Intelligent hospital guide system for oral health

Abstract

The invention discloses an intelligent diagnosis guide system for oral health, relates to the technical field of intelligent medical treatment, and solves the problems that in existing oral diagnosis guide, the description voice of a patient is not clear due to partial pathological changes of the oral cavity; the diagnosis guide analysis result error is large, and the selection accuracy of the diagnosis guide department is reduced. Comprising a data acquisition module for acquiring initial patient information, and a hospital guide analysis module for extracting test audio information in voice information and acquiring test keyword data; generating a test audio state based on the test audio information and the test keyword data; extracting description audio information in the voice information, and generating description text information from the description audio information based on the test audio state; generating a first hospital guide keyword based on the description text information and the text information, and inputting the image information into an image recognition model to obtain a second hospital guide keyword; generating a hospital guide department based on the first hospital guide keyword and the second hospital guide keyword; and the accuracy of hospital guide is improved.

Description

Technical Field

[0001] This application belongs to the field of intelligent medical technology, specifically an oral health intelligent guidance system. Background Art

[0002] The intelligent medical guidance system integrates advanced technologies such as intelligent voice interaction, medical knowledge recognition and reasoning, and implements core functions such as intelligent triage, AI-assisted self-examination, precise medical guidance, and educational knowledge recommendations. It aims to help patients accurately find the right department and doctor, thereby avoiding registration errors. It also helps hospitals reduce the workload of medical guidance and achieve precise matching of medical and patient resources. For specific regions or cities, the system can classify and divert patients according to their condition, promoting the efficient utilization and balanced development of regional medical resources, and thus supporting the implementation of tiered diagnosis and treatment policies.

[0003] The prior art (publication number CN118230985A, an invention patent application) discloses an oral medical guidance method and system based on a plug-in instruction fine-tuning large model, which belongs to the field of artificial intelligence technology. The present invention introduces medical-related knowledge to the general field large model while making its dialogue thinking conform to medical standards. By constructing an instruction data set that conforms to medical dialogue thinking, the plug-in instruction fine-tuning large model acquires medical dialogue thinking and can simulate doctors conducting multiple rounds of inquiries to patients. At the same time, the constructed medical field instruction data set is classified according to medical sub-disciplines, and medical professional knowledge is supplemented to the model through plug-in instruction fine-tuning. The present invention has a stronger ability to solve actual guidance problems, especially solving the problems of traditional knowledge graph methods with small coverage, inaccurate understanding of user intentions, and cumbersome guidance steps.

[0004] The above-mentioned oral medicine guidance system analyzes the information provided by relevant personnel through a training model and recommends departments based on the analysis results. However, different patients have different descriptive abilities. Especially in the field of dentistry, patients may have inaccurate pronunciation due to lesions in certain areas of the mouth, or due to the unclear pronunciation of certain specific words in the dialect used by the patients, which will lead to a decrease in the accuracy of the analysis results based on the patient's description and the recommended departments based on them. Therefore, there is an urgent need to develop an intelligent oral health guidance system. Summary of the Invention

[0005] The present application provides an intelligent oral health guidance system and device, which solves the technical problem in the existing oral guidance technology that the patient's description voice is unclear due to oral lesions, which leads to large errors in the guidance analysis results and reduces the accuracy of the guidance department selection.

[0006] To achieve the above objectives, this application adopts the following technical solutions: In a first aspect, an oral health intelligent guidance system is provided, comprising: a data acquisition module, a guidance analysis module and a database; The data acquisition module is used to obtain initial patient information, including text information, voice information and image information; The guidance analysis module extracts the test audio information from the voice information and obtains the test keyword data; generates the test audio status based on the test audio information and the test keyword data; Extracting descriptive audio information from the voice information, generating descriptive text information from the descriptive audio information based on the test audio state; generating a first diagnostic guidance keyword based on the descriptive text information and the text information, and inputting the image information into the image recognition model to obtain a second diagnostic guidance keyword; Generate a guidance department based on the first guidance keyword and the second guidance keyword.

[0007] Based on the above technical solution, in an oral health intelligent guidance system provided by the present application, by obtaining initial patient information, test audio information in the voice information is extracted to obtain test keyword data; a test audio state is generated based on the test audio information and the test keyword data; descriptive audio information in the voice information is extracted, and descriptive text information is generated from the descriptive audio information based on the test audio state; a first guidance keyword is generated based on the descriptive text information and the text information, and the image information is input into the image recognition model to obtain a second guidance keyword; a guidance department is generated based on the first guidance keyword and the second guidance keyword; the patient's descriptive audio is calibrated through the analysis results of the test audio, so that some unclear or incorrect pronunciations in the descriptive audio will be corrected, so that the accuracy of the data finally analyzed is improved, thereby improving the accuracy of the guidance.

[0008] In combination with the first aspect above, in a possible implementation, generating the test audio state based on the test audio information and the test keyword data includes: Extracting test audio corresponding to several test keywords from the test audio data; extracting standard audio corresponding to several test keywords from the test keyword data; the standard audio is the standard audio data set corresponding to the test keyword; Integrate the test audio and standard audio corresponding to the test keyword into a keyword analysis group corresponding to the test keyword; input the standard audio in the keyword analysis group as the reference audio and the test audio as the variable audio into the audio state analysis model to obtain the keyword label corresponding to the test keyword; the keyword state is the label of the test audio corresponding to the test keyword, and the pronunciation type of the corresponding test keyword can be found based on the label; Obtaining the keyword status corresponding to each of the test keywords, and arranging the keyword status labels in each of the keyword statuses in a set order to integrate them into a test feature vector; extracting a test audio state feature vector lookup table in the test keyword data; the test audio state feature vector lookup table comprises a corresponding relationship between a plurality of audio states and corresponding feature vectors thereof; finding a pronunciation type based on the test feature vector as a type feature vector in the test audio state feature vector lookup table, and taking the pronunciation type as the test audio state.

