comprises an intelligent medical guidance method program, and when the intelligent medical guidance method program is executed by the processor, the following steps are implemented:
[0038] Receive the patient's medical information, analyze it through the deep learning interactive model, obtain the corresponding vector information, and perform medical terminology normalization on the vector information to generate a standardized medical vector;
[0039] According to the known symptoms in the standardized medical vector, searching in the knowledge graph engine for other symptoms that are highly correlated with the known symptoms, and generating relevant problem descriptions by combining the known symptoms and other symptoms;
[0040] receiving response information of the patient to the problem description, and then updating the standardized medical vector according to the response information;
[0041] Inputting the updated standardized medical vector into a disease prediction model, and using the disease prediction model to predict the disease corresponding to the patient;
[0042] The corresponding department is matched according to the corresponding symptoms of the patient, and a list of doctors sitting in the corresponding department is exported to provide it to the patient for selection and registration.
[0043] In this solution, after the registration is completed, the intelligent diagnosis guidance method program is executed by the processor to further implement the following steps:
[0044] Determine the relationship between the various contents in the updated standardized medical vector in combination with the knowledge graph engine, select corresponding associated words for connection, and generate a draft of the chief complaint and the history of present illness;
[0045] Associating the chief complaint and the draft history of present illness with the patient's identification information and storing them in a database;
[0046] A query request from a doctor's workstation front end is received, and based on the query request, corresponding chief complaints and draft medical history are searched in the database for the doctor to diagnose and treat, wherein the query request at least includes the patient's identity information.
[0047] In this solution, the updated standardized medical vector is input into the disease prediction model, and the disease prediction model predicts the disease corresponding to the patient, specifically including:
[0048] Calculate the information quantity of the updated standardized medical vector;
[0049] Comparing the calculated amount of information with a preset threshold;
[0050] If the amount of information calculated exceeds the preset threshold, a disease prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease; if the amount of information calculated does not exceed the preset threshold, a disease group prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease group, wherein the disease prediction model and the disease group prediction model both belong to the disease prediction model.
[0051] The third aspect of the present invention further proposes a computer-readable storage medium, which includes an intelligent diagnosis guidance method program. When the intelligent diagnosis guidance method program is executed by a processor, the steps of the intelligent diagnosis guidance method as described above are implemented.
[0052] The present invention proposes an intelligent medical guidance method, system and computer-readable storage medium, by which patients can describe their medical conditions as detailed as possible while waiting for medical treatment, and these words are converted into standardized medical vector information and integrated into electronic medical records, thus saving doctors' consultation time and improving medical treatment efficiency.
[0053] The intelligent diagnosis guidance system of the present invention is based on the patient's natural text description, supplemented by question-and-answer prompts, to help patients more accurately and comprehensively describe the development of the disease. Questions and answers are mostly generated based on symptoms in the knowledge graph and the relationship between symptoms, which is more in line with the patient's medical knowledge level.
[0054] The present invention selectively uses disease group models or disease models to predict patient diseases according to the richness of the information already obtained, thereby achieving a more accurate prediction effect.
[0055] The present invention generates the chief complaint and the history of the present illness based on the standardized medical vector, and combines the knowledge graph to complete the sentence fluency and natural modification. At the same time, the chief complaint and the history of the present illness are imported into the front end of the doctor's workstation, which is convenient for outpatient doctors to quickly understand the patient and improve the efficiency of the doctor's consultation.
[0056] Additional aspects and advantages of the present invention will be given in part in the following description, and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 A flow chart of an intelligent diagnosis guidance method of the present invention is shown;
[0058] Figure 2 A block diagram of an intelligent diagnosis guidance system of the present invention is shown;
[0059] Figure 3 A schematic diagram of the software architecture of an intelligent medical guidance system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION
[0060] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.
[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.
[0062] Figure 1 A flow chart of an intelligent diagnosis guidance method of the present invention is shown.
[0063] like Figure 1 As shown, the first aspect of the present invention provides an intelligent diagnosis guidance method, the method comprising:
[0064] S102, receiving the patient's medical information, parsing it through the deep learning interactive model to obtain corresponding vector information, and performing medical terminology normalization operation on the vector information to generate a standardized medical vector;
[0065] S104, searching the knowledge graph engine for other symptoms that are highly correlated with the known symptoms in the standardized medical vector, and generating a related problem description by combining the known symptoms and the other symptoms;
[0066] S106, receiving response information made by the patient to the problem description, and then updating the standardized medical vector according to the response information;
[0067] S108, inputting the updated standardized medical vector into a disease prediction model, and using the disease prediction model to predict the disease corresponding to the patient;
[0068] S110, matching the corresponding department according to the corresponding symptoms of the patient, and exporting the corresponding department's attending doctor list to provide it to the patient for selection and registration.
