AI analysis-integrated traditional Chinese medicine four-diagnosis comprehensive diagnosis method and equipment thereof

Abstract

The invention discloses a traditional Chinese medicine four diagnosis comprehensive diagnosis method and device integrated with AI analysis, and relates to the field of intelligent medical treatment, and the method comprises the steps: obtaining user identity information and basic physiological parameters at least comprising age, gender, height and weight, carrying out the correlation binding of the two, and constructing a user exclusive diagnosis file; a preset camera is combined with a color card correction mechanism to collect tongue surface, tongue bottom and face images of a user, the shooting angle and focusing parameters are automatically adjusted in the collection process, and ambient light interference is eliminated through the color card correction mechanism; according to the method, the traditional Chinese medicine syndrome element mapping matrix is converted into the syndrome score, the personalized health report containing multi-dimensional information is generated based on the analysis result, and the file is associated to support multi-mode acquisition, so that traditional Chinese medicine four-diagnosis data acquisition precision, analysis intelligence and diagnosis personalization are integrally realized, the traditional Chinese medicine four-diagnosis efficiency and accuracy are effectively improved, and the traditional Chinese medicine four-diagnosis efficiency and accuracy are improved. And comprehensive and adaptive health assessment is provided for the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical technology, in particular to a traditional Chinese medicine four-examination comprehensive diagnosis method integrated with AI analysis and equipment thereof. BACKGROUND

[0002] The traditional Chinese medicine four-examination instrument is a device that combines the theory of traditional Chinese medicine "observation, listening, questioning and palpation" with modern technology, which can collect information such as tongue appearance, complexion and pulse condition, assist traditional Chinese medicine syndrome differentiation through data analysis, standardize diagnosis and treatment process, and improve diagnosis efficiency, and is commonly used for traditional Chinese medicine clinical auxiliary diagnosis and health management, providing objective data support for traditional Chinese medicine diagnosis and treatment.

[0003] The invention patent application with the application number 202210540736.2 discloses a traditional Chinese medicine model training method based on knowledge structure and four-examination characteristics, which comprises: performing entity recognition on a pre-constructed traditional Chinese medicine corpus dataset to obtain entity results, and classifying the entity results to obtain a medical entity set of each type; constructing a chain sample set according to a pre-set entity superior-inferior relationship for the medical entity set; performing knowledge structure training on the chain sample set to obtain a primary traditional Chinese medicine diagnosis model; obtaining a diagnosis information set of four-examination characteristics, performing text conversion on the diagnosis information set of four-examination characteristics according to a pre-set preprocessing strategy to obtain a four-examination text set; and training the primary traditional Chinese medicine diagnosis model according to a pre-set mask masking method and the four-examination text set to obtain a traditional Chinese medicine diagnosis model. The application aims to solve the problem that the existing training method uses a large amount of text corpus with uneven quality, and the trained model cannot well integrate the logical thinking of traditional Chinese medicine syndrome differentiation and treatment, so it is difficult to be called a traditional Chinese medicine diagnosis model. Therefore, a more effective training method is needed to obtain a traditional Chinese medicine diagnosis model that can reflect the energy of traditional Chinese medicine and increase the diagnosis accuracy.

[0004] However, the operation of the existing traditional Chinese medicine four-examination instrument is biased towards running according to a specified program, and its intelligence level and diagnosis accuracy are relatively low.

[0005] Therefore, we propose a traditional Chinese medicine four-examination comprehensive diagnosis method integrated with AI analysis and equipment thereof. SUMMARY

[0006] In view of the above-mentioned shortcomings of the prior art, the present application provides a traditional Chinese medicine four-examination comprehensive diagnosis method integrated with AI analysis and equipment thereof, which can effectively solve the problems of the prior art.

[0007] To achieve the above purpose, the present application is realized by the following technical solutions:

[0008] The application discloses a traditional Chinese medicine four-examination comprehensive diagnosis method integrated with AI analysis, and comprises the following steps:

[0009] User identity information and basic physiological parameters including at least age, gender, height and weight are acquired, and the two are associated and bound to construct a user-specific diagnosis file; user tongue surface, tongue bottom and face images are collected through a preset camera combined with a color card correction mechanism, the collection process automatically adjusts the shooting angle and focusing parameters, and the color card correction mechanism is used to eliminate environmental light interference; a pulse diagnosis component is started, connection between the pulse diagnosis component and a host device is established through wireless communication, left and right hand pulse data of a user are sequentially collected, and after the collection is completed, the host device performs length checking on the pulse data, and when the data length reaches a preset threshold, the pulse data is determined as valid pulse diagnosis data; the collected tongue image, face image and valid pulse diagnosis data are input into an AI analysis model, the AI analysis model generates multi-round inquiry contents for the user based on traditional Chinese medicine syndrome differentiation theory; multi-round inquiry feedback information of the user is collected, tongue image, face feature data, valid pulse diagnosis data and inquiry feedback information are multi-dimensionally fused, the AI analysis model cross-verify and logically reasons the fused data to determine the traditional Chinese medicine constitution type and specific syndrome type of the user, and simultaneously performs comprehensive analysis of the fused information data; based on the constitution type, syndrome type and comprehensive analysis result output by the AI analysis model, an electronic health report is simultaneously generated, and the report is associated to the user-specific diagnosis file, so that the report supports electronic retrieval, two-dimensional code viewing and paper printing acquisition.