[0009] In combination with the first aspect, in a possible implementation manner, the test audio state feature vector lookup table is constructed in the following manner: obtaining a plurality of pronunciation types and type audios corresponding to a plurality of test keywords under the pronunciation types through a database; obtaining standard audios corresponding to the test keywords; taking the standard audios corresponding to the test keywords as reference audios, taking the type audios corresponding to the test keywords as variable audios, and inputting the reference audios and the variable audios into an audio state analysis model to obtain corresponding keyword states of the test keywords under the corresponding pronunciation types; obtaining keyword states corresponding to a plurality of type audios under the pronunciation types in sequence, and arranging and integrating keyword state labels in each keyword state in a set order to form a type feature vector corresponding to the pronunciation type; integrating each pronunciation type and the corresponding type feature vector to form a test audio state feature vector lookup table.

[0010] In combination with the first aspect, in a possible implementation manner, a training manner of the audio state analysis model comprises: obtaining standard audios of a plurality of test keywords and a plurality of type audios corresponding to the test keywords through a database; generating keyword states of the test keywords under the corresponding pronunciation types based on the standard audios of the test keywords and the type audios corresponding to the test keywords; taking the standard audios corresponding to the test keywords as reference audios and taking the type audios as variable audios; and integrating a plurality of training data and verification data based on the reference audios, the variable audios, and the corresponding keyword states; training an artificial intelligence model using the training data, verifying the trained artificial intelligence model using the verification data, and finally obtaining an audio state analysis model with input of the reference audios and the variable audios and output of the keyword states corresponding to the variable audios.

[0011] In combination with the first aspect, in a possible implementation manner, a keyword state is obtained in the following manner: obtaining standard audios corresponding to the test keywords and a plurality of type audios corresponding to each pronunciation type; Extracting a plurality of audio features from the standard audio, and integrating the audio features into a standard feature vector corresponding to the test keyword in a set order; Extracting multiple audio features from the type audio, and integrating the audio features into a type feature vector corresponding to the test keyword in a set order; Performing cluster analysis on each standard feature vector and type feature vector to obtain a number of clusters, setting different cluster numbers for each cluster in turn, and obtaining a cluster center feature vector corresponding to the cluster center in each cluster; the cluster center feature vector includes a standard feature vector or a type feature vector; The number of the cluster is used as the keyword status label corresponding to the test keyword under the corresponding pronunciation type; the cluster center feature vector is used as the pronunciation feature vector corresponding to the test keyword under the corresponding pronunciation type; Obtain the keyword status label and pronunciation feature vector of each test keyword under each pronunciation type in turn; The keyword state label and pronunciation feature vector are integrated into the keyword state of the corresponding test keyword under the corresponding pronunciation type.

[0012] In conjunction with the first aspect above, in a possible implementation, generating the descriptive text information from the descriptive audio information based on the test audio state includes: When the test audio state is a normal type, obtaining description audio information, and converting the description audio information into description text information; the description audio information can be converted into text information through natural speech recognition technology; When the test audio state is a pathological type or a dialect type, the pronunciation feature vector corresponding to each test keyword is obtained, and the description audio information is corrected based on the pronunciation feature vector to obtain the corrected audio information; and the corrected audio information is converted into a description text.

[0013] In conjunction with the first aspect above, in a possible implementation, correcting the description audio information based on the pronunciation feature vector to obtain the corrected audio information includes: Obtaining a pronunciation feature vector corresponding to each test keyword, and when the pronunciation feature vector is a standard feature vector, marking the test keyword as a normal keyword; otherwise, marking the test keyword as a keyword to be corrected; Dividing the described audio information into a plurality of sub-audios; wherein one audio segment contains a syllable; extracting a plurality of audio features from the sub-audios, and integrating the plurality of audio features into a sub-audio feature vector corresponding to the sub-audio in a set order; calculating Euclidean distances between the sub-audio feature vector and pronunciation feature vectors corresponding to each of the to-be-corrected keywords, and marking the sub-audio as to-be-replaced audio when the Euclidean distance is less than a set distance threshold, and setting standard audio of a test keyword corresponding to the to-be-corrected keyword as replacement audio of the to-be-replaced audio; replacing sub-audio of the to-be-replaced audio with replacement audio corresponding to the replacement audio; replacing sub-audio of the to-be-replaced audio with replacement audio corresponding to the replacement audio; replacing sub-audio of the to-be-replaced audio with replacement audio corresponding to the replacement audio;