[0069] It can be understood that the vector information may include a variety of information such as text, type, value, time and extended attributes; search for symptoms that are highly correlated with known symptoms in the knowledge graph engine, obtain their common descriptions, and generate follow-up questions as a guide for patients to continue to refine their disease-related descriptions.
[0070] It should be noted that the above department matching process can be divided into two steps. First, the corresponding disease group (that is, the specialist department is responsible for the disease group) is queried based on the symptoms, and then the corresponding department's list of doctors is queried based on the disease group and provided to the patient for selection.
[0071] According to an embodiment of the present invention, after the registration is completed, the method further includes:
[0072] Determine the relationship between the various contents in the updated standardized medical vector in combination with the knowledge graph engine, select corresponding associated words for connection, and generate a draft of the chief complaint and the history of present illness;
[0073] Associating the chief complaint and the draft history of present illness with the patient's identification information and storing them in a database;
[0074] A query request from a doctor's workstation front end is received, and based on the query request, corresponding chief complaints and draft medical history are searched in the database for the doctor to diagnose and treat, wherein the query request at least includes the patient's identity information.
[0075] Preferably, the above-mentioned identity identification information can be the patient's ID number, medical card number, etc., but is not limited thereto.
[0076] Specifically, the present invention generates a draft of the chief complaint and the history of the present illness based on the text, value, time and other information of the standardized medical vector, and then combines the interrelationships of these contents in the knowledge graph engine and uses appropriate Chinese vocabulary to connect them, making the entire text more fluent and natural. The generated text (i.e., the chief complaint and the draft of the history of the present illness) will be stored in a database associated with the consultation number, which is convenient for the front-end query of the doctor's workstation.
[0077] The front end of the doctor's workstation can query the generated text (i.e., the chief complaint and draft of current medical history) based on the consultation number, and the back-end service group will provide it to the electronic medical record system at the front end of the doctor's workstation, and use it as the preset content of the chief complaint and current medical history columns in the patient's electronic medical record for the doctor to diagnose and treat.
[0078] According to an embodiment of the present invention, the updated standardized medical vector is input into a disease prediction model, and the disease prediction model predicts the disease corresponding to the patient, specifically including:
[0079] Calculate the information quantity of the updated standardized medical vector;
[0080] Comparing the calculated amount of information with a preset threshold;
[0081] If the amount of information calculated exceeds the preset threshold, a disease prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease; if the amount of information calculated does not exceed the preset threshold, a disease group prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease group, wherein the disease prediction model and the disease group prediction model both belong to the disease prediction model.
[0082] Specifically, the diseases predicted by the disease prediction model may be fever, pharyngitis, etc., and the disease groups predicted by the symptom prediction model may be upper respiratory tract disease groups, etc.
[0083] It can be understood that in the process of disease prediction, given the great differences in the cultural level of patients, the richness and accuracy of the input content also vary greatly, so different prediction models (such as diseases or disease groups) are selectively selected according to the richness of their effective standardized medical vectors. When the input standardized medical vectors are small, a coarse-grained prediction model, that is, a disease group prediction model, is used; when the input standardized medical vectors are large, a fine-grained prediction model, that is, a disease group disease model, is used; the present invention can select a suitable prediction model based on the medical vectors provided by different patients, thereby facilitating the prediction of more accurate disease information.
[0084] According to a specific embodiment of the present invention, receiving the patient's diagnosis information and parsing it through a deep learning interaction model specifically includes:
[0085] receiving a patient's selection instruction of an interaction type;
[0086] Based on the selection instruction, a corresponding deep learning interaction model is selected to parse the medical information expressed by the patient, and the deep learning interaction model includes at least a deep learning speech interaction model and a deep learning sign language interaction model.
[0087] It should be noted that patients usually use voice to interact, but some deaf-mute people are unable to interact with the system through voice. In this case, they can choose to interact with the system through sign language. When different types of interaction are used, the corresponding deep learning interaction model will also change accordingly to enhance the patient's interaction experience.
