Personal health information prediction system based on AI voice recognition

Abstract

The present invention relates to a personal health information prediction system based on AI voice recognition. According to the present invention, a personal health information prediction system based on AI voice recognition comprises: a symptom prediction unit that applies text conversion data regarding a user's voice signal related to the user's main symptoms to a first prediction model to predict probability values ​​for each minor symptom, and applies the probability values ​​for each minor symptom to a second prediction model to predict probability values ​​for each major symptom; a first survey unit that presents a candidate list of sub-minor symptoms for the major symptom derived with the highest probability value and receives a selection of a plurality of corresponding minor symptoms from the user; a disease prediction unit that inputs the selected plurality of minor symptoms into a disease prediction model to predict a plurality of expected diseases associated with the input data and a major symptom with the highest correlation to the plurality of expected diseases; a second survey unit that presents a candidate list of sub-minor symptoms for the major symptom with the highest correlation to the plurality of expected diseases and receives a selection of a plurality of corresponding minor symptoms; and a control unit that inputs all minor symptoms selected by the first and second survey units into the disease prediction model to finally derive n expected diseases of the user. According to the present invention, a user's voice signal related to the user's main symptoms can be analyzed based on artificial intelligence to ultimately derive and provide the user's predicted disease, and a medical department can be recommended to the user based on the predicted disease.

Description

Technology Field

[0001] The present invention relates to an AI voice recognition-based personal health information prediction system, and more specifically, to an AI voice recognition-based personal health information prediction system capable of receiving a user voice signal related to the user's major symptoms, analyzing it through artificial intelligence, and deriving and providing the user's predicted disease. Background Technology

[0002] Recently, active research and development is underway on systems that use artificial intelligence to analyze data regarding patients' biological information or various symptoms to predict or match their expected diseases.

[0003] Among these, a system has been introduced that diagnoses health conditions and predicts diseases by analyzing vital signs or exercise status collected from patients through various sensors. However, while the reliability of the prediction results is relatively high due to its reliance on vital signs, this system has limitations; it requires separate sensor devices to collect vital signs from patients, and it is difficult to predict diseases in areas or conditions that cannot be diagnosed through vital sign measurement alone.

[0004] In addition, based on these factors, there is a need for a non-face-to-face medical diagnostic system that can collect information on a patient's current symptoms on a mobile device without additional sensors before the patient visits a medical institution such as a hospital, derive a predicted disease based on this information, and recommend an appropriate medical department.

[0005] The technology forming the background of the present invention is disclosed in Korean Published Patent No. 10-2022-0068858 (published May 26, 2022). The problem to be solved

[0006] The present invention aims to provide an AI voice recognition-based personal health information prediction system capable of deriving a user's expected disease by analyzing voice signals related to the user's major symptoms. means of solving the problem

[0007] The present invention provides a personal health information prediction system based on AI voice recognition, comprising: a symptom prediction unit that applies text conversion data regarding a user's voice signal related to the user's major symptoms to a first prediction model to predict probability values ​​for each minor symptom and applies the probability values ​​for each minor symptom to a second prediction model to predict probability values ​​for each major symptom; a first survey unit that presents a candidate list of sub-minor symptoms for the major symptom derived with the highest probability value and receives a selection of a plurality of corresponding minor symptoms from the user; a disease prediction unit that inputs the selected plurality of minor symptoms into a disease prediction model to predict a plurality of expected diseases associated with the input data and a major symptom with the highest correlation to the plurality of expected diseases; a second survey unit that presents a candidate list of sub-minor symptoms for the major symptom with the highest correlation to the plurality of expected diseases and receives a selection of a plurality of corresponding minor symptoms; and a control unit that inputs all minor symptoms selected by the first and second survey units into the disease prediction model to finally derive n expected diseases of the user.

[0008] In addition, the first and second questionnaire sections may extract a list of sub-symptoms corresponding to the corresponding major symptom from a database that has previously stored a list of sub-symptoms for each of the multiple major symptoms and present it as a candidate list.

[0009] In addition, the personal health information prediction system may further include a voice signal processing unit that receives a user voice signal related to the user's major symptoms through a user terminal and converts it into text.

[0010] In addition, the voice signal processing unit can convert a sentence corresponding to the input voice signal into text, separate morphemes, encode them as integers, and then perform padding processing.

[0011] In addition, the control unit can provide the n predicted diseases derived in the final result to the user terminal as a disease prediction result.

