A Big Data-Based Intelligent Analysis System and Method for Tongue Diagnosis
By acquiring tongue features through a tongue diagnosis instrument and constructing a tongue-lesion correlation data model, the subjectivity and data management problems of traditional tongue diagnosis are solved, enabling precise auxiliary analysis of tongue diagnosis and prediction of lesion risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING DAJING TCM INFORMATION TECH CO LTD
- Filing Date
- 2025-03-10
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional tongue diagnosis relies on physician experience, is highly subjective, has poor diagnostic accuracy and consistency, lacks systematic data recording and analysis tools, and current tongue diagnosis instruments have low analytical precision and cannot make accurate predictions by combining the patient's historical lesion diagnosis information.
Tongue images are obtained using a tongue diagnostic instrument. Tongue features are extracted using image recognition technology. A tongue image-lesion correlation data model is constructed by combining historical data, and cluster analysis is performed to generate a lesion risk fluctuation map. Early warning information is then fed back to assist in diagnosis.
It improves the diagnostic accuracy and consistency of tongue diagnosis, enables effective prediction of the risk of lesions in patients, and ensures the accuracy of tongue diagnosis auxiliary analysis results.
Smart Images
Figure CN120108706B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tongue diagnosis auxiliary analysis technology, specifically to a tongue diagnosis auxiliary intelligent analysis system and method based on big data. Background Technology
[0002] Tongue diagnosis, which involves observing the tongue's color, shape, and coating, is an important tool for traditional Chinese medicine practitioners to assess a patient's health and disease type. However, traditional tongue diagnosis relies on the physician's experience and subjective judgment, and has the following problems:
[0003] Highly subjective: Different doctors may have different interpretations of the same tongue appearance, affecting the accuracy and consistency of diagnosis;
[0004] Experience dependence: High-level tongue diagnosis requires long-term accumulation, and young doctors may lack the diagnostic ability;
[0005] Inconvenient data management: Traditional tongue diagnosis lacks systematic data recording and analysis tools, making it difficult to conduct large-scale case studies and follow-ups.
[0006] With the rapid development of image analysis technology, tongue diagnosis instruments have gradually become an important auxiliary tool for TCM doctors in disease diagnosis. However, the existing tongue diagnosis instruments do not have high accuracy in analyzing lesions in tongue diagnosis images, and they cannot combine the patient's historical lesion diagnosis information to accurately predict the current tongue diagnosis results during the auxiliary analysis process. Therefore, the existing technology has significant shortcomings. Summary of the Invention
[0007] The purpose of this invention is to provide a big data-based tongue diagnosis-assisted intelligent analysis system and method to solve the problems mentioned in the background art.
[0008] To address the aforementioned technical problems, this invention provides the following technical solution: a tongue diagnosis-assisted intelligent analysis method based on big data, the method comprising the following steps:
[0009] S1. Obtain the tongue diagnosis image of the patient through the tongue diagnosis instrument, extract the tongue image features based on image recognition technology, and combine the disease diagnosis information of the doctor for the corresponding tongue diagnosis image in the historical data to construct the tongue image-lesion association data model through cluster analysis technology.
[0010] S2. Based on time-series analysis technology, the binding results of tongue features and corresponding multivariate detection information of the same patient at different times are summarized to construct the health record of the corresponding patient;
[0011] S3. Based on the constructed tongue image-lesion association data model, perform lesion association matching on the tongue image features at different times in the health records of the corresponding patients, generate a lesion risk fluctuation map based on time-series information for the corresponding patients, and predict the comorbid lesion risk warning information of the patients at the current time based on the patient's most recent medical diagnosis information.
[0012] S4. Feedback the patient's risk warning information for complications at the current time to the diagnosing doctor of the corresponding tongue diagnosis image at the current time, to assist the doctor in making diagnostic decisions for the patient's condition.
[0013] Furthermore, in the process of extracting tongue features from the tongue diagnosis image acquired by the tongue diagnosis instrument based on image recognition technology, the tongue diagnosis image is separated by color thresholding method, and the tongue coating area and the tongue body area in the tongue diagnosis image are marked in different ways; the color thresholding method identifies the tongue coating area and the tongue body area in the tongue diagnosis image by querying the preset tongue coating color in the database.
[0014] The tongue appearance features include tongue coating color, tongue body color, and tongue coating texture features;
[0015] The tongue coating color represents the average grayscale value of each pixel within the tongue coating area; the tongue body color represents the average grayscale value of each pixel within the tongue body area; the tongue coating texture feature represents the texture pattern formed by each tongue coating texture node and the corresponding tongue coating texture node within the tongue coating area; the grayscale value corresponding to each pixel within the tongue coating area is obtained, and the set of pixels whose corresponding grayscale values belong to the preset grayscale threshold range of tongue coating texture in the database is recorded as the candidate set of tongue coating texture nodes; the grayscale values of each pixel within the surrounding n×n pixel area of each pixel within the tongue coating area are obtained, and if the difference between the grayscale value of the center pixel within the corresponding n×n pixel area and the grayscale values of the other pixels within the corresponding n×n pixel area is less than or equal to the preset threshold, then the center pixel within the corresponding n×n pixel area is taken as an element in the tongue coating texture node comparison set; each pixel in the candidate set of tongue coating texture nodes that does not belong to the tongue coating texture node comparison set is taken as a tongue coating texture node.