[0014] In a possible implementation manner of the first aspect, the first guide keyword is generated based on the description text information and the text information, and the method comprises the following steps. splitting the description text information and the text information into a plurality of keywords, when the keyword belongs to the oral cavity vocabulary, marking the keyword as a to-be-classified keyword; when the to-be-classified keyword belongs to the conventional vocabulary, obtaining an attribute of the to-be-classified keyword; the attribute comprises a position attribute and a symptom attribute; when the to-be-classified keyword belongs to the dialect conversion vocabulary, obtaining a professional vocabulary corresponding to the to-be-classified keyword in the dialect conversion vocabulary and an attribute corresponding to the professional vocabulary; when the to-be-classified keyword belongs to the colloquialism conversion vocabulary, obtaining a professional vocabulary corresponding to the to-be-classified keyword in the colloquialism conversion vocabulary and an attribute corresponding to the professional vocabulary; The oral cavity vocabulary comprises a conventional vocabulary, a dialect conversion vocabulary and a colloquialism conversion vocabulary, the conventional vocabulary comprises a plurality of professional vocabularies related to the oral cavity and attributes corresponding to the professional vocabularies; the dialect conversion vocabulary comprises a plurality of professional vocabularies, attributes corresponding to the professional vocabularies and a plurality of dialect vocabularies of the professional vocabularies in dialects, and a corresponding relationship between the professional vocabularies and the dialect vocabularies; the colloquialism conversion vocabulary comprises a plurality of professional vocabularies, attributes corresponding to the professional vocabularies and a plurality of colloquial vocabularies of the professional vocabularies in colloquialisms, and a corresponding relationship between the professional vocabularies and the colloquial vocabularies; the professional vocabulary is a professional vocabulary related to the oral cavity; sorting each to-be-classified keyword and its corresponding attribute according to an order of appearance of each to-be-classified keyword in the description text information or the text information to obtain a to-be-classified keyword sorting table; Determining whether the attributes of adjacent keywords to be classified are the same according to the order of the keyword to be classified sorting table; when the attributes of adjacent keywords to be classified are the same, deleting the keyword to be classified that is ranked earlier in the keyword to be classified sorting table, and updating the keyword to be classified sorting table; and continuing to determine whether the attributes of adjacent keywords to be classified are the same according to the order of the keyword to be classified sorting table; When the attributes of adjacent keywords to be classified are different, the professional vocabulary and attributes corresponding to the two adjacent keywords to be classified are integrated into a guidance phrase; and the two adjacent keywords to be classified in the keyword ranking table to be classified are deleted, and the keyword ranking table to be classified is updated; and the attributes of adjacent keywords to be classified are continuously determined in sequence according to the order of the keyword ranking table to be classified; End when there is only one keyword to be classified in the keyword sorting table; Obtain several guidance phrases in sequence; and integrate each guidance phrase into a first guidance keyword.

[0015] In conjunction with the first aspect above, in one possible implementation, a training method of the image recognition model includes: Acquire multiple oral image information, as well as positional attributes and presentation attributes corresponding to various parts in the image information; the presentation attributes are the presentation states of the corresponding parts; presentation attributes include labels set by professionals based on the states of the corresponding parts in the oral image information, such as normal and swollen presentation attributes corresponding to the gums; integrate each image information and its corresponding positional attributes and presentation attributes into several sets of training data and test data; The artificial intelligence model is trained using training data, and the trained artificial intelligence model is tested using test data, so that the input is image information and the output is position attributes and expression attributes corresponding to several parts in the image information; each position attribute and expression attribute is integrated into a guidance phrase, and each guidance phrase is integrated into a second guidance keyword, and the second guidance keyword is used as the final output to obtain an image recognition model with image information as input and second guidance keyword as output; wherein the artificial intelligence model includes a deep convolutional neural network, etc.

[0016] In combination with the first aspect above, in a possible implementation, generating the guidance department based on the first guidance keyword and the second guidance keyword includes: Obtain each guidance phrase in the first guidance keyword and the second guidance keyword; integrate each guidance phrase into a guidance feature phrase, Obtain departmental characteristic phrases set by each department; the departmental characteristic phrases include several diagnostic phrases corresponding to the department; the diagnostic phrases include location attributes and diagnostic attributes; the location attribute is the location corresponding to the oral cavity part, and the diagnostic attribute is the manifestation attribute and symptom attribute of the corresponding oral cavity part, that is, the departmental characteristic phrases are phrases consisting of several oral cavity parts that the corresponding department is responsible for diagnosing and treating and the states of the corresponding parts; Obtain the number of guidance phrases in the guidance feature phrase group that are identical to the number of diagnosis phrases in each department feature phrase group, and record the ratio of the number to the number of guidance phrases in the guidance feature phrase group as the matching degree of the department feature phrase group corresponding to the department; and select the department with the highest matching degree as the guidance department.

[0017] The present application provides an intelligent oral health guidance system, which can obtain initial patient information, extract test audio information from voice information, and obtain test keyword data; generate a test audio state based on the test audio information and the test keyword data; extract descriptive audio information from the voice information, and generate descriptive text information from the descriptive audio information based on the test audio state; generate a first guidance keyword based on the descriptive text information and the text information, input image information into an image recognition model to obtain a second guidance keyword; generate a guidance department based on the first guidance keyword and the second guidance keyword; calibrate the patient's descriptive audio through the analysis results of the test audio, so that some unclear or incorrect pronunciations in the descriptive audio will be corrected, so that the accuracy of the data finally analyzed is improved, thereby improving the accuracy of the guidance.

[0018] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in this application does not imply that all features and advantages can be realized in any single embodiment. On the contrary, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution or beneficial effect is included in at least one embodiment. Therefore, the description of a technical feature, technical solution or beneficial effect in this specification does not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in the present embodiment can also be combined in any appropriate manner. Those skilled in the art will understand that the embodiment can be implemented without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in specific embodiments that do not embody all embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 This is a schematic diagram of the module connection of the oral health intelligent guidance system in this application; Figure 2 This is a schematic diagram of the execution process of the oral health intelligent guidance system in this application; Figure 3 A schematic diagram of the process for generating text information described in this application. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions of this application in conjunction with the embodiments. Obviously, the embodiments described are only a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0022] Example 1 See also Figures 1-2 ,In a first aspect, an oral health intelligent guidance system is provided, comprising: a data acquisition module, a guidance analysis module and a database; The data acquisition module includes a text acquisition unit, a voice acquisition unit and an image acquisition unit; the text acquisition unit is used to acquire text information provided by the patient; the voice acquisition unit is used to acquire voice information provided by the patient; the image acquisition unit is used to acquire image information provided by the patient; the image information is an image of the patient's oral cavity; The guidance analysis module includes a test audio analysis unit, a guidance keyword generation unit, and a guidance department division unit; The test audio analysis unit is used to extract test audio information from the voice information, where the test audio information is the audio of the patient's specific keywords recorded by the reception desk; obtain test keyword data; the test keyword data is the relevant data of several keywords set to be extracted; and generate a test audio state based on the test audio information and the test keyword data; The guidance keyword generation unit is used to extract the descriptive audio information from the voice information, and generate descriptive text information from the descriptive audio information based on the test audio state; generate a first guidance keyword based on the descriptive text information and the text information, the first guidance keyword being the guidance keyword obtained through the patient's subjective description; input the image information into the image recognition model to obtain a second guidance keyword, the second guidance feature keyword being the guidance keyword obtained through the patient's objective image; The guidance department division unit generates guidance departments based on the first guidance keyword and the second guidance keyword; the guidance departments are departments within the stomatology department, including oral medicine, oral and maxillofacial surgery, prosthodontics, orthodontics, and oral mucosa department; The database is used to store the relevant data required in this application and the generated relevant data.