[0088] According to an embodiment of the present invention, receiving the patient's medical information, parsing it through a deep learning interaction model, and obtaining corresponding vector information specifically includes:
[0089] Receiving the patient's voice information about seeking a diagnosis, and preprocessing the voice information about seeking a diagnosis to obtain each frame signal;
[0090] Extract signal feature parameters of each frame signal based on Mel frequency cepstral coefficients;
[0091] The extracted signal feature parameters are input into the deep learning voice interaction model to obtain the probability that each frame signal belongs to each state in the preset state set;
[0092] Based on each frame signal, the state with the highest probability is selected as the sound path of the frame signal, and the corresponding text information is extracted from the state with the highest probability;
[0093] The text information extracted from all frame signals is combined into vector information based on the word order sequence.
[0094] It should be noted that before the speech signal is recognized, in order to eliminate the problems of the patient's vocal organs themselves, a series of processing operations must be performed to reduce the impact on the accuracy of speech recognition, and ultimately obtain a signal that is as uniform and smooth as possible, thereby improving the quality of speech recognition.
[0095] According to a specific embodiment of the present invention, the pre-processing process includes: pre-emphasis, frame division and windowing processing sub-processes.
[0096] The pre-emphasis sub-process includes: using a pre-emphasis function to process the patient-seeking voice information, and the expression of the pre-emphasis function is: H(z)=1-αz -1 , 0.9≤α≤1, where H(z) represents the sound frequency after pre-emphasis, z represents the sound frequency before pre-emphasis, and α represents the pre-emphasis coefficient;
[0097] It is understandable that the voice signal (i.e., the diagnosis voice information) will be affected by the mouth and nose, and the high frequency band above about 800Hz will be attenuated at 6dB / octave, so it is necessary to enhance the high frequency part of the voice signal through the pre-emphasis function.
[0098] The framing sub-process includes: framing the pre-emphasized patient-seeking voice information into multiple frame signals, and storing the overlapping parts between two adjacent frame signals after framing;
[0099] It can be understood that the voice information for seeking medical advice is a time-varying signal. Due to the inertial motion of the patient's vocal organs, its characteristics are relatively stable within a short time range (10ms to 30ms). The signal characteristic parameters of the voice information for seeking medical advice are extracted through a short-term steady-state process. The late windowing weakens the two ends of the frame signal. In order to maintain the continuity of the data information, there must be overlapping parts between frames when dividing the frames. Preferably, the frame length of each frame signal can be 25ms, and the frame shift is 10ms. The frame shift is the time sequence shift length between two adjacent frame signals.
[0100] The windowing sub-process includes: strengthening the sampling signal of the diagnosis voice information and weakening other signals by windowing, and smoothing the edge of the frame signal;
[0101] It should be noted that windowing is to transform or operate on short-time signals. Windowing is used to enhance the sampled signal and weaken other signals, smooth the edges of the frame signal, and achieve a smooth transition. Preferably, the window function used in the windowing sub-process is a Hanning window, but is not limited to this.
[0102] According to an embodiment of the present invention, receiving the patient's medical information, parsing it through a deep learning interaction model, and obtaining corresponding vector information, specifically also includes:
[0103] The image collector collects the diagnostic image information of continuous frames in real time, and pre-processes each frame of the diagnostic image information;
[0104] Detect the patient's hand information from each frame of preprocessed diagnostic image information, determine the hand position information in the image, and segment the hand features;
[0105] Based on the segmented hand features from the diagnostic image information of continuous frames, the hand change features are tracked in time sequence;
[0106] Based on the hand features and hand change features obtained above, semantic recognition is performed through a deep learning sign language interaction model, and the corresponding vector information is output.
[0107] It should be noted that, in the process of image preprocessing, the diagnostic image information is firstly gray-thresholded to obtain a binary image, and then the noise is removed by median filtering and Gaussian filtering, and finally the preprocessed diagnostic image information is obtained.
[0108] It should be noted that in the process of dynamic sign language recognition, in addition to hand detection and segmentation, it is also necessary to track the hand to identify its dynamic features. Tracking is to track the segmented hand feature information and the time-series continuous diagnostic image information to facilitate understanding of the observed changes in the patient's hand. Successful tracking can improve recognition accuracy and interpret the semantics conveyed by the hand position, posture or gesture.