[0016] In the process of acquiring tongue image features, this invention analyzes the tongue coating color, tongue body color, and tongue coating texture features from three perspectives, thereby providing data support for the subsequent acquisition of the first and second tongue image cluster features corresponding to each lesion. When acquiring tongue coating texture features, the tongue coating texture node comparison set is obtained in order to filter the elements in the tongue coating texture node candidate set and use the elements in the tongue coating texture node comparison set as interference items for the tongue coating texture nodes.
[0017] Furthermore, in step S1, during the construction of the tongue image-lesion association analysis data model using clustering analysis technology, the diagnostic information of the disease for the corresponding patient in the historical data is obtained. The patient's tongue image is then bound to the corresponding diagnostic information, whereby the diagnostic information represents the set of diagnoses made by the corresponding doctor based on the tongue image. Each element in the diagnostic set corresponds to a type of lesion. The tongue image features of the tongue images corresponding to the same lesion in the historical data are summarized. The tongue image association data corresponding to the lesion type in the tongue image-lesion association data model is recorded as tongue image clustering features. The tongue image clustering features include a first tongue image clustering feature and a second tongue image clustering feature.
[0018] The steps for obtaining the clustering features of the first tongue image are as follows:
[0019] S111. Obtain the grayscale difference between the tongue coating color and the tongue body color corresponding to each tongue diagnosis image; summarize the grayscale difference between the tongue coating color and the tongue body color of each tongue diagnosis image corresponding to the same lesion in ascending order to generate the first tongue image clustering feature analysis sequence of the corresponding lesion.
[0020] S112. Calculate the average value of the gray level difference of each element in the first tongue image cluster feature analysis sequence of the corresponding lesion, and record it as the mean deviation of the corresponding lesion.
[0021] S113. Calculate the standard deviation of the gray-level difference values corresponding to each element in the first tongue image clustering feature analysis sequence of the corresponding lesion, and record it as the tongue image clustering degree of the corresponding lesion. Then compare the tongue image clustering degree of the corresponding lesion with the preset value.
[0022] If the tongue image clustering degree of the corresponding lesion is less than or equal to the preset value, or if the number of elements in the first tongue image clustering feature analysis sequence of the corresponding lesion is unique, then the mean deviation of the corresponding lesion is transmitted to step S114.
[0023] If the clustering degree of the corresponding lesion is greater than the preset value, the element with the largest absolute value of the difference between the corresponding gray level difference and the mean deviation of the corresponding lesion in the first tongue image clustering feature analysis sequence of the corresponding lesion will be removed to obtain a new first tongue image clustering feature analysis sequence of the corresponding lesion, and the process will jump to step S112.
[0024] S114. Use the received results as the first tongue image clustering feature corresponding to the lesion;
[0025] The steps for obtaining the clustering features of the second tongue image are as follows:
[0026] S121. Summarize the tongue coating texture features of each tongue diagnosis image corresponding to the same lesion in sequence to generate the second tongue image clustering feature analysis sequence of the corresponding lesion.
[0027] S122. Take the first element in the cluster feature analysis sequence of the second tongue image of the corresponding lesion as the initial reference texture feature;
[0028] S123. Obtain the element corresponding to the current reference texture feature in the second tongue image clustering feature analysis sequence of the corresponding lesion and mark it. Obtain the first unmarked element closest to the marked element in the second tongue image clustering feature analysis sequence of the corresponding lesion and record it as the fusion element.
[0029] S124. Adjust the position of the blending element to maximize the overlap area between the adjusted blending element and the corresponding area of the reference texture feature. Save the position of the adjusted blending element and use the union area of the adjusted blending element and the reference texture feature as the new reference texture feature.
[0030] If the fusion element is the last element in the second tongue image clustering feature analysis sequence of the corresponding lesion, then the new reference texture feature is transmitted to step S125;
[0031] If the fusion element is not the last element in the second tongue image clustering feature analysis sequence of the corresponding lesion, then the new reference texture feature is transmitted to step S123.
[0032] S125. Receive the reference texture features transmitted in step S124, and analyze the ratio of the number of times each pixel position in the received reference texture features overlaps with the positions of each adjusted fusion element stored to the total number of elements in the second tongue image clustering feature analysis sequence of the corresponding lesion. This ratio is recorded as the clustering index corresponding to the corresponding element in the received reference texture features. The pixel positions in the received reference texture features whose corresponding clustering index is greater than the preset clustering index are grouped into a set and recorded as the second tongue image clustering feature of the corresponding lesion.
[0033] In this invention, when acquiring the first and second tongue image clustering features, the different tongue image features are screened through clustering. Specifically, when acquiring the first tongue image clustering feature, the elements within the analysis sequence of the first tongue image clustering feature of the lesion are dynamically updated through iteration, thereby continuously optimizing the mean deviation of the corresponding lesion and ensuring the accuracy of the first tongue image clustering feature corresponding to the acquired lesion.
[0034] Furthermore, in the binding results of the tongue image features and corresponding multivariate detection information corresponding to the same patient at different times, the interval between the acquisition time of the tongue diagnosis image corresponding to the tongue image features and the detection time corresponding to the bound multivariate detection information is less than or equal to a preset value; the multivariate detection information includes a summary set of one or more corresponding detection results from various preset body detection items.