[0023] Based on the above technical solution, in an oral health intelligent guidance system provided by the present application, by obtaining initial patient information, extracting test audio information from voice information, and obtaining test keyword data; generating a test audio state based on the test audio information and the test keyword data; extracting descriptive audio information from the voice information, and generating descriptive text information from the descriptive audio information based on the test audio state; generating a first guidance keyword based on the descriptive text information and the text information, and inputting the image information into the image recognition model to obtain a second guidance keyword; generating a guidance department based on the first guidance keyword and the second guidance keyword; calibrating the patient's descriptive audio through the analysis results of the test audio, so that some unclear or incorrect pronunciations in the descriptive audio will be corrected, so that the accuracy of the data finally analyzed is improved; thereby improving the accuracy of the guidance; avoiding errors caused by unclear or inaccurate description data due to lesions in some parts of the patient's mouth or dialect reasons.

[0024] In one possible implementation, the test audio state is generated based on the test audio information and the test keyword data, including: extracting test audio corresponding to several test keywords in the test audio data; the test keywords are keywords set by professionals for testing the pronunciation of patients, including "a", "s", "sh", "f" and "v", etc., or Chinese characters corresponding to letter pronunciations, and the test audio is audio data corresponding to the test keywords entered by the patient at the reception desk; extracting standard audio corresponding to several test keywords in the test keyword data; the standard audio is standard audio data set corresponding to the test keywords; Integrate the test audio and standard audio corresponding to the test keyword into a keyword analysis group corresponding to the test keyword; input the standard audio in the keyword analysis group as the reference audio and the test audio as the variable audio into the audio state analysis model to obtain the keyword label corresponding to the test keyword; the keyword state includes the label of the test audio corresponding to the test keyword, and the type of pronunciation corresponding to the test keyword can be found based on the label; Obtain the keyword status corresponding to each of the test keywords, and arrange the keyword status labels in each of the keyword statuses in a set order to form a test feature vector; the test feature vector is a vector generated by the test audio corresponding to each of the test keywords of this patient, and is used to represent the audio features of this test; extracting a test audio state feature vector lookup table in the test keyword data; the test audio state feature vector lookup table comprises a corresponding relationship between a plurality of audio states and corresponding feature vectors thereof; finding a pronunciation type based on the test feature vector as a type feature vector in the test audio state feature vector lookup table, taking the pronunciation type as a test audio state; the pronunciation type is a pronunciation characteristic that the patient is likely to have, such as a normal type, a type with a dialect, or a type with a certain oral lesion, which is analyzed from the pronunciation characteristic of the test audio of the test keyword.

[0025] In a possible implementation, the test audio state feature vector lookup table is constructed in the following manner: a plurality of pronunciation types and type audios corresponding to a plurality of test keywords under the pronunciation types are obtained from a database; the pronunciation types include a plurality of normal types, lesion types, and dialect types; the type audio is audio of each test keyword recorded by a plurality of related persons under a corresponding state; obtaining standard audios corresponding to the test keywords; taking the standard audios corresponding to the test keywords as reference audios, taking the type audios corresponding to the test keywords as variable audios, and inputting the reference audios and the variable audios into an audio state analysis model to obtain corresponding keyword states of the test keywords under corresponding pronunciation types; for example, a patient with missing front teeth or a cleft lip has a large difference from normal pronunciation when pronouncing "f" and "v", so the keyword states corresponding to "f" and "v" are different from the normal state; when the patient has soft palate paralysis, the pronunciation of "m" and "n" is different from the normal pronunciation; obtaining keyword states corresponding to a plurality of type audios under the pronunciation type in sequence, and arranging and integrating keyword state labels in each keyword state in a set order to form a type feature vector corresponding to the pronunciation type; the type feature vector is a vector for representing the characteristics of the pronunciation type, which is composed of pronunciation characteristics of each test keyword corresponding to the pronunciation type; integrating each pronunciation type and the corresponding type feature vector into the test audio state feature vector lookup table.

[0026] In a possible implementation, a training method of the audio state analysis model includes: obtaining standard audios of several test keywords through a database, and several types of audios from several pronunciation types corresponding to the test keywords; the pronunciation types include several normal types, lesion types and dialect types; the type audios include normal audios of keywords corresponding to the normal type, lesion audios of keywords corresponding to the lesion type and dialect audios of keywords corresponding to the dialect type; the lesion audios of keywords are audios of test keywords recorded by patients with lesions in a certain part of the oral cavity under the condition of lesions in the oral cavity; the dialect type of keywords is audios of test keywords recorded in a certain dialect condition; Generate a keyword state of the test keyword under the corresponding pronunciation type based on the standard audio of the test keyword and the type audio corresponding to the test keyword; The standard audio corresponding to the test keyword is used as the reference audio, and the type audio is used as the variable audio; based on the reference audio, the variable audio and the corresponding keyword status, a plurality of training data and test data are integrated; The artificial intelligence model is trained using training data, and the trained artificial intelligence model is tested using test data, ultimately obtaining an audio state analysis model whose input is reference audio and variable audio, and whose output is the keyword state corresponding to the variable audio; wherein the artificial intelligence model includes a neural network model, etc.