[0109] According to an embodiment of the present invention, based on the hand features segmented from the diagnostic image information of consecutive frames, the hand change features are tracked and acquired in time sequence, specifically including:
[0110] It is preset that the diagnostic image information has M frames, and corresponding bone graphs are extracted from the M frames of diagnostic image information respectively;
[0111] Calculate the information volume I between two adjacent frames of skeleton images and the average value of all information volumes I The calculation formula of information volume I is: I = h(S t )+h(S t+1 )-h(S t , S t+1 ), where S represents the skeleton graph, h represents the entropy function, and entropy h(S t ) represents the activity level of the skeleton graph in the tth frame, and the entropy h(S t+1 ) represents the activity level of the skeleton graph in frame t+1, and the entropy h(S t , S t+1 ) represents the similarity between two adjacent skeleton images;
[0112] Mark the code value of the bone graph corresponding to each amount of information I. During the marking process, if The value of the skeleton diagram code corresponding to the mark is 1; otherwise, the value of the skeleton diagram code corresponding to the mark is 0;
[0113] Preset a fixed value d that can be divided by M, and a variable integer a, Calculate the sum of the code values of the corresponding skeleton graphs from the [d×(a-1)+1]th frame to the [d×a]th frame, and determine whether the sum of the code values is 0. If it is not 0, retain all the corresponding skeleton graphs with code values of 1 between the [d×(a-1)+1]th frame to the [d×a]th frame as the key frame skeleton graph; otherwise, retain the skeleton graph of the [d / 2]th frame as the key frame skeleton graph;
[0114] The variable integer a is increased from the initial value 1 to Take values in sequence and obtain all key frame skeleton graphs in the M frame skeleton graph;
[0115] Based on all the key frame skeleton graphs, the hand change features are tracked in time sequence.
[0116] It should be noted that since the diagnostic image information includes many frames, the action changes between adjacent frames are not much different, and can even be approximately regarded as the same gesture action. However, the deep learning sign language interaction model needs to perform calculations on the input images. When the amount of data is large, the neural network calculation takes a long time. The present invention first converts the diagnostic image information into a skeleton graph, further extracts the key frame skeleton graph, and deletes redundant frames. The so-called key frame refers to a representative image frame in a video image sequence. By extracting the key frame, the calculation amount of the neural network can be reduced and the calculation efficiency is improved.
[0117] At the same time, the present invention uses mutual information to determine the similarity between two adjacent frames of skeleton images, selects multiple groups of data (i.e., d is the number of a group) and calculates the average mutual information through statistical methods. The larger the average mutual information value, the higher the similarity between the two frames. On the contrary, the smaller the average mutual information value, the greater the difference between the two frames.
[0118] According to a specific embodiment of the present invention, based on the known symptoms in the standardized medical vector, searching for other symptoms with a high correlation with the known symptoms in the knowledge graph engine specifically includes:
[0119] Predicting the subject in the standardized medical vector by a subject linking module in the knowledge graph engine;
[0120] According to the subject, find candidate relations corresponding to the subject from the knowledge graph of the knowledge graph engine, and the candidate relations are represented into multiple levels according to their structural information;
[0121] The candidate relationship is input into the relationship extraction module of the knowledge graph engine and matched with the standardized medical vector to predict other symptoms that are highly correlated with the known symptoms.
[0122] According to a specific embodiment of the present invention, before receiving the patient's diagnosis information, the method further includes:
[0123] The distance S between the patient and the intelligent medical guidance device is sensed in real time by the sensor;
[0124] When the distance S is less than or equal to the preset distance, the intelligent medical guidance device is adjusted from the sleep mode to the awakening mode, and in the awakening mode, the intelligent medical guidance device is supported to receive the patient's medical information.
[0125] It can be understood that the sensor can be an infrared sensor, an ultrasonic sensor, etc.
[0126] According to a specific embodiment of the present invention, after the intelligent medical guidance device is adjusted from the sleep mode to the wake-up mode, the method further includes:
[0127] Receiving identity information input by the patient;
[0128] and querying whether there is an electronic medical record of the patient in the hospital information system based on the identity information;
[0129] If so, the information generated during the registration process will be recorded in the patient's electronic medical record to facilitate the doctor's query of the diagnosis and treatment process.
[0130] According to a specific embodiment of the present invention, after receiving the identity information input by the patient, the method further includes:
[0131] The display module of the preset intelligent medical guidance device has multiple display modes;
[0132] Extracting the patient's age and education level based on the identity information;
[0133] A target display mode is matched among multiple display modes according to the age and educational background of the patient, and the display module adopts the target display mode for display.
[0134] It is understandable that most older patients have symptoms of dizziness, so a display mode with enlarged text can be used so that older patients can interact normally. At the same time, some patients have a low level of education, so a display mode of Chinese characters plus pinyin can be used so that patients with low education can interact normally. In addition, the intelligent medical guidance system of the present invention can also support multi-language interaction to expand the scope of application.
[0135] Figure 2 A block diagram of an intelligent diagnosis guidance system of the present invention is shown.