[0010] Further, the identity information is acquired through user certificate recognition or user self-input;

[0011] The user identity information includes a user ID number and a biological feature identifier, the biological feature identifier is collected through an optical sensor built in the host device, and is stored after irreversible hash processing;

[0012] The basic physiological parameters further include a body mass index (BMI), a resting heart rate, a preset historical period weight fluctuation value and a basic medical history;

[0013] The user-specific diagnosis file is stored through a hierarchical structure of basic information + dynamic data, the basic information layer stores the identity and fixed physiological parameters, i.e., age and gender, and the dynamic data layer updates the variable physiological parameters in real time, and the update trigger condition is that the physiological parameter change amplitude exceeds a preset value or the preset period is automatically updated;

[0014] The biological feature identifier includes a fingerprint or a face feature code, and the variable physiological parameters include weight, resting heart rate and weight fluctuation value.

[0015] Further, the stage of collecting the user tongue surface, tongue bottom and face images is subject to the following conditions:

[0016] Camera resolution ≥ 4K, tongue surface shooting pitch angle range 30°-45°, tongue bottom shooting pitch angle range 15°-25°, face shooting yaw angle ≤ ± 5°, auto focus accuracy ≤ 0.1mm, focus response time ≤ 0.3s;

[0017] The color card correction mechanism adopts a 24-color standard color card, which includes 6 gray blocks and 18 color blocks, and the correction process is:

[0018] Synchronize the acquisition of the original pixel data of the color card area and the target image area, extract the average original pixel value of each color block in the color card, and mark it as I raw_card (c);

[0019] Based on the standard reflectivity value of the color card factory calibration, the ambient light influence coefficient is calculated

[0020] Color correction is performed on each pixel point of the target image to obtain the corrected image, and the correction formula is:

[0021]

[0022] In the formula: S(m) is the standard reflectivity value of the mth gray block; I raw_card (m) is the average original pixel value of the mth gray block; I corr (i,j,C) is the pixel value of pixel point (i,j) in the corrected image in color channel C; S(c) is the standard reflectivity value of the cth color block calibrated when the 24-color standard color card leaves the factory; I raw (i,j,C) is the original pixel value of pixel point (i,j) in the target image in color channel C.

[0023] Wherein, c represents the serial number of 1 to 6 gray blocks and 7 to 24 color blocks.

[0024] Further, the AI analysis model architecture is: syndrome feature extraction-questioning logic generation-feedback adaptation;

[0025] The syndrome feature extraction layer maps the tongue image, face image and effective pulse diagnosis data to feature vectors, and outputs the core feature vectors of TCM syndrome differentiation dimensions;

[0026] The questioning logic generation layer dynamically generates multiple rounds of questioning content based on the matching degree of the core feature vector and the TCM syndrome differentiation theory library, and the calculation of the matching degree and the questioning parameters is subject to:

[0027]

[0028] In the formula: M is the matching degree of the core feature vector and the standard feature vector of the syndrome differentiation theory library, M ∈ [0.1]; n is the number of dimensions of the core feature vector; w kis the diagnostic weight of the kth core feature, w k ∈ [0.5, 2]; F k is the actual extraction value of the kth feature in the core feature vector; T k is the standard threshold value in the diagnostic theory library corresponding to the kth feature; Q is the information depth coefficient of the current round of inquiry content; Q0 is the basic information depth coefficient when the matching degree is zero; Q1 is the highest information depth coefficient when the matching degree is 1;

[0029] wherein w k The value is more important for the core feature corresponding to the value, and the value of the core feature corresponding to the value is smaller, Q0 and Q1 are preset values, and the values are 1 and 5 respectively, and the inquiry parameter is represented by Q.

[0030] Further, in the inquiry stage, the user can manually input a symptom keyword not covered, and the host device automatically matches and supplements the corresponding inquiry option based on the input keyword;

[0031] The processing of the host device on the effective pulse diagnosis data further includes preprocessing and feature quantization steps:

[0032] Feature quantization is to convert the extracted time domain and frequency domain features into pulse image quantitative values:

[0033]

[0034] In the formula: F is the floating and sinking amount; A max , A min is the maximum and minimum amplitude of the denoised pulse signal; A ref is the preset standard pulse amplitude reference value; C' is the delay amount; T avg is the average period of the denoised pulse signal in 5 consecutive periods; S is the slippery amount; E main is the frequency band energy corresponding to the main frequency; E sub is the frequency band energy corresponding to the secondary main frequency; E total is the total energy of the pulse signal in the 0-20Hz frequency range.

[0035] Further, when the pulse diagnosis component collects pulse data of the user's left and right hands, it synchronously performs signal interference recognition and elimination processing on the pulse data:

[0036] The three-axis acceleration data A x , A y , A z of the pulse diagnosis component are collected, and the interference intensity score is calculated synchronously:

[0037] A0 represents a preset acceleration reference value;

[0038] Set the interference judgment threshold [a, b], a < b, when I < a, it is judged that there is no significant interference, and the pulse data is normally collected, when a ≤ I < b, it is judged that there is slight interference, the preset vibration feedback unit on the pulse diagnosis component is started, the user is prompted to keep the wrist stable through 50Hz low-frequency micro-vibration, and the collected pulse signal is processed through adaptive filtering, when I ≥ b, it is judged that there is serious interference, the pulse collection is immediately suspended, the user is prompted to adjust the wrist posture through the voice of the host device, and the collection is restarted after I < a;

[0039] The pulse signal after adaptive filtering is secondly checked through calculation of signal-to-noise ratio SNR:

[0040] E s represents the pulse signal main frequency band energy value, that is, 0.5-3Hz, E n represents the energy value of the noise frequency band in the signal, that is, >10Hz;

[0041] When SNR ≥ 15dB, it is judged that the filtered signal is effective, and is included in the pulse diagnosis effective data;

[0042] When SNR < 15dB, it is judged that the filtered signal is still invalid, the re-collection process is triggered, and the maximum re-collection times are set to 2;

[0043] If the SNR ≥ 15dB is still not met after 2 times of re-collection, the text information of pulse diagnosis collection exception is recorded to the user diagnosis file, and the initial weight of tongue appearance and inquiry information is increased by 0.1 when the subsequent AI analysis model is inferred.