[0027] This embodiment trains the audio state analysis model through a large amount of data, simplifies the analysis steps through the model, and greatly improves the operating efficiency during analysis.

[0028] In a possible implementation, a method for obtaining the keyword status includes: obtaining a standard audio corresponding to the test keyword and a plurality of types of audio corresponding to each pronunciation type; Extracting several audio features from the standard audio, including frequency domain features and time domain features such as fundamental frequency, voice-to-noise ratio, formant frequency, and harmonic structure; integrating each of the audio features into a standard feature vector corresponding to the test keyword in a set order; Extracting multiple audio features from the type audio, and integrating each of the audio features into a type feature vector corresponding to the test keyword in a set order; it is understandable that the arrangement order of the corresponding audio features in the type feature vector and the standard feature vector is the same; Performing cluster analysis on each standard feature vector and type feature vector to obtain a number of clusters, setting different cluster numbers for each cluster in turn, and obtaining a cluster center feature vector corresponding to the cluster center in each cluster; the cluster center feature vector includes a standard feature vector or a type feature vector; The number of the cluster is used as the keyword status label corresponding to the test keyword under the corresponding pronunciation type; the cluster center feature vector is used as the pronunciation feature vector corresponding to the test keyword under the corresponding pronunciation type; Obtain the keyword status label and pronunciation feature vector of each test keyword under each pronunciation type in turn; integrate the keyword status label and pronunciation feature vector into the keyword status of the corresponding test keyword under the corresponding pronunciation type.

[0029] See also Figure 3 In a possible implementation, generating the descriptive text information from the descriptive audio information based on the test audio state includes: When the test audio state is a normal type, obtaining description audio information, and converting the description audio information into description text information; the description audio information can be converted into text information through natural speech recognition technology; When the test audio state is a pathological type or a dialect type, the pronunciation feature vector corresponding to each test keyword is obtained, and the description audio information is corrected based on the pronunciation feature vector to obtain the corrected audio information; and the corrected audio information is converted into a description text.

[0030] This embodiment calibrates the patient's descriptive speech so that the final converted text will not be inaccurate due to unclear pronunciation caused by the patient's oral lesions or other reasons, which in turn will lead to the final triage error.

[0031] In a possible implementation, the step of correcting the description audio information based on the pronunciation feature vector to obtain the corrected audio information includes: Obtaining a pronunciation feature vector corresponding to each test keyword, and when the pronunciation feature vector is a standard feature vector, marking the test keyword as a normal keyword; otherwise, marking the test keyword as a keyword to be corrected; Dividing the described audio information into a plurality of sub-audios; each sub-audio contains a syllable; extracting a plurality of audio features from the sub-audio, and integrating the plurality of audio features into a sub-audio feature vector corresponding to the sub-audio in a set order; Calculating the Euclidean distance between the sub-audio feature vector and the pronunciation feature vector corresponding to each of the keywords to be corrected; when the Euclidean distance is less than a set distance threshold, marking the sub-audio as the audio to be replaced, and setting the standard audio of the test keyword corresponding to the keyword to be corrected as the replacement audio for the audio to be replaced; Using the replacement audio as the sub-audio to replace the sub-audio of the audio to be replaced corresponding to the replacement audio; Replace the sub-audios corresponding to the audios to be replaced in sequence; The sub-audios are re-integrated into the corrected audio information in the original order; the corrected audio information is converted into a description text; this embodiment makes the description text converted from the corrected audio information more accurate by replacing the incorrect pronunciation.

[0032] In a possible implementation, generating the first diagnosis guidance keyword based on the descriptive text information and the text information includes: splitting the descriptive text information and the text information into a plurality of keywords, When the keyword belongs to the oral vocabulary, the keyword is marked as a keyword to be classified; When the keyword to be classified belongs to a conventional word library, obtaining attributes of the keyword to be classified; the attributes include position attributes and symptom attributes; When the keyword to be classified belongs to the dialect conversion vocabulary, obtaining the professional vocabulary corresponding to the keyword to be classified in the dialect conversion vocabulary and the attributes corresponding to the professional vocabulary; When the keyword to be classified belongs to the colloquialism conversion lexicon, obtaining the professional vocabulary corresponding to the keyword to be classified in the colloquialism conversion lexicon and the attributes corresponding to the professional vocabulary; The oral vocabulary includes a conventional vocabulary, a dialect conversion vocabulary and a colloquial conversion vocabulary. The conventional vocabulary contains a number of professional vocabulary related to oral cavity and the attributes corresponding to the professional vocabulary; the dialect conversion vocabulary includes a number of professional vocabulary, the attributes corresponding to the professional vocabulary and a number of dialect vocabulary corresponding to the professional vocabulary in the dialect, as well as the corresponding relationship between the professional vocabulary and the dialect vocabulary; the colloquial vocabulary includes a number of professional vocabulary, the attributes corresponding to the professional vocabulary and a number of colloquial vocabulary corresponding to the professional vocabulary in colloquial language, as well as the corresponding relationship between the professional vocabulary and the colloquial vocabulary; the professional vocabulary is professional vocabulary related to oral cavity; Sorting each keyword to be classified and its corresponding attribute according to the order in which each keyword to be classified appears in the description text information or text information to obtain a keyword sorting table to be classified; Determining whether the attributes of adjacent keywords to be classified are the same according to the order of the keyword to be classified sorting table; when the attributes of adjacent keywords to be classified are the same, deleting the keyword to be classified that is ranked earlier in the keyword to be classified sorting table, and updating the keyword to be classified sorting table; and continuing to determine whether the attributes of adjacent keywords to be classified are the same according to the order of the keyword to be classified sorting table; When the attributes of adjacent keywords to be classified are different, the professional vocabulary and attributes corresponding to the two adjacent keywords to be classified are integrated into a guidance phrase; and the two adjacent keywords to be classified in the keyword ranking table to be classified are deleted, and the keyword ranking table to be classified is updated; and the attributes of adjacent keywords to be classified are continuously determined in sequence according to the order of the keyword ranking table to be classified; End when there is only one keyword to be classified in the keyword sorting table; Obtain several guidance phrases in sequence; and integrate each guidance phrase into a first guidance keyword.