[0035] In the process of constructing the health record of the corresponding patient in S2, the acquisition time of the tongue diagnosis image corresponding to the tongue image features is used as the time sequence reference. The binding result of the tongue image features, the corresponding multivariate detection information and the corresponding medical diagnosis information corresponding to each acquisition time is used as a health record summary element. The health record summary elements are sorted and summarized in the order of acquisition time to obtain the health record of the corresponding patient. The medical diagnosis information is the diagnosis result of the corresponding doctor on the type of lesion of the patient based on the corresponding tongue image features and multivariate detection information in the historical data.
[0036] Furthermore, in the process of generating the lesion risk fluctuation map based on time-series information for the corresponding patient in S3, tongue image features at different times in the corresponding patient's health record are obtained. The grayscale difference between the tongue coating color and the tongue body color in the tongue image features at time t in the corresponding patient's health record is denoted as At, and the tongue coating texture feature in the tongue image features at time t in the corresponding patient's health record is denoted as Bt. (At, Bt) is input into the tongue image-lesion association data model corresponding to each lesion to obtain the lesion risk value based on the m-th lesion, denoted as F. (m，t) The calculation formula is as follows:
[0037] F (m，t) =P1 (m，t) +P2 (m，t)
[0038] Among them, P1 (m，t) This represents the similarity between At and the first tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; P1 (m，t) =1-|At-AC m | / AC m AC m P2 represents the first tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; (m，t) P2 represents the similarity between Bt and the second tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; (m，t) The value is equal to the quotient of the maximum value of the overlapping pixels between the region corresponding to the second tongue image clustering feature in the tongue image-lesion association data model of the m-th lesion after the translation transformation of the region corresponding to Bt, divided by the total number of pixels in the region corresponding to Bt, and the total number of pixels in the region corresponding to Bt is greater than 0; if the total number of pixels in the region corresponding to Bt is equal to 0, then P2 is determined. (m，t) The value is 0;
[0039] Constructing lesion risk association data pairs (t, F) (m，t) ), and assign each (t, F) to different values of t. (m，t)Mark the time and lesion risk value on the coordinate system, and connect adjacent marked points in the coordinate system in chronological order. The resulting line graph is the risk fluctuation graph of the m-th lesion based on the time information of the corresponding patient.
[0040] Furthermore, the risk warning information of the patient's comorbidities at the current time in S3 includes a predicted set of comorbidities and a risk prediction value for each comorbidity.
[0041] Obtain the lesion diagnosis set from the patient's most recent medical diagnosis information, where each element in the lesion diagnosis set corresponds to a type of lesion diagnosed by a doctor; combine the patient's most recent medical diagnosis information to analyze the lesion matching evaluation value of the patient at the current time based on the m-th type of lesion, calculated using the following formula:
[0042] Em = F (m，) ·(1+g·μ)
[0043] Where Em represents the lesion matching assessment value of the patient based on the m-th lesion at the current time; F (m，) The current time of the patient's tongue diagnosis image is based on the lesion risk value corresponding to the m-th lesion; μ represents the preset conversion coefficient; g represents the weighting coefficient. If the lesion diagnosis set in the patient's most recent medical diagnosis information contains the m-th lesion, then g = 1; otherwise, g = 0.
[0044] The prediction set of the comorbidities represents the set of all lesion types in each Em corresponding to different values of m, where the corresponding value is greater than or equal to the preset matching assessment value; the risk prediction value of each comorbidity is equal to the lesion matching assessment value of the corresponding lesion type in the prediction set of the comorbidities for the patient at the current time.
[0045] A big data-based tongue diagnosis-assisted intelligent analysis system, the system comprising the following modules:
[0046] The lesion association model building module acquires the tongue diagnosis images of the corresponding patients through the tongue diagnosis instrument, extracts tongue features from the tongue diagnosis images acquired by the tongue diagnosis instrument based on image recognition technology, and combines the disease diagnosis information of doctors for the patients to whom the corresponding tongue diagnosis images belong in historical data to construct a tongue image-lesion association data model through cluster analysis technology.
[0047] The patient record construction module, based on time-series analysis technology, summarizes the binding results of tongue features and corresponding multivariate detection information of the same patient at different times to construct the health record of the corresponding patient.
[0048] The lesion risk warning analysis module matches the tongue image features at different times in the corresponding patient's health record according to the constructed tongue image-lesion association data model, generates a lesion risk fluctuation map based on time-series information for the corresponding patient, and predicts the patient's comorbid lesion risk warning information at the current time based on the patient's most recent medical diagnosis information.
[0049] The feedback management module feeds back the patient's risk warning information for complications at the current time to the diagnosing doctor of the corresponding tongue diagnosis image at the current time, assisting the doctor in making diagnostic decisions for the patient's condition.
[0050] Furthermore, the lesion risk early warning analysis module includes a lesion risk fluctuation map construction unit and a risk early warning information generation unit.
[0051] The lesion risk fluctuation map construction unit performs lesion correlation matching on the tongue image features at different times in the health records of the corresponding patients according to the constructed tongue image-lesion association data model, and generates the lesion risk fluctuation map of the corresponding patients based on time-series information.
[0052] The risk warning information generation unit predicts the risk warning information of the patient's comorbidities at the current time based on the patient's most recent medical diagnosis information.