[0044] Further, the multi-dimensional fusion is subject to:

[0045] The initial weights of the dimensional data are set as follows: tongue feature 0.3, facial feature 0.2, pulse diagnosis data 0.3, and inquiry information 0.2;

[0046] The weight dynamic adjustment formula is

[0047] In the formula, W i ′ is the adjusted weight, W i is the original weight, β is the adjustment coefficient, R i is the reliability index of the i-th dimensional data; is the average value of the dimensional reliability indexes; σ(R) is the global standard deviation of all dimensional reliability indexes; and ε is a minimum value, which is defined by the user of the system end and initially set as 0.05;

[0048] The fused data is converted into syndrome score through a preset traditional Chinese medicine syndrome element mapping matrix, and the matrix element M s,e represents the association strength between the s-th fused feature and the e-th syndrome element, and the syndrome score Se = M s,e × F s ;

[0049] F = F s ;

[0050] When S e > 0.6, it is determined that the syndrome element is established.

[0051] wherein the adjustment coefficient β ∈ [0.2, 0.6], when the difference of the reliability indexes of the dimensional data is greater, the value is greater, when the difference of the reliability indexes of the dimensional data is smaller, the reliability index R i The difference between the proportion of effective features and the proportion of noise signals in the dimensional data is calculated, and the normalized value is obtained.

[0052] Further, the standardized fusion feature value F s is obtained by the following formula:

[0053]

[0054] F = F s,raw ; δ s is a feature-specific adjustment coefficient; F s,max , F s,min are the maximum and minimum values of the s-th fusion feature in the healthy population.

[0055] wherein δ s is set according to the close degree of the feature and the syndrome in the theory of traditional Chinese medicine, and the value range is 0.1 ≤ δ s ≤ 0.25, the higher the correlation degree, the greater the value, and vice versa.

[0056] When F s,raw < F s,min + δ s × (F s,max - F s,min ), F s is forcibly set to 0.

[0057] When F s,raw > F s,max - δ s × (F s,max - F s,min ), F s is forcibly set to 1.

[0058] Further, the camera is an RGB multispectral imaging camera, and the environmental light intensity control for image acquisition is 500-800 lux.

[0059] The TCM syndrome differentiation theory library is a preset database, the TCM syndrome differentiation theory library is built-in not less than 1000 latest standard syndrome characteristic data sets verified by clinical verification of a three-level Class A TCM institution, and the number of dimensions of core characteristic vectors is in the range of 12-20;

[0060] When the pulse condition data is preprocessed, a wavelet threshold denoising algorithm is used to complete the preprocessing, time domain characteristics include pulse wave conduction velocity, pulse width and pulse amplitude variation rate, and frequency domain characteristics include main frequency, secondary main frequency and main frequency band energy proportion.

[0061] On the other hand, an AI TCM four-examination instrument device, the device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, when the computer program is executed by the processor, the execution steps of a TCM four-examination comprehensive diagnosis method integrated with AI analysis are realized.

[0062] Compared with the known prior art, the technical scheme provided by the application has the following beneficial effects:

[0063] The application provides a TCM four-examination comprehensive diagnosis method integrated with AI analysis, in the execution process, a special diagnosis file is constructed by associating and binding a user identity and physiological parameters, the data pertinence and continuity are guaranteed, color card correction and accurate shooting parameters are combined to collect tongue surface, tongue bottom and face images, environmental light interference is eliminated and image definition is improved, a pulse diagnosis component is started to collect left and right hand pulse condition data, length verification, interference identification and elimination and signal-to-noise ratio secondary verification are carried out to ensure the effectiveness of the pulse diagnosis data, the collected data is input into an AI analysis model to generate multiple rounds of inquiry contents, the user manually inputs symptom keywords to supplement inquiry options, various data are fused in multiple dimensions and the weight is dynamically adjusted, the syndrome score is converted through a TCM syndrome element mapping matrix, a personalized health report containing multiple dimension information is generated based on the analysis result and the file is associated to support multiple mode acquisition, the TCM four-examination data collection is accurately realized, the analysis is intelligently realized and the diagnosis is personalized, the TCM four-examination efficiency and accuracy are effectively improved, and the user is provided with comprehensive and adaptive health evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical scheme in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or the prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creating any inventive labor.

[0065] Figure 1 It is a flowchart of a TCM four-examination comprehensive diagnosis method integrated with AI analysis;

[0066] Figure 2 The AI TCM four diagnostic instruments device appearance schematic diagram of the present application is shown in Figure 1.

[0067] Figure 3 The device operation interface schematic diagram of the present application is shown in Figure 2. Figure 1

[0068] Figure 4 The device operation interface schematic diagram of the present application is shown in Figure 3. Figure 2

[0069] Figure 5 The device operation interface schematic diagram of the present application is shown in Figure 4. Figure 3

[0070] Figure 6 The device image acquisition component appearance schematic diagram of the present application is shown in Figure 5.