[0012] In addition, the control unit may provide the top n candidate diseases with the highest probability values ​​among the multiple candidate disease probability values ​​derived by deep learning analysis of the input symptoms through the disease prediction model, sorted in descending order.

[0013] In addition, the disease prediction model may include a first model that outputs probability values ​​for each candidate disease by deep learning analysis of a plurality of input minor symptoms, and a second model that outputs probability values ​​for a plurality of major symptoms by deep learning analysis of a plurality of candidate disease probability values ​​received from the first model.

[0014] In addition, the sub-symptoms of the above major symptom may include information regarding accompanying symptoms related to the above major symptom and the site of symptom manifestation.

[0015] In addition, the personal health information prediction system may further include a recommendation unit that recommends a medical department to the user based on the finally derived predicted disease.

[0016] In addition, the personal health information prediction system described above can be implemented to be linked with a mobile app executable on a user terminal. Effects of the invention

[0017] According to the present invention, a user's voice signal related to the user's major symptoms can be analyzed based on artificial intelligence to ultimately derive and provide the user's predicted disease. In addition, based on the predicted disease, a medical department for a hospital visit can be recommended to the user. Brief explanation of the drawing

[0018] FIG. 1 is a diagram showing the configuration of an AI voice recognition-based personal health information prediction system according to an embodiment of the present invention. Figure 2 is a diagram illustrating the major symptoms and minor symptoms of a specific disease as examples. FIG. 3 is a diagram specifically illustrating a symptom prediction model according to an embodiment of the present invention. FIG. 4 is a diagram illustrating the operation of a first prediction model in an embodiment of the present invention. FIG. 5 is a diagram illustrating a personal health information prediction process according to an embodiment of the present invention. Specific details for implementing the invention

[0019] Then, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification have been given similar reference numerals.

[0020] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected," but also cases where they are "electrically connected" with other components interposed between them. Furthermore, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0021] FIG. 1 is a diagram showing the configuration of an AI voice recognition-based personal health information prediction system according to an embodiment of the present invention.

[0022] As shown in FIG. 1, an AI voice recognition-based personal health information prediction system (100) according to an embodiment of the present invention includes a symptom prediction unit (120), a first survey unit (130), a disease prediction unit (140), a second survey unit (150), and a control unit (160), and may further include a voice signal processing unit (110), a recommendation unit (170), and a database (180). The operation of each unit (110~150, 170, 180) and the data flow between each unit can be controlled by the control unit (160).

[0023] First, the voice signal processing unit (110) can receive a user voice signal related to the user's main symptoms through the user terminal (10) and convert it into text. Here, the user terminal (10) may include a smartphone, laptop, pad, tablet OC, desktop, etc., which has a built-in process capable of inputting and processing microphone signals.

[0024] Here, the voice signal processing unit (110) can convert a sentence corresponding to the input voice signal into text, perform morpheme separation, then integer encode the separated morphemes and provide padding. As a simple example, when a voice signal such as "My stomach hurts so much today, and my head hurts too" is received, it is converted into text, each morpheme is separated, integer encoding is performed, and then padding is performed to adjust the length of all data to a set length.

[0025] Since various speech signal processing techniques for converting speech signals into text have been previously disclosed, various known data preprocessing techniques may be utilized in the embodiments of the present invention. For morphological separation, a Python package for Korean information processing called KoNLPy may be utilized.

[0026] The voice signal processing unit (110) can transmit text conversion data for a user voice signal related to the user's main symptoms to the symptom prediction unit (120). The symptom prediction unit (120) can predict minor symptoms expected of the user from the text conversion data and predict major symptoms based on this.

[0027] Specifically, the symptom prediction unit (120) can predict probability values ​​for each minor symptom by applying text conversion data for user voice signals related to the user's major symptoms to a previously trained first prediction model, and predict probability values ​​for each major symptom by applying the probability values ​​for each minor symptom to a previously trained second prediction model.

[0028] FIG. 2 is a diagram illustrating the major symptoms and minor symptoms of a specific disease. FIG. 2 illustrates the major symptoms associated with upper respiratory tract infections and the types of minor symptoms subordinate thereto.

[0029] As shown in FIG. 2, sub-symptoms of a major symptom may include information regarding accompanying symptoms related to the major symptom and the site of symptom manifestation. For example, if the major symptom is a headache, the sub-symptoms of that symptom may include skin heat sensation, high fever, fever, chills, left side pain, right side pain, tingling sensation, etc.