[0136] like Figure 2 As shown, the second aspect of the present invention further proposes an intelligent medical guidance system 2, including a memory 21 and a processor 22, wherein the memory includes an intelligent medical guidance method program, and when the intelligent medical guidance method program is executed by the processor, the following steps are implemented:
[0137] Receive the patient's medical information, analyze it through the deep learning interactive model, obtain the corresponding vector information, and perform medical terminology normalization on the vector information to generate a standardized medical vector;
[0138] According to the known symptoms in the standardized medical vector, searching in the knowledge graph engine for other symptoms that are highly correlated with the known symptoms, and generating relevant problem descriptions by combining the known symptoms and other symptoms;
[0139] receiving response information of the patient to the problem description, and then updating the standardized medical vector according to the response information;
[0140] Inputting the updated standardized medical vector into a disease prediction model, and using the disease prediction model to predict the disease corresponding to the patient;
[0141] The corresponding department is matched according to the corresponding symptoms of the patient, and a list of doctors sitting in the corresponding department is exported to provide it to the patient for selection and registration.
[0142] According to an embodiment of the present invention, after the registration is completed, the intelligent diagnosis guidance method program further implements the following steps when executed by the processor:
[0143] Determine the relationship between the various contents in the updated standardized medical vector in combination with the knowledge graph engine, select corresponding associated words for connection, and generate a draft of the chief complaint and the history of present illness;
[0144] Associating the chief complaint and the draft history of present illness with the patient's identification information and storing them in a database;
[0145] A query request from a doctor's workstation front end is received, and based on the query request, corresponding chief complaints and draft medical history are searched in the database for the doctor to diagnose and treat, wherein the query request at least includes the patient's identity information.
[0146] According to an embodiment of the present invention, the updated standardized medical vector is input into a disease prediction model, and the disease prediction model predicts the disease corresponding to the patient, specifically including:
[0147] Calculate the information quantity of the updated standardized medical vector;
[0148] Comparing the calculated amount of information with a preset threshold;
[0149] If the amount of information calculated exceeds the preset threshold, a disease prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease; if the amount of information calculated does not exceed the preset threshold, a disease group prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease group, wherein the disease prediction model and the disease group prediction model both belong to the disease prediction model.
[0150] like Figure 3 As shown, according to a specific embodiment of the present invention, the software architecture of the intelligent medical guidance system also includes the following modules:
[0151] User interaction module: provides patients with the ability to input disease-related information and register;
[0152] Business API module: As the main backend service provider, it implements business-related logic, including the following functions: calling various model services, editing and maintaining knowledge graphs, and connecting to the hospital HIS (Hospital Information System), i.e., the hospital information system.
[0153] Symptom prediction model: includes disease group prediction model and disease prediction model.
[0154] Knowledge graph engine: contains various business-related graphs, such as disease-symptom graph, symptom-symptom graph, symptom-description graph, etc.
[0155] According to a specific embodiment of the present invention, when training the disease prediction model, 80,000 outpatient medical records of a certain hospital are used to complete the structured operation and capture the medical vectors.
[0156] Only the chief complaint medical vector + the diagnosed disease group was used to train the disease group prediction model;
[0157] Use the medical vector of current medical history + diagnosis to train disease prediction models.
[0158] When constructing the knowledge graph, 80,000 outpatient medical records of a certain hospital were used to complete structured operations and capture their medical vectors.
[0159] Construct a disease-symptom map based on the diagnosis vector and the symptom vector in the medical record;
[0160] Based on the coexisting symptom vectors in the same medical record, a symptom-symptom map is constructed;
[0161] Based on the symptom vector and its description information, a symptom-description map is constructed.
[0162] According to a specific embodiment of the present invention, the patient inputs: fever for one day, cough for three days
[0163] Medical vector capture: (preset today 2021-07-28)
[0164] {text:fever,type:symptom,value:1,time:2020-07-28}
[0165] {text:cough,type:symptom,value:1,time:2020-07-28}
[0166] Recommended associated symptoms and descriptions:
[0167] Description: Coughing frequency, main coughing period, coughing severity
[0168] Associated symptoms: runny nose (if patient ticks), abdominal discomfort
[0169] Symptom prediction: In view of the small amount of medical vector information, the disease group prediction model is selected, and the predicted disease group is: upper respiratory tract disease group;
[0170] Department matching: Matching departments according to the predicted upper respiratory tract disease group: pediatrics, respiratory specialists;
[0171] Generate a draft of the chief complaint and history of present illness: cough for three days, fever for one day with runny nose.
[0172] The third aspect of the present invention further proposes a computer-readable storage medium, which includes an intelligent diagnosis guidance method program. When the intelligent diagnosis guidance method program is executed by a processor, the steps of the intelligent diagnosis guidance method as described above are implemented.