[0053] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention utilizes image recognition technology to extract tongue features from historical tongue diagnosis images, and continuously optimizes the reference standards for tongue clustering features in the tongue image association data corresponding to each lesion based on cluster analysis, thereby ensuring the accuracy of the constructed tongue image-lesion association analysis data model; at the same time, in the process of tongue diagnosis auxiliary analysis, it takes into account the patient's own medical diagnosis information, realizes effective prediction of the patient's lesion risk, and ensures the accuracy of the tongue diagnosis auxiliary analysis results for the patient. Attached Figure Description
[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0055] Figure 1 This is a schematic diagram of the structure of a tongue diagnosis-assisted intelligent analysis system based on big data according to the present invention;
[0056] Figure 2 This is a flowchart illustrating a big data-based tongue diagnosis-assisted intelligent analysis method according to the present invention. Detailed Implementation
[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0058] Please see Figure 1 The present invention provides a technical solution: a tongue diagnosis-assisted intelligent analysis system based on big data, the system comprising the following modules:
[0059] The lesion association model building module acquires the tongue diagnosis images of the corresponding patients through the tongue diagnosis instrument, extracts tongue features from the tongue diagnosis images acquired by the tongue diagnosis instrument based on image recognition technology, and combines the disease diagnosis information of doctors for the patients to whom the corresponding tongue diagnosis images belong in historical data to construct a tongue image-lesion association data model through cluster analysis technology.
[0060] The patient record construction module, based on time-series analysis technology, summarizes the binding results of tongue features and corresponding multivariate detection information of the same patient at different times to construct the health record of the corresponding patient.
[0061] The lesion risk early warning analysis module includes a lesion risk fluctuation map construction unit and a risk early warning information generation unit.
[0062] The lesion risk fluctuation map construction unit performs lesion correlation matching on the tongue image features at different times in the health records of the corresponding patients according to the constructed tongue image-lesion association data model, and generates the lesion risk fluctuation map of the corresponding patients based on time-series information.
[0063] The risk warning information generation unit predicts the risk warning information of the patient's comorbidities at the current time based on the patient's most recent medical diagnosis information.
[0064] The feedback management module feeds back the patient's risk warning information for complications at the current time to the diagnosing doctor of the corresponding tongue diagnosis image at the current time, assisting the doctor in making diagnostic decisions for the patient's condition.
[0065] like Figure 2 As shown, a tongue diagnosis-assisted intelligent analysis method based on big data is described, the method comprising the following steps:
[0066] S1. Obtain the tongue diagnosis image of the patient through the tongue diagnosis instrument, extract the tongue image features based on image recognition technology, and combine the disease diagnosis information of the doctor for the corresponding tongue diagnosis image in the historical data to construct the tongue image-lesion association data model through cluster analysis technology.
[0067] In the process of extracting tongue features from the tongue diagnosis image acquired by the tongue diagnosis instrument based on image recognition technology, the tongue diagnosis image is separated by color thresholding method, and the tongue coating area and the tongue body area in the tongue diagnosis image are marked in different ways; the color thresholding method identifies the tongue coating area and the tongue body area in the tongue diagnosis image by querying the preset tongue coating color in the database.
[0068] The tongue appearance features include tongue coating color, tongue body color, and tongue coating texture features;
[0069] The tongue coating color represents the average grayscale value of each pixel within the tongue coating area; the tongue body color represents the average grayscale value of each pixel within the tongue body area; the tongue coating texture feature represents the texture pattern formed by each tongue coating texture node and the corresponding tongue coating texture node within the tongue coating area; the grayscale value corresponding to each pixel within the tongue coating area is obtained, and the set of pixels whose corresponding grayscale values belong to the preset grayscale threshold range of tongue coating texture in the database is recorded as the candidate set of tongue coating texture nodes; the grayscale values of each pixel within the surrounding n×n pixel area of each pixel within the tongue coating area are obtained, and if the difference between the grayscale value of the center pixel within the corresponding n×n pixel area and the grayscale values of the other pixels within the corresponding n×n pixel area is less than or equal to the preset threshold, then the center pixel within the corresponding n×n pixel area is taken as an element in the tongue coating texture node comparison set; each pixel in the candidate set of tongue coating texture nodes that does not belong to the tongue coating texture node comparison set is taken as a tongue coating texture node.
[0070] In this embodiment, the elements in the tongue texture node comparison set are the relatively discrete interference pixels corresponding to the tongue texture nodes.
[0071] In step S1, during the construction of the tongue image-lesion association analysis data model using clustering analysis technology, the following steps are taken: First, the diagnostic information of patients associated with corresponding tongue images from historical data is obtained. The patient's tongue image is then linked to the corresponding diagnostic information, where the diagnostic information represents the set of diagnoses made by the doctor based on the tongue image. Each element in the diagnostic set corresponds to a type of lesion. Second, the tongue image features of the same lesion in the historical data are summarized. Third, the tongue image association data corresponding to the lesion type in the tongue image-lesion association data model is recorded as tongue image clustering features. These tongue image clustering features include a first tongue image clustering feature and a second tongue image clustering feature.