[0030] These symptom prediction units (120) may be implemented by including two symptom prediction models, wherein the first prediction model is a minor symptom prediction module that receives text conversion data for a user's voice signal and performs deep learning analysis to predict minor symptoms related to the user's disease, and the second prediction model is a major symptom prediction module that receives the minor symptom prediction results and performs deep learning analysis to predict major symptoms related to the user's disease.

[0031] Here, the first prediction model is trained in advance and can be pre-trained using text conversion data of voice signals related to the patient's major symptoms input from multiple patients with various diseases and symptom cases, and data on at least one sub-symptom of the patient. In addition, the second prediction model can be pre-trained using data on at least one sub-symptom of the patient and major symptom information of the patient.

[0032] FIG. 3 is a diagram specifically illustrating a symptom prediction model according to an embodiment of the present invention. As shown in FIG. 3, the first prediction model (121) included in the symptom prediction unit (120) can classify minor symptoms from text conversion data obtained by natural language processing of a user voice signal. Specifically, it can output expected probability values ​​for each of the multiple minor symptoms through each output node by deep learning analysis of the input text conversion data. The second prediction model (122) can receive the prediction results of the first prediction model (121), perform deep learning analysis, and output expected probability values ​​for each of the multiple major symptoms through each output node.

[0033] The first prediction model (121) can receive each word constituting text conversion data through multiple input nodes, perform machine learning, and output and transmit multiple prediction values ​​for each symptom through multiple output nodes. An embodiment of the present invention can embed data for voice signals, encode them, and then pre-train.

[0034] FIG. 4 is a diagram illustrating the operation of a first prediction model according to an embodiment of the present invention. As shown in FIG. 4, the first prediction model may utilize a BERT model. This first prediction model can be trained by receiving data that has undergone a tokenization process through the KoNLPy tool. Tokenized data for a sentence may be represented as Tok1, Tok2, …, TokN, etc.

[0035] Here, the left figure of Fig. 4 illustrates an example of classifying labels by receiving tokenized data for two sentences and performing learning to match associations, and the right figure illustrates an example of classifying labels (sub-symptoms) by receiving tokenized text data for a single sentence, and the output layer can provide probability values ​​for each sub-symptom.

[0036] The BERT model is an AI-based natural language processing model that learns by reading and learning a large amount of data on its own, and classifies or predicts results by identifying the context and relationships of words. Through this, the first prediction model (121) can learn the relationship between various contextual information of text converted from a speech signal and various sub-symptoms of the user. In addition, after learning is completed, the user's sub-symptoms can be classified using text conversion data for the input speech signal, and more specifically, probability values ​​can be provided for each sub-symptom. The higher the probability value of a sub-symptom, the more likely it is to be a sub-symptom with a high degree of association with the corresponding sentence.

[0037] In the case of the embodiment of the present invention, a BERT model was utilized that was pre-trained with a large number of corpora received from St. Mary's Hospital. An example of the description of the corpora trained with BERT is shown in Table 1.

[0038]

[0039] Of course, this is merely one example, and the present invention is not necessarily limited to being learned by the data described above. In addition, in an embodiment of the present invention, the second prediction model may be implemented with a structure combining Bi-LSTM (bidirectional LSTM) and CRF. This second prediction model can output predicted values ​​for each symptom through a plurality of output nodes by machine learning the data received from the first prediction model.

[0040] The first survey unit (130) identifies the major symptom derived from the highest probability value among the probability values ​​for each major symptom output from the first prediction model (121), presents the sub-symptoms of the identified major symptom in the form of options to the user terminal (10) as a candidate list, and receives multiple corresponding sub-symptoms from the user.

[0041] At this time, the first questionnaire (130) may extract a list of sub-symptoms corresponding to the corresponding major symptom from a database (180) that has previously stored a list of sub-symptoms for each major symptom and present it as a candidate list.

[0042] The user selects the sub-symptoms that currently apply to them from a list of candidate sub-symptoms provided as a survey on the user terminal (10), and depending on the user's condition, all or part of the list of candidate sub-symptoms currently provided may be selected. Additionally, the first survey unit (130) can transmit the sub-symptom data selected through the user terminal (10) to the disease prediction unit (140).

[0043] The disease prediction unit (140) inputs a number of selected minor symptoms into a pre-trained disease prediction model to predict a number of expected diseases associated with the input data and a major symptom that is most closely associated with the number of expected diseases.