[0173] The present invention proposes an intelligent medical guidance method, system and computer-readable storage medium, by which patients can describe their medical conditions as detailed as possible while waiting for medical treatment, and these words are converted into standardized medical vector information and integrated into electronic medical records, thus saving doctors' consultation time and improving medical treatment efficiency.
[0174] The intelligent diagnosis guidance system of the present invention is based on the patient's natural text description, supplemented by question-and-answer prompts, to help patients more accurately and comprehensively describe the development of the disease. Questions and answers are mostly generated based on symptoms in the knowledge graph and the relationship between symptoms, which is more in line with the patient's medical knowledge level.
[0175] The present invention selectively uses disease group models or disease models to predict patient diseases according to the richness of the information already obtained, thereby achieving a more accurate prediction effect.
[0176] The present invention generates the chief complaint and the history of the present illness based on the standardized medical vector, and combines the knowledge graph to complete the sentence fluency and natural modification. At the same time, the chief complaint and the history of the present illness are imported into the front end of the doctor's workstation, which is convenient for outpatient doctors to quickly understand the patient and improve the efficiency of the doctor's consultation.
[0177] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.
[0178] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0179] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.
[0180] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.
[0181] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.
[0182] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. An intelligent diagnosis guidance method, characterized in that: The method comprises: Receive the patient's medical information, analyze it through the deep learning interactive model, obtain the corresponding vector information, and perform medical terminology normalization on the vector information to generate a standardized medical vector; According to the known symptoms in the standardized medical vector, searching in the knowledge graph engine for other symptoms that are highly correlated with the known symptoms, and generating relevant problem descriptions by combining the known symptoms and other symptoms; receiving response information of the patient to the problem description, and then updating the standardized medical vector according to the response information; Inputting the updated standardized medical vector into a disease prediction model, and using the disease prediction model to predict the disease corresponding to the patient; Match the corresponding department according to the corresponding symptoms of the patient, and derive a list of doctors in the corresponding department to provide to the patient for selection and registration; Among them, the patient's diagnosis information is received and analyzed by the deep learning interaction model to obtain the corresponding vector information, including: Receiving the patient's voice information about seeking a diagnosis, and preprocessing the voice information about seeking a diagnosis to obtain each frame signal; Extract signal feature parameters of each frame signal based on Mel frequency cepstral coefficients; The extracted signal feature parameters are input into the deep learning voice interaction model to obtain the probability that each frame signal belongs to each state in the preset state set; Based on each frame signal, the state with the highest probability is selected as the sound path of the frame signal, and the corresponding text information is extracted from the state with the highest probability; The text information extracted from all frame signals is combined into vector information based on the word order sequence; The pre-processing process includes: pre-emphasis, frame division and windowing processing sub-processes; The pre-emphasis sub-process includes: using a pre-emphasis function to process the patient-seeking voice information, and the expression of the pre-emphasis function is: H(z)=1-αz -1 , 0.9≤α≤1, where H(z) represents the sound frequency after pre-emphasis, z represents the sound frequency before pre-emphasis, and α represents the pre-emphasis coefficient; The framing sub-process includes: framing the pre-emphasized patient-seeking voice information into multiple frame signals, and storing the overlapping parts between two adjacent frame signals after framing; The windowing sub-process includes: strengthening the sampling signal of the diagnosis voice information and weakening other signals by windowing, and smoothing the edge of the frame signal; Among them, the patient's diagnosis information is received and analyzed by the deep learning interaction model to obtain the corresponding vector information, which specifically includes: The image collector collects the diagnostic image information of continuous frames in real time, and pre-processes each frame of the diagnostic image information; Detect the patient's hand information from each frame of preprocessed diagnostic image information, determine the hand position information in the image, and segment the hand features; Based on the segmented hand features from the diagnostic image information of continuous frames, the hand change features are tracked in time sequence; Based on the hand features and hand change features obtained above, semantic recognition is performed through a deep learning sign language interaction model, and the corresponding vector information is output; Among them, based on the segmentation of the hand features from the diagnostic image information of the continuous frames, the hand change features are tracked in time sequence, specifically including: It is preset that the diagnostic image information has M frames, and corresponding bone graphs are extracted from the M frames of diagnostic image information respectively; Calculate the information volume I between two adjacent frames of skeleton images and the average value of all information volumes I The calculation formula of information volume I is: I = h(S