[0071] Figure 7 The device operation interface schematic diagram of the present application is shown in Figure 6. Figure 4

[0072] The device operation interface schematic diagram of the present application is shown in Figure 7. Figure 8 Figure 5 The device operation interface schematic diagram of the present application is shown in Figure 8.

[0073] Figure 9 The device operation interface schematic diagram of the present application is shown in Figure 9. Figure 6 The device operation interface schematic diagram of the present application is shown in Figure 10.

[0074] Figure 10 The device operation interface schematic diagram of the present application is shown in Figure 11. Figure 7 The device operation interface schematic diagram of the present application is shown in Figure 12.

[0075] Figure 11 The AI TCM four diagnostic electronic health detection report example schematic diagram of the present application is shown in Figure 13.

[0076] The AI TCM four diagnostic paper health detection report example schematic diagram of the present application is shown in Figure 14. Figure 12 DETAILED DESCRIPTION In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0077] The present application will be further described below with reference to the embodiments.

[0078] The present application will be further described below with reference to the embodiments.

[0079] Embodiment:

[0080] ​​​The integrated AI analysis-based traditional Chinese medicine four-examination comprehensive diagnosis method of the embodiment, as shown in Figure 1 includes:

[0081] Obtain user identity information and at least basic physiological parameters including age, gender, height, and weight, associate and bind the two, and construct a user-specific diagnosis file;

[0082] The identity information is obtained through user certificate recognition or user self-entry;

[0083] The user identity information includes a user ID number and a biological feature identifier. The biological feature identifier is collected by an optical sensor built into the main device, and is stored after irreversible hash processing;

[0084] The basic physiological parameters also include body mass index (BMI), resting heart rate, preset historical period weight fluctuation value, and basic medical history;

[0085] The user-specific diagnosis file is stored in a hierarchical structure of basic information + dynamic data. The basic information layer stores the identity and fixed physiological parameters, i.e., age and gender. The dynamic data layer updates the variable physiological parameters in real time. The update trigger condition is that the physiological parameter change amplitude exceeds the preset value or the preset period is automatically updated;

[0086] The biological feature identifier includes a fingerprint or a face feature code, and the variable physiological parameters include weight, resting heart rate, and weight fluctuation value.

[0087] The user's tongue surface, tongue bottom, and face images are collected through a preset camera combined with a color card correction mechanism. The shooting angle and focusing parameters are automatically adjusted during the collection process, and the color card correction mechanism eliminates environmental light interference;

[0088] During the collection of the user's tongue surface, tongue bottom, and face images, the following is followed:

[0089] The camera resolution is ≥4K, the tongue surface shooting angle range is 30°-45°, the tongue bottom shooting angle range is 15°-25°, the face shooting angle is ≤±5°, the automatic focusing accuracy is ≤0.1mm, and the focusing response time is ≤0.3s;

[0090] The color card correction mechanism uses a 24-color standard color card, which includes 6 gray blocks and 18 color blocks. The correction process is as follows:

[0091] Synchronously collect the original pixel data of the color card area and the target image area, extract the average original pixel value of each color block in the color card, and record it as I raw_card (c);

[0092] Based on the standard reflectivity value calibrated by the color card factory, the environmental light influence coefficient

[0093] Color correction is performed on each pixel point of the target image to obtain a corrected image, and the correction formula is:

[0094]

[0095] In the formula, S(m) is the standard reflectivity value of the mth gray block; I raw_card (m) is the average original pixel value of the mth gray block; I corr (i,j,C) is the pixel value of pixel point (i,j) in the corrected image in the color channel C; S(c) is the standard reflectivity value of the cth color block calibrated when the 24-color standard color card is shipped; I raw (i,j,C) is the original pixel value of pixel point (i,j) in the target image in the color channel C;

[0096] wherein c represents the serial number of the 1st to 6th gray blocks and the 7th to 24th color blocks;

[0097] The above formula is based on the factory standard reflectivity of the gray blocks and the color blocks in the 24-color standard color card, combined with the average original pixel value of the collected color card area to calculate the ambient light influence coefficient, and then the coefficient is used to adjust each pixel of the target image (tongue surface, tongue bottom, facial image) in the corresponding color channel to eliminate the interference of ambient light on the image color, and to ensure that the collected image color is close to the real physiological state. This correction method combining the reflectivity of gray blocks and color blocks can accurately adapt to the correction needs of different color channels, avoid color deviation caused by single color card correction, and provide high-quality image data support for the accurate extraction of tongue image and facial features by the subsequent AI model;

[0098] The pulse diagnosis component is started, and a connection between the pulse diagnosis component and the host device is established through wireless communication. The pulse image data of the left and right hands of the user are collected in sequence. After the collection is completed, the host device performs length checking on the pulse image data. When the data length reaches a preset threshold, the pulse image data is determined as valid pulse diagnosis data.

[0099] The collected tongue image, facial image and valid pulse diagnosis data are input into an AI analysis model. The AI analysis model generates multi-round inquiry contents for the user based on the theory of traditional Chinese medicine syndrome differentiation.

[0100] The AI analysis model architecture is: syndrome feature extraction-inquiry logic generation-feedback adaptation.

[0101] The syndrome feature extraction layer maps the features of the tongue image, facial image and valid pulse diagnosis data, and outputs the core feature vector of the traditional Chinese medicine syndrome differentiation dimension.