[0044] Here, the disease prediction unit (140) may include a first model that outputs probability values ​​for candidate diseases by deep learning analysis of multiple input minor symptoms, and a second model that outputs probability values ​​for multiple major symptoms by deep learning analysis of multiple candidate disease probability values ​​received from the first model. To this end, the first model and the second model may classify or predict expected diseases by machine learning analysis of pre-trained and input data, or classify or predict expected major symptoms by machine learning analysis of classified expected diseases.

[0045] The disease prediction unit (140) can identify the major symptom most closely associated with a plurality of predicted diseases from the results derived from the second model and provide it to the second questionnaire unit (150). At this time, the major symptom most closely associated may be the same as or different from the major symptom initially derived from the symptom prediction unit (120) mentioned earlier.

[0046] The second survey section (150) can identify the major symptoms most closely associated with multiple expected diseases, present a list of sub-symptoms corresponding to the identified major symptoms as candidates, and receive the selection of the corresponding sub-symptoms. At this time, the second survey section (150), like the first survey section (130), can extract a list of sub-symptoms corresponding to the major symptoms from the database (180) and provide them as a candidate list. In this way, the first and second survey sections (150) can receive feedback from the user regarding sub-symptoms related to the user's current condition by linking with the database. Additionally, the accuracy of the prediction model can be improved through two rounds of feedback. According to the present invention, user feedback information is reflected in artificial intelligence learning, thereby enabling more accurate disease prediction.

[0047] Next, the control unit (160) can input all the minor symptoms selected by the first and second survey units (150) into a disease prediction model to finally derive n predicted diseases of the user.

[0048] The control unit (160) can input the data of all minor symptoms selected by the first and second survey units (150) back into the previously described disease prediction model to finally derive and provide n predicted diseases from the first model.

[0049] The control unit (160) can provide the n predicted diseases derived in the final stage as disease prediction results through the user terminal (10). At this time, the control unit (160) can provide the top n candidate diseases with the highest probability values ​​among the multiple candidate disease probability values ​​derived by deep learning analysis of the input symptoms through a disease prediction model, sorted in descending order. For example, if n is 3, the top 3 candidate diseases with the highest probability can be provided in descending order.

[0050] In addition, the recommendation unit (170) can recommend a medical department to the user based on the predicted disease finally derived in this way. At this time, it can recommend a corresponding medical department for each of the n candidate diseases, and can additionally provide a list of nearby hospitals where the corresponding medical department exists based on the location of the user terminal.

[0051] The personal health information prediction system (100) according to the embodiment of the present invention may be implemented to be linked with a mobile app executable on a user terminal. Additionally, the personal health information prediction system (100) may correspond to a server providing related services and may provide a voice-based health information prediction service platform by being connected to a network with the user terminal (10).

[0052] The present invention can predict what disease is suspected of a user based on voice information and survey information regarding the user's current major symptoms, and furthermore, can recommend medical departments and hospital information related to the predicted disease by linking with at least one hospital.

[0053] FIG. 5 is a diagram illustrating a personal health information prediction process according to an embodiment of the present invention.

[0054] First, the voice signal processing unit (110) of the health information prediction system (100) can receive a voice signal regarding the user's main symptoms from the user terminal (10) and convert it into text (S1). The conversion result is transmitted to the symptom prediction unit (120).

[0055] The symptom prediction unit (120) can predict minor symptoms by deep learning analysis of text-converted voice signal data and predict major symptoms therefrom (S2). The symptom prediction unit (120) can predict minor symptoms associated with the user's voice signal by applying the text-converted voice signal to a pre-trained first prediction model and predict major symptoms associated therewith by applying the prediction result to a second prediction model. At this time, probability values ​​are provided for each major symptom label, and the major symptom information with the highest probability is utilized in the first survey unit (130).

[0056] The first survey unit (130) can conduct a user survey and response through the major symptom prediction results obtained from the symptom prediction unit (120) (S3). At this time, the first survey unit (130) can retrieve information on sub-symptoms regarding the major symptom classified with the highest probability in the symptom prediction unit (120) from the database (180), present it as a candidate item for the user survey, and receive the corresponding sub-symptoms from the user. Information on the sub-symptoms selected by the user can be transmitted to the disease prediction unit (140).

[0057] The disease prediction unit (140) inputs the minor symptoms responded to in the first survey unit (130) into a pre-trained disease prediction model to predict multiple expected diseases associated with the minor symptoms and major symptoms highly associated with these multiple expected diseases (S4). The information on the major symptoms most highly associated can be transmitted to the second survey unit (150).