[0072] The steps for obtaining the clustering features of the first tongue image are as follows:
[0073] S111. Obtain the grayscale difference between the tongue coating color and the tongue body color corresponding to each tongue diagnosis image; summarize the grayscale difference between the tongue coating color and the tongue body color of each tongue diagnosis image corresponding to the same lesion in ascending order to generate the first tongue image clustering feature analysis sequence of the corresponding lesion.
[0074] S112. Calculate the average value of the gray level difference of each element in the first tongue image cluster feature analysis sequence of the corresponding lesion, and record it as the mean deviation of the corresponding lesion.
[0075] S113. Calculate the standard deviation of the gray-level difference values corresponding to each element in the first tongue image clustering feature analysis sequence of the corresponding lesion, and record it as the tongue image clustering degree of the corresponding lesion. Then compare the tongue image clustering degree of the corresponding lesion with the preset value.
[0076] If the tongue image clustering degree of the corresponding lesion is less than or equal to the preset value, or if the number of elements in the first tongue image clustering feature analysis sequence of the corresponding lesion is unique, then the mean deviation of the corresponding lesion is transmitted to step S114.
[0077] If the clustering degree of the corresponding lesion is greater than the preset value, the element with the largest absolute value of the difference between the corresponding gray level difference and the mean deviation of the corresponding lesion in the first tongue image clustering feature analysis sequence of the corresponding lesion will be removed to obtain a new first tongue image clustering feature analysis sequence of the corresponding lesion, and the process will jump to step S112.
[0078] S114. Use the received results as the first tongue image clustering feature corresponding to the lesion;
[0079] In this embodiment, if the first tongue image cluster feature analysis sequence of the lesion contains 4 elements, they are denoted as r1, r2, r3 and r4 respectively, and r1 < r2 < r3 < r4;
[0080] If the mean deviation of the corresponding lesion is r5, and r1 < r2 < r3 < r5 < r4;
[0081] If the tongue appearance aggregation degree of the corresponding lesion is r6,
[0082] but
[0083] If r6 is greater than the preset value, and r6-r1>r4-r6, then r1 in the first tongue image cluster feature analysis sequence of the lesion will be deleted, and the elements in the new first tongue image cluster feature analysis sequence of the corresponding lesion will be r2, r3 and r4.
[0084] The steps for obtaining the clustering features of the second tongue image are as follows:
[0085] S121. Summarize the tongue coating texture features of each tongue diagnosis image corresponding to the same lesion in sequence to generate the second tongue image clustering feature analysis sequence of the corresponding lesion.
[0086] S122. Take the first element in the cluster feature analysis sequence of the second tongue image of the corresponding lesion as the initial reference texture feature;
[0087] S123. Obtain the element corresponding to the current reference texture feature in the second tongue image clustering feature analysis sequence of the corresponding lesion and mark it. Obtain the first unmarked element closest to the marked element in the second tongue image clustering feature analysis sequence of the corresponding lesion and record it as the fusion element.
[0088] S124. Adjust the position of the blending element to maximize the overlap area between the adjusted blending element and the corresponding area of the reference texture feature. Save the position of the adjusted blending element and use the union area of the adjusted blending element and the reference texture feature as the new reference texture feature.
[0089] If the fusion element is the last element in the second tongue image clustering feature analysis sequence of the corresponding lesion, then the new reference texture feature is transmitted to step S125;
[0090] If the fusion element is not the last element in the second tongue image clustering feature analysis sequence of the corresponding lesion, then the new reference texture feature is transmitted to step S123.
[0091] S125. Receive the reference texture features transmitted in step S124, and analyze the ratio of the number of times each pixel position in the received reference texture features overlaps with the positions of each adjusted fusion element stored to the total number of elements in the second tongue image clustering feature analysis sequence of the corresponding lesion. This ratio is recorded as the clustering index corresponding to the corresponding element in the received reference texture features. The pixel positions in the received reference texture features whose corresponding clustering index is greater than the preset clustering index are grouped into a set and recorded as the second tongue image clustering feature of the corresponding lesion.
[0092] S2. Based on time-series analysis technology, the binding results of tongue features and corresponding multivariate detection information of the same patient at different times are summarized to construct the health record of the corresponding patient;
[0093] In the binding results of the tongue image features and corresponding multivariate detection information of the same patient at different times, the interval between the acquisition time of the tongue diagnosis image corresponding to the tongue image features and the detection time corresponding to the bound multivariate detection information is less than or equal to a preset value; the multivariate detection information includes a summary set of one or more corresponding detection results from various preset body detection items.
[0094] In the process of constructing the health record of the corresponding patient in S2, the acquisition time of the tongue diagnosis image corresponding to the tongue image features is used as the time sequence reference. The binding result of the tongue image features, the corresponding multivariate detection information and the corresponding medical diagnosis information corresponding to each acquisition time is used as a health record summary element. The health record summary elements are sorted and summarized in the order of acquisition time to obtain the health record of the corresponding patient. The medical diagnosis information is the diagnosis result of the corresponding doctor on the type of lesion of the patient based on the corresponding tongue image features and multivariate detection information in the historical data.
[0095] S3. Based on the constructed tongue image-lesion association data model, perform lesion association matching on the tongue image features at different times in the health records of the corresponding patients, generate a lesion risk fluctuation map based on time-series information for the corresponding patients, and predict the comorbid lesion risk warning information of the patients at the current time based on the patient's most recent medical diagnosis information.