[0102] It should be noted that:

[0103] The syndrome feature extraction layer uses a multi-modal feature alignment mechanism for feature mapping of the tongue image, facial image and valid pulse diagnosis data:

[0104] Firstly, the tongue image is preprocessed (including tongue surface area segmentation, color correction, and light normalization), and 12-dimensional visual features such as tongue fur thickness (pixel ratio), tongue color (HSV channel mean), and dampness state (gray standard deviation) are extracted through a convolutional neural network and mapped to the "tongue diagnosis syndrome dimension" (such as tongue color corresponding to cold and heat, and tongue fur thickness corresponding to deficiency and excess) in traditional Chinese medicine;

[0105] Key point detection (positioning eyes, nose, lips, and facial color area) is performed on the collected facial image, and 8-dimensional features such as facial color gloss (Lab* space L value), eyelid color (whether pale or purple), and lip color saturation are extracted and mapped to the "facial color diagnosis dimension" (such as red face corresponding to heat syndrome and pale face corresponding to deficiency syndrome);

[0106] Effective pulse diagnosis data (after denoising and cycle alignment) are extracted to obtain 6-dimensional time and frequency domain features such as pulse frequency, amplitude, and waveform distortion rate, which are mapped to the "pulse diagnosis dimension" (such as pulse rate corresponding to tardiness and amplitude corresponding to deficiency and excess);

[0107] Finally, the three types of features are converted into 13-dimensional core feature vectors corresponding to the eight syndrome differentiation (8-dimensional) and the five zang-fu differentiation (5-dimensional) through the traditional Chinese medicine syndrome differentiation rule library (containing 32 expert consensus mapping relationships), where each dimension value is standardized to [0, 1] (0 indicates that the syndrome differentiation dimension feature is not significant, and 1 indicates that the feature is significant), realizing the structured mapping of cross-modal data to the traditional Chinese medicine syndrome differentiation system.

[0108] The inquiry logic generation layer dynamically generates multiple rounds of inquiry content based on the matching degree of the core feature vector and the traditional Chinese medicine syndrome differentiation theory library, and the calculation of the matching degree and the inquiry parameters is subject to:

[0109]

[0110] wherein: M is the matching degree of the core feature vector and the standard feature vector of the syndrome differentiation theory library, M ∈ [0, 1]; n is the number of dimensions of the core feature vector; w k is the syndrome differentiation weight of the kth core feature, w k ∈ [0.5, 2]; F k is the actual extraction value of the kth feature in the core feature vector; T k is the standard threshold value range median of the corresponding kth feature in the syndrome differentiation theory library; Q is the information depth coefficient of the current round of inquiry content; Q0 is the basic information depth coefficient when the matching degree is zero; Q1 is the highest information depth coefficient when the matching degree is 1;

[0111] wherein, w kThe greater the value corresponding to the core feature that plays a decisive role in the differentiation conclusion, the smaller the value corresponding to the core feature that plays a minor role in the differentiation conclusion. Q0 and Q1 are preset values, and the values are 1 and 5 respectively. The inquiry parameter is denoted by Q.

[0112] The above formula allocates corresponding weights according to the importance of each core feature in the differentiation conclusion in the TCM differentiation theory. The matching degree M of the core feature vector and the differentiation theory library is calculated by calculating the ratio of the actual extraction value of each core feature to the median value of the theoretical standard threshold, combined with the number of core feature dimensions. Then, the information depth coefficient Q is linearly adjusted according to the matching degree. When the matching degree is low, the basic information depth is used to ensure comprehensive inquiry. When the matching degree is high, the information depth is improved to dig out key differentiation information. This design of linking feature weight with differentiation importance and self-adapting inquiry depth with matching degree can reduce the interference of irrelevant features in matching degree calculation, avoid redundant inquiry, accurately obtain key information required by AI differentiation, and improve the efficiency and pertinence of inquiry.

[0113] It should be noted that T k is the median value of the standard threshold range of TCM features. Since all TCM four-examination features (tongue appearance, pulse appearance, facial features, etc.) are based on physical measurement or quantitative scoring, T k > 0; if the system is abnormal and T k is 0, the feature is automatically excluded from the matching degree calculation; the design of matching degree M ensures that M ∈ [0, 1]. When the calculation result M < 0, the system automatically sets M to 0, indicating that the matching degree is very low; when M > 1, the system automatically sets M to 1, indicating complete matching;

[0114] Collecting user multi-round inquiry feedback information, multi-dimensional fusion of tongue appearance, facial feature data, effective pulse diagnosis data and inquiry feedback information, AI analysis model cross-verification and logical reasoning on the fusion data to determine the user's TCM constitution type and specific syndrome, and simultaneously performing comprehensive analysis of the fusion information data;

[0115] During the inquiry stage, the user can manually input keywords not covered by the symptoms, and the host device automatically matches and supplements the corresponding inquiry options based on the input keywords;

[0116] The processing of the host device on the effective pulse diagnosis data also includes preprocessing and feature quantization steps:

[0117] Feature quantization is to convert the extracted time domain and frequency domain features into pulse appearance quantitative values:

[0118]

[0119] In the formula: F is the floating and sinking amount; A max , A min are the maximum and minimum amplitudes of the denoised pulse appearance signal; Aref is a preset standard pulse amplitude reference value; C' is the number of delays; T avg is the average period of the de-noised continuous 5 pulse periods; S is the amount of slip; E main is the frequency band energy corresponding to the main frequency; E sub is the frequency band energy corresponding to the secondary main frequency; E total is the total energy of the entire pulse signal in the 0-20Hz frequency domain;