[0058] Next, the second survey section (150) retrieves information on sub-symptoms for the major symptom derived with the highest probability from the disease prediction section (140) from the database (180) and presents it as a candidate list for the user survey, and the corresponding sub-symptoms can be selected through options (S5).

[0059] Next, the control unit (160) inputs all the minor symptoms received through the user survey in steps S3 and S5 into a disease prediction model to finally derive n predicted diseases of the user (S6). At this time, the minor symptoms responded to the survey in each step can be combined and input into the model, and duplicate minor symptoms can be aggregated.

[0060] Additionally, the control unit (160) can output and provide the final derived predicted disease results to the user terminal (10) (S7). At this time, n predicted diseases can be sorted in descending order of probability value and provided.

[0061] In addition, the control unit (160) can recommend a medical department related to the expected disease to the user (S8). At this time, relevant hospital information can also be provided.

[0062] According to the present invention as described above, a user's voice signal related to the user's major symptoms can be analyzed based on artificial intelligence to ultimately derive and provide the user's predicted disease. In addition, based on the predicted disease, a medical department for a hospital visit can be recommended to the user.

[0063] The present invention has been described with reference to embodiments illustrated in the drawings, but this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims. Explanation of the symbols

[0064] 100: Personal Health Information Prediction System 110: Voice signal processing unit 120: Symptom prediction unit 130: Questionnaire No. 1 140: Disease Prediction Section 150: 2nd Survey Section 160: Control Section 170 : Recommendation Section 180: Database

Claims

Claim 1 In a personal health information prediction system based on AI voice recognition, the system comprises: a symptom prediction unit that applies text conversion data regarding a user's voice signal related to the user's major symptoms to a first prediction model to predict probability values ​​for each minor symptom, and applies the probability values ​​for each minor symptom to a second prediction model to predict probability values ​​for each major symptom; a first survey unit that presents a candidate list of sub-minor symptoms for the major symptom derived with the highest probability value and receives a selection of a plurality of corresponding minor symptoms from the user; a disease prediction unit that inputs the selected plurality of minor symptoms into a disease prediction model to predict a plurality of expected diseases associated with the input data and the major symptom with the highest correlation to the plurality of expected diseases; and a second survey unit that presents a candidate list of sub-minor symptoms for the major symptom with the highest correlation to the plurality of expected diseases and receives a selection of a plurality of corresponding minor symptoms. A personal health information prediction system comprising a control unit that inputs all minor symptoms selected by the first and second survey units into the disease prediction model to finally derive n predicted diseases of the user, wherein the disease prediction model includes a first model that outputs probability values ​​for each candidate disease by deep learning analysis of a plurality of input minor symptoms, and a second model that outputs probability values ​​for a plurality of major symptoms by deep learning analysis of a plurality of probability values ​​for each candidate disease received from the first model. Claim 2 A personal health information prediction system according to claim 1, wherein the first and second questionnaires extract a list of sub-symptoms corresponding to a corresponding major symptom from a database that has previously stored a list of sub-symptoms for each of a plurality of major symptoms and present it as a candidate list. Claim 3 A personal health information prediction system according to claim 1, further comprising a voice signal processing unit that receives a user voice signal related to the user's main symptoms through a user terminal and converts it into text. Claim 4 In claim 3, the voice signal processing unit converts a sentence corresponding to an input voice signal into text, separates morphemes, encodes them as integers, and then performs padding processing; a personal health information prediction system. Claim 5 A personal health information prediction system according to claim 1, wherein the control unit provides the final derived n predicted diseases to a user terminal as a disease prediction result. Claim 6 A personal health information prediction system according to claim 1, wherein the control unit provides the top n candidate diseases with the highest probability values ​​among a plurality of candidate disease-specific probability values ​​derived by deep learning analysis of input symptoms through the disease prediction model, sorted in descending order. Claim 7 delete Claim 8 A personal health information prediction system according to claim 1, wherein the sub-symptoms for the major symptom include information regarding accompanying symptoms related to the major symptom and the site of symptom manifestation. Claim 9 A personal health information prediction system according to claim 1, further comprising a recommendation unit that recommends a medical department to a user based on the finally derived predicted disease. Claim 10 A personal health information prediction system according to claim 1, implemented to link with a mobile app executable on a user terminal.

Images