[0096] In the process of generating the lesion risk fluctuation map based on time-series information for the corresponding patient in S3, tongue features at different times in the patient's health record are obtained. The grayscale difference between the tongue coating color and the tongue body color in the tongue features at time t in the patient's health record is denoted as At, and the tongue coating texture feature in the tongue features at time t in the patient's health record is denoted as Bt. (At, Bt) are input into the tongue image-lesion association data model corresponding to each lesion to obtain the lesion risk value based on the m-th lesion, denoted as F. (m，t) The calculation formula is as follows:
[0097] F (m，t) =P1 (m，t) +P2 (m，t)
[0098] Among them, P1 (m，t) This represents the similarity between At and the first tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; P1 (m，t) =1-|At-AC m | / AC m AC m P2 represents the first tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; (m，t) P2 represents the similarity between Bt and the second tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; (m，t) The value is equal to the quotient of the maximum value of the overlapping pixels between the region corresponding to the second tongue image clustering feature in the tongue image-lesion association data model of the m-th lesion after the translation transformation of the region corresponding to Bt, divided by the total number of pixels in the region corresponding to Bt, and the total number of pixels in the region corresponding to Bt is greater than 0; if the total number of pixels in the region corresponding to Bt is equal to 0, then P2 is determined.(m，t) The value is 0;
[0099] Constructing lesion risk association data pairs (t, F) (m，t) ), and assign each (t, F) to different values of t. (m，t) Mark the time and lesion risk value on the coordinate system, and connect adjacent marked points in the coordinate system in chronological order. The resulting line graph is the risk fluctuation graph of the m-th lesion based on the time information of the corresponding patient.
[0100] In this embodiment, the coordinate system corresponding to time and lesion risk value is constructed with o as the origin, time as the x-axis and lesion risk value as the y-axis; the lesion risk fluctuation chart of the same patient's health record based on different lesions may differ;
[0101] The risk warning information of the patient's comorbidities in S3 at the current time includes a predicted set of comorbidities and a risk prediction value for each comorbidity.
[0102] Obtain the lesion diagnosis set from the patient's most recent medical diagnosis information, where each element in the lesion diagnosis set corresponds to a type of lesion diagnosed by a doctor; combine the patient's most recent medical diagnosis information to analyze the lesion matching evaluation value of the patient at the current time based on the m-th type of lesion, calculated using the following formula:
[0103] Em = F (m，) ·(1+g·μ)
[0104] Where Em represents the lesion matching assessment value of the patient based on the m-th lesion at the current time; F (m，) The current time of the patient's tongue diagnosis image is based on the lesion risk value corresponding to the m-th lesion; μ represents the preset conversion coefficient; g represents the weighting coefficient. If the lesion diagnosis set in the patient's most recent medical diagnosis information contains the m-th lesion, then g = 1; otherwise, g = 0.
[0105] The prediction set of the comorbidities represents the set of all lesion types in each Em corresponding to different values of m, where the corresponding value is greater than or equal to the preset matching assessment value; the risk prediction value of each comorbidity is equal to the lesion matching assessment value of the corresponding lesion type in the prediction set of the comorbidities for the patient at the current time.
[0106] S4. Feedback the patient's risk warning information for complications at the current time to the diagnosing doctor of the corresponding tongue diagnosis image at the current time, to assist the doctor in making diagnostic decisions for the patient's condition.
[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0108] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A tongue diagnosis-assisted intelligent analysis method based on big data, characterized in that, The method includes the following steps: S1. Obtain the tongue diagnosis image of the corresponding patient through the tongue diagnosis instrument, extract the tongue image features based on image recognition technology, and combine the disease diagnosis information of the doctor for the corresponding tongue diagnosis image in historical data to construct the tongue image-lesion association data model through cluster analysis technology. In step S1, during the construction of the tongue image-lesion association analysis data model using clustering analysis technology, the following steps are taken: First, the diagnostic information of patients associated with corresponding tongue images from historical data is obtained. The patient's tongue image is then linked to the corresponding diagnostic information, where the diagnostic information represents the set of diagnoses made by the doctor based on the tongue image. Each element in the diagnostic set corresponds to a type of lesion. Second, the tongue image features of the same lesion in the historical data are summarized. Third, the tongue image association data corresponding to the lesion type in the tongue image-lesion association data model is recorded as tongue image clustering features. These tongue image clustering features include a first tongue image clustering feature and a second tongue image clustering feature. The steps for obtaining the clustering features of the first tongue image are as follows: S111. Obtain the grayscale difference between the tongue coating color and the tongue body color corresponding to each tongue diagnosis image; summarize the grayscale difference between the tongue coating color and the tongue body color of each tongue diagnosis image corresponding to the same lesion in ascending order to generate the first tongue image clustering feature analysis sequence of the corresponding lesion. S112. Calculate the average value of the gray level difference of each element in the first tongue image cluster feature analysis sequence of the corresponding lesion, and record it as the mean deviation of the corresponding lesion. S113. Calculate the standard deviation of the gray-level difference values corresponding to each element in the first tongue image clustering feature analysis sequence of the corresponding lesion, and record it as the tongue image clustering degree of the corresponding lesion. Then compare the tongue image clustering degree of the corresponding lesion with the preset value. If the tongue image clustering degree of the corresponding lesion is less than or equal to the preset value, or if the number of elements in the first tongue image clustering feature analysis sequence of the corresponding lesion is unique, then the mean deviation of the corresponding lesion is transmitted to step S114. If the clustering degree of the corresponding lesion is greater than the preset value, the element with the largest absolute value of the difference between the corresponding