[0120] The above formula quantifies the floating and sinking characteristics by the ratio of the maximum and minimum amplitude difference of the de-noised pulse signal to the standard amplitude reference value, reflects the number of delays with the average period of the continuous 5 pulse periods, and embodies the slip characteristics with the ratio of the main frequency band energy to the secondary main frequency band energy. The traditional pulse of the qualitative description is converted into an objective and calculable numerical value. This way of quantitatively converting abstract pulse indicators solves the problem of strong subjectivity and difficulty in standardization of manual pulse diagnosis, and enables the pulse diagnosis data to be effectively integrated with other dimensional data such as tongue and interrogation, providing objective and quantitative pulse diagnosis basis for cross-validation and logical reasoning of AI models;

[0121] When the pulse diagnosis component collects pulse data of the user's left and right hands, it synchronously performs signal interference recognition and rejection processing on the pulse data:

[0122] Collecting three-axis acceleration data A x , A y , A z of the pulse diagnosis component, and synchronously calculating the interference intensity score:

[0123] A0 represents a preset acceleration reference value;

[0124] Set the interference judgment threshold [a, b], a < b. When I < a, it is judged as no significant interference, and the pulse data is normally collected. When a≤I < b, it is judged as slight interference, and the preset vibration feedback unit on the pulse diagnosis component is started to prompt the user to keep the wrist stable through 50Hz low-frequency micro-vibration, and the collected pulse signal is synchronously processed by adaptive filtering. When I≥b, it is judged as serious interference, and the pulse collection is immediately suspended. The user is prompted by the main device to adjust the wrist posture, and after I < a, the collection is restarted;

[0125] For the pulse signal after adaptive filtering, the signal-to-noise ratio SNR is calculated for secondary verification:

[0126] E s represents the main frequency band energy value of the pulse signal, i.e. 0.5-3Hz, E n represents the energy value of the noise frequency band in the signal, i.e. >10Hz;

[0127] When SNR≥15dB, it is determined that the filtered signal is valid, and the pulse diagnosis valid data is included;

[0128] When SNR<15dB, it is determined that the filtered signal is still invalid, the re-acquisition process is triggered, and the maximum re-acquisition is set to 2 times;

[0129] If the SNR≥15dB is still not met after 2 times of re-acquisition, the text information of pulse diagnosis acquisition exception is recorded to the user diagnosis archive, and the initial weight of tongue appearance and inquiry information is increased by 0.1 during the inference of the subsequent AI analysis model;

[0130] The above formula calculates the interference intensity score by collecting the three-axis acceleration data of the pulse diagnosis component and the preset acceleration reference value, and according to whether the score is in the preset threshold [a, b], the measures such as no processing, vibration prompt + adaptive filtering, pause collection + voice prompt are taken, and then the ratio of the energy of the main frequency band (0.5-3Hz) to the energy of the noise frequency band (>10Hz) of the filtered pulse signal is calculated, and the signal validity standard is SNR≥15dB, and the maximum re-acquisition limit is set to 2 times, and the weight of tongue appearance and inquiry information is increased when re-acquisition fails. This hierarchical processing mechanism combining three-axis acceleration interference recognition and signal-to-noise ratio secondary verification can effectively filter different degrees of interference, reduce invalid pulse diagnosis data, and through weight adjustment to cope with acquisition exceptions, ensuring the quality of pulse diagnosis data and the accuracy of subsequent AI analysis;

[0131] Based on the body constitution type, syndrome type and comprehensive analysis results output by the AI analysis model, an electronic health report is generated, and the report is associated with the user's exclusive diagnosis archive, so that the report supports electronic retrieval, two-dimensional code viewing and paper printing acquisition;

[0132] It should be noted that:

[0133] When generating an electronic health report, first, based on the output results of the AI analysis model, the results are derived from the comprehensive processing of multi-dimensional data:

[0134] First, the user's tongue surface (inclination angle 30°-45°), tongue bottom (inclination angle 15°-25°) and face (inclination angle ≤±5°) images are collected and processed through a preset camera combined with a 24-color standard color card correction mechanism (including 6 gray blocks and 18 color blocks, which eliminates light interference through environmental light influence coefficient calculation and pixel correction), and the pulse diagnosis component is connected to the main device through wireless communication, and the user's left and right hand pulse data is collected in turn, and the valid pulse diagnosis data is determined through length verification, interference identification elimination (through three-axis acceleration to calculate interference intensity score and hierarchical processing) and signal-to-noise ratio (SNR≥15dB) secondary verification;

[0135] The tongue image, facial image, and effective pulse diagnosis data are then input into the AI ​​analysis model. Based on the TCM syndrome differentiation theory database (which includes a clinical validation dataset of no less than 1,000 cases from top-tier TCM institutions), the model generates multi-round consultation content. After collecting user feedback, it performs multi-dimensional fusion of tongue image features, facial features, effective pulse diagnosis data, and consultation information (initial weights are 0.3, 0.2, 0.3, and 0.2, respectively, and then dynamically adjusts the weights according to the reliability index). The fused data is converted into syndrome scores through a TCM syndrome element mapping matrix. After cross-validation and logical reasoning of the fused data, the AI ​​analysis model determines the user's TCM constitution type and specific syndrome type. Finally, based on the constitution type, syndrome type, and comprehensive analysis results output by the AI ​​model, an electronic health report containing multi-dimensional information such as user basic information, constitution analysis, syndrome type judgment, and health assessment is generated simultaneously.

[0136] Multi-dimensional fusion follows the following rules:

[0137] The initial weights for each dimension of data are set as follows: tongue features 0.3, facial features 0.2, pulse diagnosis data 0.3, and consultation information 0.2.