[0033] For example, the part of the keyword ranking table to be classified is:

[0034] According to the above table, we first determine whether the attributes corresponding to the keywords to be classified with sequence number 1 and sequence number 2 are the same. If they are the same, we delete the keyword to be classified with sequence number 1 from the keyword to be classified sorting table and update the keyword to be classified sorting table. The updated keyword to be classified sorting table is:

[0035] Then determine whether the attributes corresponding to sequence number 1 and sequence number 2 in the updated keyword sorting table to be classified are the same; if they are not the same, generate a guidance phrase with the professional vocabulary and attributes corresponding to sequence number 1 and sequence number 2, and the guidance phrase is {location attribute: gums; symptom attribute: pain}; delete the keywords to be classified corresponding to sequence number 1 and sequence number 2 in the current keyword sorting table to be classified; update the keyword sorting table to be classified, and the updated keyword sorting table to be classified is:

[0036] Then determine whether the attributes corresponding to sequence number 1 and sequence number 2 in the updated keyword sorting table to be classified are the same; if they are not the same, the professional vocabulary and attributes corresponding to sequence number 1 and sequence number 2 are used to generate a guidance phrase, and the guidance phrase is {position attribute: front teeth; symptom attribute: soreness}; thereby obtaining several groups of guidance phrases, and integrating each guidance phrase into the first guidance keyword.

[0037] Since the text information or voice information described by the patient has a corresponding logical relationship, especially the logical relationship between the location attribute and the symptom attribute, the location attribute and the symptom attribute generally appear one after another. This embodiment generates the first guidance keyword through the logical relationship between the location attribute and the symptom attribute to avoid inaccurate generation of guidance keywords due to the patient's stuttering or other reasons.

[0038] In one possible implementation, a training method for the image recognition model includes: obtaining a plurality of oral cavity image information, and positional attributes and performance attributes corresponding to various parts in the image information; the performance attributes are the presentation states of the corresponding parts; the performance attributes include labels set by professionals based on the states of the corresponding parts in the oral cavity image information, such as normal and swollen performance attributes corresponding to the gums; and integrating the various image information and the corresponding positional attributes and performance attributes into several sets of training data and test data; The artificial intelligence model is trained using training data, and the trained artificial intelligence model is tested using test data, so that the input is image information and the output is position attributes and expression attributes corresponding to several parts in the image information; each position attribute and expression attribute is integrated into a guidance phrase, and each guidance phrase is integrated into a second guidance keyword, and the second guidance keyword is used as the final output to obtain an image recognition model with image information as input and second guidance keyword as output; wherein the artificial intelligence model includes a deep convolutional neural network, etc.

[0039] In a possible implementation, generating the guidance department based on the first guidance keyword and the second guidance keyword includes: obtaining each guidance phrase in the first guidance keyword and the second guidance keyword; integrating each guidance phrase into a guidance feature phrase; Obtain departmental characteristic phrases set by each department; the departmental characteristic phrases include several diagnostic phrases corresponding to the department; the diagnostic phrases include location attributes and diagnostic attributes; the location attribute is the location corresponding to the oral cavity part, and the diagnostic attribute is the manifestation attribute and symptom attribute of the corresponding oral cavity part, that is, the departmental characteristic phrases are phrases consisting of several oral cavity parts that the corresponding department is responsible for diagnosing and treating and the states of the corresponding parts; Obtain the number of guidance phrases in the guidance feature phrase group that are identical to the number of diagnosis phrases in each department feature phrase group, and record the ratio of the number to the number of guidance phrases in the guidance feature phrase group as the matching degree of the department feature phrase group corresponding to the department; and select the department with the highest matching degree as the guidance department.

[0040] Example 2 Different from Example 1, the data collection module is also used to obtain patient personal information; patient personal information includes the patient's education level, age, and medical history, etc. The guidance department division unit in the guidance analysis module is used to generate a guidance department based on the patient's personal information, the first guidance keyword, and the second guidance keyword. Specifically, the patient's personal information is input into the descriptive ability evaluation model to obtain the patient's descriptive ability score. The descriptive ability score is a score calculated by experts based on the patient's ability to accurately describe symptoms based on the data analysis of the patient's personal information. The higher the descriptive ability score, the more accurate the corresponding symptom description; the more referenceable the subjective description; the descriptive ability score value ranges from 0 to 1. Obtain each guidance phrase in the first guidance keyword; Obtain the department characteristic phrases set by each department; obtain the number of diagnosis phrases in each guidance phrase in the first guidance keyword that are identical to the diagnosis phrases in the department characteristic phrases, and record it as K1; Obtain each guidance phrase in the second guidance keyword; obtain the number of each guidance phrase in the second guidance keyword that is identical to the diagnosis phrase in the department characteristic phrase, and record it as K2; The matching degree of the corresponding department is calculated using the formula PD=MP×K1+K2, where MP is the descriptive ability score of the corresponding patient; the department with the highest matching degree is used as the guide department.