gray level difference and the mean deviation of the corresponding lesion in the first tongue image clustering feature analysis sequence of the corresponding lesion will be removed to obtain a new first tongue image clustering feature analysis sequence of the corresponding lesion, and the process will jump to step S112. S114. Use the received results as the first tongue image clustering feature corresponding to the lesion; The steps for obtaining the clustering features of the second tongue image are as follows: S121. Summarize the tongue coating texture features of each tongue diagnosis image corresponding to the same lesion in sequence to generate the second tongue image clustering feature analysis sequence of the corresponding lesion. S122. Take the first element in the cluster feature analysis sequence of the second tongue image of the corresponding lesion as the initial reference texture feature; S123. Obtain the element corresponding to the current reference texture feature in the second tongue image clustering feature analysis sequence of the corresponding lesion and mark it. Obtain the first unmarked element closest to the marked element in the second tongue image clustering feature analysis sequence of the corresponding lesion and record it as the fusion element. S124. Adjust the position of the blending element to maximize the overlap area between the adjusted blending element and the corresponding area of the reference texture feature. Save the position of the adjusted blending element and use the union area of the adjusted blending element and the reference texture feature as the new reference texture feature. If the fusion element is the last element in the second tongue image clustering feature analysis sequence of the corresponding lesion, then the new reference texture feature is transmitted to step S125; If the fusion element is not the last element in the second tongue image clustering feature analysis sequence of the corresponding lesion, then the new reference texture feature is transmitted to step S123. S125. Receive the reference texture features transmitted in step S124, and analyze the ratio of the number of times each pixel position in the received reference texture features overlaps with the positions of each adjusted fusion element stored to the total number of elements in the second tongue image clustering feature analysis sequence of the corresponding lesion. This ratio is recorded as the clustering index corresponding to the corresponding element in the received reference texture features. The pixel positions in the received reference texture features whose corresponding clustering index is greater than the preset clustering index are grouped into a set and recorded as the second tongue image clustering feature of the corresponding lesion. S2. Based on time-series analysis technology, the binding results of tongue features and corresponding multivariate detection information of the same patient at different times are summarized to construct the health record of the corresponding patient; S3. Based on the constructed tongue image-lesion association data model, perform lesion association matching on the tongue image features at different times in the health records of the corresponding patients, generate a lesion risk fluctuation map based on time-series information for the corresponding patients, and predict the comorbid lesion risk warning information of the patients at the current time based on the patient's most recent medical diagnosis information. S4. Feedback the patient's risk warning information for complications at the current time to the diagnosing doctor of the corresponding tongue diagnosis image at the current time, to assist the doctor in making diagnostic decisions for the patient's condition.
2. The tongue diagnosis-assisted intelligent analysis method based on big data according to claim 1, characterized in that: In the process of extracting tongue features from the tongue diagnosis image acquired by the tongue diagnosis instrument based on image recognition technology, the tongue diagnosis image is separated by color thresholding method, and the tongue coating area and the tongue body area in the tongue diagnosis image are marked in different ways. The color thresholding method identifies the tongue coating area and tongue body area in the tongue diagnosis image by querying the preset tongue coating color in the database. The tongue appearance features include tongue coating color, tongue body color, and tongue coating texture features; The tongue coating color represents the average grayscale value of each pixel within the tongue coating area; the tongue body color represents the average grayscale value of each pixel within the tongue body area; the tongue coating texture feature represents the texture pattern formed by each tongue coating texture node and the corresponding tongue coating texture node within the tongue coating area; the grayscale value corresponding to each pixel within the tongue coating area is obtained, and the set of pixels whose corresponding grayscale values belong to the preset grayscale threshold range of tongue coating texture in the database is recorded as the candidate set of tongue coating texture nodes; the grayscale values of each pixel within the surrounding n×n pixel area of each pixel within the tongue coating area are obtained, and if the difference between the grayscale value of the center pixel within the corresponding n×n pixel area and the grayscale values of the other pixels within the corresponding n×n pixel area is less than or equal to the preset threshold, then the center pixel within the corresponding n×n pixel area is taken as an element in the tongue coating texture node comparison set; each pixel in the candidate set of tongue coating texture nodes that does not belong to the tongue coating texture node comparison set is taken as a tongue coating texture node.
3. The tongue diagnosis-assisted intelligent analysis method based on big data according to claim 2, characterized in that: In the binding results of the tongue image features and corresponding multivariate detection information of the same patient at different times, the interval between the acquisition time of the tongue diagnosis image corresponding to the tongue image features and the detection time corresponding to the bound multivariate detection information is less than or equal to a preset value; the multivariate detection information includes a summary set of one or more corresponding detection results from various preset body detection items. In the process of constructing the health record of the corresponding patient in S2, the acquisition time of the tongue diagnosis image corresponding to the tongue image features is used as the time sequence reference. The binding result of the tongue image features, the corresponding multivariate detection information and the corresponding medical diagnosis information corresponding to each acquisition time is used as a health record summary element. The health record summary elements are sorted and summarized in the order of acquisition time to obtain the health record of the corresponding patient. The medical diagnosis information is the diagnosis result of the corresponding doctor on the type of lesion of the patient based on the corresponding tongue image features and multivariate detection information in the historical data.