[0138] The formula for dynamic weight adjustment is:

[0139] In the formula: W i ′ represents the adjusted weight; W i β is the original weight; β is the adjustment coefficient; R i The reliability index of the i-th dimension data; σ(R) is the average of the reliability indices for each dimension; σ(R) is the global standard deviation of the reliability indices for all dimensions; ε is the minimum value, which is defined by the system user and is initially set to 0.05.

[0140] The fused data is converted into syndrome scores through a pre-defined TCM syndrome element mapping matrix, where matrix element M... s,e The symptom score S represents the correlation strength between the s-th fusion feature and the e-th symptom element. e =M s,e ×F s ;

[0141] In the formula: F s These are the standardized fusion feature values;

[0142] When S e A value greater than 0.6 indicates that the syndrome element is established;

[0143] Wherein, the adjustment coefficient β∈[0.2,0.6], the value is larger when the difference in the reliability index of each dimension is greater, and the value is smaller when the difference in the reliability index of each dimension is smaller. iThe difference between the proportion of effective features and the proportion of noise signals in the dimension data is calculated, and the normalized value is obtained;

[0144] In the above formula, the initial weight of each dimension data of tongue appearance, face, pulse diagnosis, and inquiry is first set, and then the difference between the reliability index of each dimension data and the average reliability index is combined with the adjustment coefficient β changing with the difference of reliability index to dynamically adjust the weight of each dimension, so that the data with high reliability has a higher proportion in the fusion. Subsequently, the standardized fusion feature value is multiplied by the feature-syndrome association strength through the TCM syndrome element mapping matrix to obtain the syndrome score. The establishment of the syndrome element is determined by the score > 0.6. This dynamic weight adjustment design based on the reliability index avoids the problem of misleading the syndrome differentiation result caused by low reliability data under fixed weight. At the same time, the mapping matrix is used to quantify the association between features and syndromes, and the multi-source heterogeneous data is converted into an intuitive syndrome score, which improves the logicality and accuracy of AI differentiation;

[0145] It should be noted that:

[0146] The effective feature determination standard for reliability index calculation is that the absolute value of the deviation of the feature detection value from the standard threshold range is ≤15%. The TCM syndrome element mapping matrix is constructed based on ≥2000 cases of confirmed cases in a tertiary A-level TCM institution, and the association strength of the matrix elements is tested by Pearson correlation coefficient;

[0147] The standardized fusion feature value F s is calculated by the following formula:

[0148]

[0149] In the formula, F s,raw is the original detection value of the s-th fusion feature; δ s is the feature-specific adjustment coefficient; F s,max , F s,min are the maximum and minimum values of the s-th fusion feature in the healthy population;

[0150] where δ s is set according to the closeness of the feature to the syndrome according to the TCM theory, and the value range is 0.1≤δ s ≤0.25. The higher the correlation degree, the larger the value, and vice versa;

[0151] When F s,raw <F s,min + δ s × (F s,max -F s,min ), F s is forced to be 0;

[0152] When F s,raw >F s,max- delta s x(F s,max - F s,min ), F s is forced to be 1;

[0153] The above formula sets a specific adjustment coefficient according to the close degree of the characteristics and syndromes in the theory of traditional Chinese medicine, and the characteristic value is standardized by the difference ratio of the original detection value of the characteristic and the maximum and minimum values of the characteristic in the healthy population, combined with the specific adjustment coefficient. At the same time, the original value lower than the minimum value of the health is forced to be 0, and the value higher than the maximum value of the health is forced to be 1, to ensure that the standardized value meets the characteristic range of traditional Chinese medicine syndrome differentiation. This standardized method with specific adjustment coefficient makes the characteristics closely related to syndromes more prominent after standardization, avoids the interference of abnormal values, and makes the standardized characteristic value more consistent with the logic of traditional Chinese medicine clinical syndrome differentiation, providing a precise quantitative basis for subsequent syndrome score calculation;

[0154] The camera is an RGB multispectral imaging camera, and the environmental light intensity control for image acquisition is 500-800 lux;

[0155] The TCM syndrome differentiation theory library is a preset database, and the TCM syndrome differentiation theory library is built-in with not less than 1000 latest standard syndrome characteristic data sets verified by clinical verification of a three-level Class A TCM institution, and the number of dimensions of the core characteristic vector is in the range of 12-20;

[0156] During the preprocessing of pulse data, a wavelet threshold denoising algorithm is used to complete the preprocessing, the time domain characteristics include pulse wave conduction velocity, pulse width and pulse amplitude variation rate, and the frequency domain characteristics include main frequency, secondary main frequency and main frequency band energy proportion;

[0157] An AI TCM four-examination instrument device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, the execution steps of a TCM four-examination comprehensive diagnosis method integrated with AI analysis are realized. As shown in Figure 2 , Figure 2 is a schematic view of the appearance of the AI TCM four-examination instrument device of the present application.

[0158] In the embodiment, the above-mentioned method can quickly establish a user-specific diagnosis file, accurately collect tongue image, facial image and pulse data and eliminate interference, dynamically generate inquiry content through an AI model, accurately determine constitution and syndrome type after multi-dimensional fusion of data, and generate a personalized health report. It not only improves the efficiency and accuracy of TCM diagnosis, reduces human error, but also enables users to conveniently obtain reports, helps them to understand their health status and avoid disease risks, and provides a scientific basis for subsequent conditioning, promoting the standardization and intelligent development of TCM diagnosis.