[0041] This embodiment sets the reference degree of the corresponding description content during the final triage by taking into account the accuracy of the patient's description content; the worse the patient's description ability is, the lower the reference degree of the corresponding description content during triage; thus, the accuracy of the final triage result is ensured.

[0042] A training method for a descriptive ability evaluation model includes: obtaining a number of different patient personal information and the descriptive ability scores corresponding to the patient personal information through a database; integrating the patient personal information and the descriptive ability scores into a number of sets of training data and test data; using the training data to train an artificial intelligence model; using the test data to test the trained artificial intelligence model; and finally obtaining a descriptive ability evaluation model whose input is the patient personal information and whose output is the descriptive ability score corresponding to the patient; wherein the artificial intelligence model includes a BP neural network model and an RBF neural network model, etc.

[0043] Some of the data in the above formula are calculated by removing the dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0044] How this application works: By obtaining initial patient information, extracting test audio information from voice information, and obtaining test keyword data; generating a test audio state based on the test audio information and test keyword data; extracting descriptive audio information from voice information, and generating descriptive text information from the descriptive audio information based on the test audio state; generating a first guidance keyword based on the descriptive text information and text information, and inputting the image information into an image recognition model to obtain a second guidance keyword; generating a guidance department based on the first guidance keyword and the second guidance keyword; calibrating the patient's descriptive audio through the analysis results of the test audio, so that some unclear or incorrect pronunciations in the descriptive audio will be corrected, thereby improving the accuracy of the data finally analyzed; thereby improving the accuracy of the guidance.

[0045] The above embodiments are only used to illustrate the technical method of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present application.

Claims

1. An oral health intelligent guidance system, characterized in that: include: Data collection module, diagnosis and analysis module and database; The data acquisition module is used to obtain initial patient information, including text information, voice information and image information; The diagnosis guidance analysis module extracts the test audio information from the voice information and obtains the test keyword data; generating a test audio state based on the test audio information and the test keyword data; Extracting descriptive audio information from the voice information, generating descriptive text information from the descriptive audio information based on the test audio state; generating a first diagnostic guidance keyword based on the descriptive text information and the text information, and inputting the image information into the image recognition model to obtain a second diagnostic guidance keyword; Generate a guidance department based on the first guidance keyword and the second guidance keyword.

2. The oral health intelligent guidance system according to claim 1, characterized in that: Generating the test audio state based on the test audio information and the test keyword data includes: Extracting test audio corresponding to several test keywords from the test audio data; extracting standard audio corresponding to several test keywords from the test keyword data; Integrate the test audio and standard audio corresponding to the test keyword into a keyword analysis group corresponding to the test keyword; input the standard audio in the keyword analysis group as the reference audio and the test audio as the variable audio into the audio state analysis model to obtain the keyword label corresponding to the test keyword; Obtaining the keyword status corresponding to each of the test keywords, and arranging the keyword status labels in each of the keyword statuses in a set order to integrate them into a test feature vector; Extracting a test audio state feature vector lookup table from the test keyword data; the test audio state feature vector lookup table includes a correspondence between a plurality of audio states and their corresponding feature vectors; The test feature vector is used as the type feature vector, and the pronunciation type is obtained by searching in the test audio state feature vector lookup table, and the pronunciation type is used as the test audio state.

3. An oral health intelligent guidance system according to claim 2, characterized in that: The test audio state feature vector lookup table is constructed in the following manner: Acquire a number of pronunciation types and type audio corresponding to a number of test keywords under the pronunciation types from a database; the pronunciation types include a number of normal types, pathological types and dialect types; Obtaining standard audio corresponding to the test keyword; using the standard audio corresponding to the test keyword as the reference audio, and the type audio corresponding to the test keyword as the variable audio, and inputting the reference audio and the variable audio into the audio state analysis model to obtain the corresponding keyword state of the test keyword under the corresponding pronunciation type; Sequentially obtain keyword states corresponding to several types of audio under the pronunciation type, and arrange keyword state labels in each keyword state in a set order to form a type feature vector corresponding to the pronunciation type; Each pronunciation type and its corresponding type feature vector are integrated into a test audio state feature vector lookup table.

4. An oral health intelligent guidance system according to claim 2 or 3, characterized in that: A training method of the audio state analysis model includes: Acquire, from a database, standard audios of several test keywords and several types of audios of several pronunciation types corresponding to the test keywords; Generate a keyword state of the test keyword under the corresponding pronunciation type based on the standard audio of the test keyword and the type audio corresponding to the test keyword; The standard audio corresponding to the test keyword is used as the reference audio, and the type audio is used as the variable audio; based on the reference audio, the variable audio and the corresponding keyword status, a plurality of training data and test data are integrated; The artificial intelligence model is trained using training data, and the trained artificial intelligence model is tested using test data, ultimately obtaining an audio state analysis model whose input is the reference audio and variable audio, and whose output is the keyword state corresponding to the variable audio.

5. The oral health intelligent guidance system according to claim 4, characterized in that: One way to obtain the keyword status includes: Obtain the standard audio corresponding to the test keyword and several types of audio corresponding to each pronunciation type; Extracting a plurality of audio features from the standard audio, and integrating the audio features into a standard feature vector corresponding to the test keyword in a set order; Extracting multiple audio features from the type audio, and integrating the audio features into a type feature vector corresponding to the test keyword in a set order; Perform cluster analysis on each standard feature vector and type feature vector to obtain several clusters, set different cluster numbers for each cluster in turn, and obtain the cluster center feature vector corresponding to the cluster center in each cluster; The number of the cluster is used as the keyword status label corresponding to the test keyword under the corresponding pronunciation type; the cluster center feature vector is used as the pronunciation feature vector corresponding to the test keyword under the corresponding pronunciation type; Obtain the keyword status label and pronunciation feature vector of each test keyword under each pronunciation type in turn; The keyword state label and pronunciation feature vector are integrated into the keyword state of the corresponding test keyword under the corresponding pronunciation type.