4. The tongue diagnosis-assisted intelligent analysis method based on big data according to claim 3, characterized in that: In the process of generating the lesion risk fluctuation map based on time-series information for the corresponding patient in S3, tongue features at different times in the patient's health record are obtained. The grayscale difference between the tongue coating color and the tongue body color in the tongue features at time t in the patient's health record is denoted as At, and the tongue coating texture feature in the tongue features at time t in the patient's health record is denoted as Bt. (At, Bt) are input into the tongue image-lesion association data model corresponding to each lesion to obtain the lesion risk value based on the m-th lesion, denoted as F. （m，t） The calculation formula is as follows: F （m，t） =P1 （m，t） +P2 （m，t） Among them, P1 （m，t） This represents the similarity between At and the first tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; P1 （m，t） =1-|At-AC m | / AC m AC m P2 represents the first tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; （m，t） P2 represents the similarity between Bt and the second tongue image clustering feature in the tongue image-lesion association data model corresponding to the m-th lesion; （m，t） The value is equal to the quotient of the maximum value of the overlapping pixels between the region corresponding to the second tongue image clustering feature in the tongue image-lesion association data model of the m-th lesion after the translation transformation of the region corresponding to Bt, divided by the total number of pixels in the region corresponding to Bt, and the total number of pixels in the region corresponding to Bt is greater than 0; if the total number of pixels in the region corresponding to Bt is equal to 0, then P2 is determined. （m，t） The value is 0; Construct lesion risk association data pairs (t, F) （m，t） ), and for each value of t, the corresponding (t, F) values. （m，t） Mark the time and lesion risk value on the coordinate system, and connect adjacent marked points in the coordinate system in chronological order. The resulting line graph is the risk fluctuation graph of the m-th lesion based on the time information of the corresponding patient.
5. The tongue diagnosis-assisted intelligent analysis method based on big data according to claim 4, characterized in that: The risk warning information of the patient's comorbidities in S3 at the current time includes a predicted set of comorbidities and a risk prediction value for each comorbidity. Obtain the lesion diagnosis set from the patient's most recent medical diagnosis information, where each element in the lesion diagnosis set corresponds to a type of lesion diagnosed by a doctor; combine the patient's most recent medical diagnosis information to analyze the lesion matching evaluation value of the patient at the current time based on the m-th type of lesion, calculated using the following formula: Em=F （m，） ·(1+g·μ) Where Em represents the lesion matching assessment value of the patient based on the m-th lesion at the current time; F （m，） The current time of the patient's tongue diagnosis image is based on the lesion risk value corresponding to the m-th lesion; μ represents the preset conversion coefficient; g represents the weighting coefficient. If the lesion diagnosis set in the patient's most recent medical diagnosis information contains the m-th lesion, then g=1 is determined; otherwise, g=0 is determined. The prediction set of the comorbidities represents the set of all lesion types in each Em corresponding to different values of m, where the corresponding value is greater than or equal to the preset matching assessment value; the risk prediction value of each comorbidity is equal to the lesion matching assessment value of the corresponding lesion type in the prediction set of the comorbidities for the patient at the current time.
6. A big data-based tongue diagnosis-assisted intelligent analysis system, wherein the tongue diagnosis-assisted intelligent analysis system operates using the big data-based tongue diagnosis-assisted intelligent analysis method according to any one of claims 1-5, characterized in that, The system includes the following modules: The lesion association model building module acquires the tongue diagnosis images of the corresponding patients through the tongue diagnosis instrument, extracts tongue features from the tongue diagnosis images acquired by the tongue diagnosis instrument based on image recognition technology, and combines the disease diagnosis information of doctors for the patients to whom the corresponding tongue diagnosis images belong in historical data to construct a tongue image-lesion association data model through cluster analysis technology. The patient record construction module, based on time-series analysis technology, summarizes the binding results of tongue features and corresponding multivariate detection information of the same patient at different times to construct the health record of the corresponding patient. The lesion risk warning analysis module matches the tongue image features at different times in the corresponding patient's health record according to the constructed tongue image-lesion association data model, generates a lesion risk fluctuation map based on time-series information for the corresponding patient, and predicts the patient's comorbid lesion risk warning information at the current time based on the patient's most recent medical diagnosis information. The feedback management module feeds back the patient's risk warning information for complications at the current time to the diagnosing doctor of the corresponding tongue diagnosis image at the current time, assisting the doctor in making diagnostic decisions for the patient's condition.
7. The tongue diagnosis-assisted intelligent analysis system based on big data according to claim 6, characterized in that: The lesion risk early warning analysis module includes a lesion risk fluctuation map construction unit and a risk early warning information generation unit. The lesion risk fluctuation map construction unit performs lesion correlation matching on the tongue image features at different times in the health records of the corresponding patients according to the constructed tongue image-lesion association data model, and generates the lesion risk fluctuation map of the corresponding patients based on time-series information. The risk warning information generation unit predicts the risk warning information of the patient's comorbidities at the current time based on the patient's most recent medical diagnosis information.