t )+h(S t+1 )-h(S t ,S t+1 ), where S represents the skeleton graph, h represents the entropy function, and entropy h(S t ) represents the activity level of the skeleton graph in the tth frame, and the entropy h(S t+1 ) represents the activity level of the skeleton graph in frame t+1, and the entropy h(S t ,S t+1 ) represents the similarity between two adjacent skeleton images; Mark the code value of the bone graph corresponding to each amount of information I. During the marking process, if The value of the skeleton diagram code corresponding to the mark is 1; otherwise, the value of the skeleton diagram code corresponding to the mark is 0; Preset a fixed value d that can be divided by M, and a variable integer a, Calculate the sum of the code values of the corresponding skeleton graphs from the [d×(a-1)+1]th frame to the [d×a]th frame, and determine whether the sum of the code values is 0. If it is not 0, retain all the corresponding skeleton graphs with code values of 1 between the [d×(a-1)+1]th frame to the [d×a]th frame as the key frame skeleton graph; otherwise, retain the skeleton graph of the [d / 2]th frame as the key frame skeleton graph; The variable integer a is increased from the initial value 1 to Take values in sequence and obtain all key frame skeleton graphs in the M frame skeleton graph; Based on all the key frame skeleton graphs, track and obtain the hand change characteristics in time sequence; According to the known symptoms in the standardized medical vector, other symptoms with a high correlation with the known symptoms are searched in the knowledge graph engine, specifically including: Predicting the subject in the standardized medical vector by a subject linking module in the knowledge graph engine; According to the subject, find candidate relations corresponding to the subject from the knowledge graph of the knowledge graph engine, and the candidate relations are represented into multiple levels according to their structural information; Inputting the candidate relationship into the relationship extraction module of the knowledge graph engine and matching it with the standardized medical vector to predict other symptoms with a high correlation with the known symptoms; Inputting the updated standardized medical vector into a disease prediction model, and predicting the disease corresponding to the patient by the disease prediction model, specifically includes: Calculate the information quantity of the updated standardized medical vector; Comparing the calculated amount of information with a preset threshold; If the amount of information calculated exceeds the preset threshold, a disease prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease; if the amount of information calculated does not exceed the preset threshold, a disease group prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease group, wherein the disease prediction model and the disease group prediction model both belong to the disease prediction model.
2. The intelligent diagnosis guidance method according to claim 1, characterized in that: After the registration is completed, the method further comprises: Determine the relationship between the various contents in the updated standardized medical vector in combination with the knowledge graph engine, select corresponding associated words for connection, and generate a draft of the chief complaint and the history of present illness; Associating the chief complaint and the draft history of present illness with the patient's identification information and storing them in a database; A query request from a doctor's workstation front end is received, and based on the query request, corresponding chief complaints and draft medical history are searched in the database for the doctor to diagnose and treat, wherein the query request at least includes the patient's identity information.
3. An intelligent medical guidance system, characterized in that: The system comprises a memory and a processor, wherein the memory comprises an intelligent diagnosis guidance method program, and when the intelligent diagnosis guidance method program is executed by the processor, the following steps are implemented: Receive the patient's medical information, analyze it through the deep learning interactive model, obtain the corresponding vector information, and perform medical terminology normalization on the vector information to generate a standardized medical vector; According to the known symptoms in the standardized medical vector, searching in the knowledge graph engine for other symptoms that are highly correlated with the known symptoms, and generating relevant problem descriptions by combining the known symptoms and other symptoms; receiving response information of the patient to the problem description, and then updating the standardized medical vector according to the response information; Inputting the updated standardized medical vector into a disease prediction model, and using the disease prediction model to predict the disease corresponding to the patient; Match the corresponding department according to the corresponding symptoms of the patient, and derive a list of doctors in the corresponding department to provide to the patient for selection and registration; Among them, the patient's diagnosis information is received and analyzed by the deep learning interaction model to obtain the corresponding vector information, including: Receiving the patient's voice information about seeking a diagnosis, and preprocessing the voice information about seeking a diagnosis to obtain each frame signal; Extract signal feature parameters of each frame signal based on Mel frequency cepstral coefficients; The extracted signal feature parameters are input into the deep learning voice interaction model to obtain the probability that each frame signal belongs to each state in the preset state set; Based on each frame signal, the state with the highest probability is selected as the sound path of the frame signal, and the corresponding text information is extracted from the state with the highest probability; The text information extracted from all frame signals is combined into vector information based on the word order sequence; The pre-processing process includes: pre-emphasis, frame division and windowing processing sub-processes; The pre-emphasis sub-process includes: using a pre-emphasis function to process the patient-seeking