6. The oral health intelligent guidance system according to claim 1, characterized in that: Generating the description text information from the description audio information based on the test audio state includes: When the test audio state is a normal type, obtaining description audio information, and converting the description audio information into description text information; When the test audio state is a pathological type or a dialect type, the pronunciation feature vector corresponding to each test keyword is obtained, and the description audio information is corrected based on the pronunciation feature vector to obtain the corrected audio information; and the corrected audio information is converted into a description text.

7. An oral health intelligent guidance system according to claim 6, characterized in that: The method of modifying the description audio information based on the pronunciation feature vector to obtain the modified audio information includes: Obtaining a pronunciation feature vector corresponding to each test keyword, and when the pronunciation feature vector is a standard feature vector, marking the test keyword as a normal keyword; otherwise, marking the test keyword as a keyword to be corrected; Dividing the described audio information into a plurality of sub-audios; extracting a plurality of audio features from the sub-audios, and integrating the plurality of audio features into a sub-audio feature vector corresponding to the sub-audios in a set order; Calculating the Euclidean distance between the sub-audio feature vector and the pronunciation feature vector corresponding to each of the keywords to be corrected; when the Euclidean distance is less than a set distance threshold, marking the sub-audio as the audio to be replaced, and setting the standard audio of the test keyword corresponding to the keyword to be corrected as the replacement audio for the audio to be replaced; Using the replacement audio as the sub-audio to replace the sub-audio of the audio to be replaced corresponding to the replacement audio; Replace the sub-audios corresponding to the audios to be replaced in sequence; Reintegrate the sub-audios into corrected audio information according to their original order; and convert the corrected audio information into description text.

8. The oral health intelligent guidance system according to claim 1, characterized in that: Generating the first diagnosis guidance keyword based on the descriptive text information and the text information includes: Split the description text information and text information into several keywords, When the keyword belongs to the oral vocabulary, the keyword is marked as a keyword to be classified; When the keyword to be classified belongs to a conventional word library, obtaining attributes of the keyword to be classified; the attributes include position attributes and symptom attributes; When the keyword to be classified belongs to the dialect conversion vocabulary, obtaining the professional vocabulary corresponding to the keyword to be classified in the dialect conversion vocabulary and the attributes corresponding to the professional vocabulary; When the keyword to be classified belongs to the colloquialism conversion lexicon, obtaining the professional vocabulary corresponding to the keyword to be classified in the colloquialism conversion lexicon and the attributes corresponding to the professional vocabulary; Sorting each keyword to be classified and its corresponding attribute according to the order in which each keyword to be classified appears in the description text information or text information to obtain a keyword sorting table to be classified; Determining whether the attributes of adjacent keywords to be classified are the same according to the order of the keyword to be classified sorting table; when the attributes of adjacent keywords to be classified are the same, deleting the keyword to be classified that is ranked earlier in the keyword to be classified sorting table, and updating the keyword to be classified sorting table; and continuing to determine whether the attributes of adjacent keywords to be classified are the same according to the order of the keyword to be classified sorting table; When the attributes of adjacent keywords to be classified are different, the professional vocabulary and attributes corresponding to the two adjacent keywords to be classified are integrated into a guidance phrase; and the two adjacent keywords to be classified in the keyword ranking table to be classified are deleted, and the keyword ranking table to be classified is updated; and the attributes of adjacent keywords to be classified are continuously determined in sequence according to the order of the keyword ranking table to be classified; End when there is only one keyword to be classified in the keyword sorting table; Obtain several guidance phrases in sequence; and integrate each guidance phrase into a first guidance keyword.

9. The oral health intelligent guidance system according to claim 1, characterized in that: One training method of the image recognition model includes: Acquire a plurality of oral cavity image information, and position attributes and presentation attributes corresponding to each part in the image information; the presentation attributes are presentation states of the corresponding parts; and integrate each image information and its corresponding position attributes and presentation attributes into a plurality of sets of training data and test data; The artificial intelligence model is trained using training data, and the trained artificial intelligence model is tested using test data, so that the input is image information, and the output is position attributes and expression attributes corresponding to several parts in the image information; each position attribute and expression attribute is integrated into a guidance phrase, and each guidance phrase is integrated into a second guidance keyword, and the second guidance keyword is used as the final output, to obtain an image recognition model with image information as input and second guidance keyword as output.

10. The oral health intelligent guidance system according to claim 1, characterized in that: Generating the guidance department based on the first guidance keyword and the second guidance keyword includes: Obtain each guidance phrase in the first guidance keyword and the second guidance keyword; integrate each guidance phrase into a guidance feature phrase, Obtain departmental characteristic phrases set by each department; the departmental characteristic phrases include several diagnostic phrases corresponding to the department; the diagnostic phrases include location attributes and diagnostic attributes; the location attribute is the location corresponding to the part in the oral cavity, and the diagnostic attribute is the manifestation attribute and symptom attribute of the corresponding part in the oral cavity, Obtain the number of guidance phrases in the guidance feature phrase group that are identical to the number of diagnosis phrases in each department feature phrase group, and record the ratio of the number to the number of guidance phrases in the guidance feature phrase group as the matching degree of the department feature phrase group corresponding to the department; and select the department with the highest matching degree as the guidance department.

Images