voice information, and the expression of the pre-emphasis function is: H(z)=1-αz -1 , 0.9≤α≤1, where H(z) represents the sound frequency after pre-emphasis, z represents the sound frequency before pre-emphasis, and α represents the pre-emphasis coefficient; The framing sub-process includes: framing the pre-emphasized patient-seeking voice information into multiple frame signals, and storing the overlapping parts between two adjacent frame signals after framing; The windowing sub-process includes: strengthening the sampling signal of the diagnosis voice information and weakening other signals by windowing, and smoothing the edge of the frame signal; Among them, the patient's diagnosis information is received and analyzed by the deep learning interaction model to obtain the corresponding vector information, which specifically includes: The image collector collects the diagnostic image information of continuous frames in real time, and pre-processes each frame of the diagnostic image information; Detect the patient's hand information from each frame of preprocessed diagnostic image information, determine the hand position information in the image, and segment the hand features; Based on the segmented hand features from the diagnostic image information of continuous frames, the hand change features are tracked in time sequence; Based on the hand features and hand change features obtained above, semantic recognition is performed through a deep learning sign language interaction model, and the corresponding vector information is output; Among them, based on the segmentation of the hand features from the diagnostic image information of the continuous frames, the hand change features are tracked in time sequence, specifically including: It is preset that the diagnostic image information has M frames, and corresponding bone graphs are extracted from the M frames of diagnostic image information respectively; Calculate the information volume I between two adjacent frames of skeleton images and the average value of all information volumes I The calculation formula of information volume I is: I = h(S t )+h(S t+1 )-h(S t ,S t+1 ), where S represents the skeleton graph, h represents the entropy function, and entropy h(S t ) represents the activity level of the skeleton graph in the tth frame, and the entropy h(S t+1 ) represents the activity level of the skeleton graph in frame t+1, and the entropy h(S t ,S t+1 ) represents the similarity between two adjacent skeleton images; Mark the code value of the bone graph corresponding to each amount of information I. During the marking process, if The value of the skeleton diagram code corresponding to the mark is 1; otherwise, the value of the skeleton diagram code corresponding to the mark is 0; Preset a fixed value d that can be divided by M, and a variable integer a, Calculate the sum of the code values of the corresponding skeleton graphs from the [d×(a-1)+1]th frame to the [d×a]th frame, and determine whether the sum of the code values is 0. If it is not 0, retain all the corresponding skeleton graphs with code values of 1 between the [d×(a-1)+1]th frame to the [d×a]th frame as the key frame skeleton graph; otherwise, retain the skeleton graph of the [d / 2]th frame as the key frame skeleton graph; The variable integer a is increased from the initial value 1 to Take values in sequence and obtain all key frame skeleton graphs in the M frame skeleton graph; Based on all the key frame skeleton graphs, track and obtain the hand change characteristics in time sequence; According to the known symptoms in the standardized medical vector, other symptoms with a high correlation with the known symptoms are searched in the knowledge graph engine, specifically including: Predicting the subject in the standardized medical vector by a subject linking module in the knowledge graph engine; According to the subject, find candidate relations corresponding to the subject from the knowledge graph of the knowledge graph engine, and the candidate relations are represented into multiple levels according to their structural information; Inputting the candidate relationship into the relationship extraction module of the knowledge graph engine and matching it with the standardized medical vector to predict other symptoms with a high correlation with the known symptoms; Inputting the updated standardized medical vector into a disease prediction model, and predicting the disease corresponding to the patient by the disease prediction model, specifically includes: Calculate the information quantity of the updated standardized medical vector; Comparing the calculated amount of information with a preset threshold; If the amount of information calculated exceeds the preset threshold, a disease prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease; if the amount of information calculated does not exceed the preset threshold, a disease group prediction model is selected to predict the updated standardized medical vector and obtain the corresponding disease group, wherein the disease prediction model and the disease group prediction model both belong to the disease prediction model.
4. The intelligent medical guidance system according to claim 3, characterized in that: After the registration is completed, the intelligent diagnosis guidance method program further implements the following steps when executed by the processor: Determine the relationship between the various contents in the updated standardized medical vector in combination with the knowledge graph engine, select corresponding associated words for connection, and generate a draft of the chief complaint and the history of present illness; Associating the chief complaint and the draft history of present illness with the patient's identification information and storing them in a database; A query request from a doctor's workstation front end is received, and based on the query request, corresponding chief complaints and draft medical history are searched in the database for the doctor to diagnose and treat, wherein the query request at least includes the patient's identity information.
5. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes an intelligent medical guidance method program. When the intelligent medical guidance method program is executed by a processor, the steps of an intelligent medical guidance method as described in any one of claims 1 to 2 are implemented.