[0159] The following will describe the method in the above embodiment in combination withFigures 3-12 , an example of an application of this method:

[0160] A community health service station of traditional Chinese medicine introduces an integrated AI analysis of traditional Chinese medicine comprehensive diagnosis system, and provides health assessment services for Mr. Zhang, a 38-year-old male user. The specific diagnosis process and results are as follows:

[0161] I. Construction of user-specific diagnosis archives

[0162] The system first starts the identity identification and basic information collection module: through the identity card recognition device, the identity card number of Mr. Zhang is obtained, and the built-in optical sensor of the main device is called to collect his fingerprint information. After the collection is completed, the fingerprint data is irreversibly hashed to protect privacy security; then guide Mr. Zhang to enter the basic physiological parameters, including age 38, male, height 175 cm, weight 70 kg (calculated BMI is 22.8, which is in the healthy range), resting heart rate 68 times per minute, body weight fluctuation in the past 6 months 0.8 kg (not exceeding the preset fluctuation threshold), and Mr. Zhang also stated that "occasionally there is stomach distension for the past 2 months, and there is no history of chronic diseases". The system associates and binds the above identity identification information and basic physiological parameters, and stores them in a hierarchical structure of "basic information layer + dynamic data layer": the basic information layer fixedly stores the identity card number, hashed fingerprint, age, gender and other unchanging information; the dynamic data layer updates the variable parameters such as weight, resting heart rate and body weight fluctuation value in real time, and finally constructs the exclusive diagnosis archives of Mr. Zhang.

[0163] II. Tongue image and facial image collection and correction

[0164] The system calls the preset 4K resolution RGB multispectral imaging camera, first adjusts the light intensity of the collection environment to 700 lux (which meets the standard range of 500-800 lux), and then automatically adjusts the shooting angle and focusing parameters according to the preset parameters: when collecting the tongue surface image, the camera is set to 35°, the automatic focusing accuracy is 0.07 mm, and the focusing response time is 0.25 s; when collecting the tongue bottom image, the angle is set to 22°, the focusing accuracy and response time are consistent with the tongue surface collection; when collecting the facial image, the shooting offset angle is controlled at-3° to ensure that the facial contour, skin color and facial details are complete.

[0165] Image acquisition synchronization start 24 color standard color card correction mechanism (the color card contains 6 gray blocks, 18 color blocks): the system first simultaneously acquires the original pixel data of the color card area and Mr. Zhang's tongue surface, tongue bottom and face, extracts the average original pixel value of each color block in the color card, and then calculates the influence coefficient of the current environmental light by combining the standard reflectivity value of each color block calibrated when the color card is factory-calibrated; subsequently, the original image of Mr. Zhang's tongue and face is color corrected pixel by pixel based on the coefficient. After correction, the problems of "tongue color darkening" and "distortion of facial skin color" caused by environmental light are completely eliminated, and the characteristics of Mr. Zhang's light red tongue surface, slight tooth marks on the tongue edge, and white and greasy moss can be clearly observed, and the facial skin color is uniform without obvious color spots or yellowing.

[0166] III. Pulse diagnosis data acquisition and effectiveness determination

[0167] The staff assists Mr. Zhang to wear the pulse diagnosis assembly on the wrist, and the system automatically establishes a wireless connection between the pulse diagnosis assembly and the main device through Bluetooth, and then acquires the pulse data of Mr. Zhang's left and right hands in turn. During the acquisition process, the system monitors the three-axis acceleration data of the pulse diagnosis assembly in real time and calculates the interference intensity score: when acquiring the pulse data of the left hand for the first time, because Mr. Zhang's wrist moves unconsciously, the interference intensity score is in the preset "slight interference" interval (i.e. the score is between a and b, a < b), the system immediately starts the vibration feedback unit on the pulse diagnosis assembly, prompts Mr. Zhang to keep the wrist stable through 50Hz low-frequency micro-vibration, and simultaneously performs adaptive filtering processing on the acquired pulse signal.

[0168] After filtering, the system performs double effectiveness verification on the signal: one is to calculate the signal-to-noise ratio (SNR) by extracting the energy value of the 0.5-3Hz main frequency band and the energy value of the >10Hz noise frequency band in the pulse signal, and the SNR is calculated to be 20dB (≥15dB effective threshold); two is to check the data length and confirm that the length of the acquired pulse data reaches the preset threshold. After double verification, the pulse data is determined as valid pulse diagnosis data.

[0169] Subsequently, the system pre-processes the valid pulse diagnosis data: removes the environmental noise and human micro-motion noise in the signal by using the wavelet threshold denoising algorithm, extracts the time domain features (pulse wave transmission speed 0.9ms, pulse width 0.28s, pulse amplitude variation rate 4.5%) and frequency domain features (main frequency 1.3Hz, secondary main frequency 0.6Hz, main frequency band energy proportion 68%), and completes feature quantization: the floating and sinking amount is calculated to be 0.5 by the maximum amplitude and the minimum amplitude of the denoised pulse signal and the preset standard pulse amplitude reference value, the delay amount is calculated to be 0.7 by the average period of the continuous 5 pulse periods, and the slippery amount is calculated to be 0.6 by the main frequency corresponding to the frequency band energy, the secondary main frequency corresponding to the frequency band energy, and the total energy in the 0-20Hz frequency domain range of the entire pulse signal.

[0170] IV. AI Model Generation Inquiry and Feedback Collection

[0171] The system inputs the corrected tongue image, facial image, and quantified effective pulse diagnosis data into the AI analysis model: the model first maps these data through the syndrome feature extraction layer, outputting a core feature vector containing 16 dimensions (covering tongue color, tongue shape, moss quality, facial skin color, pulse diagnosis floating sinking delay number and slippery etc. key features); then enters the inquiry logic generation layer, combining the TCM syndrome differentiation theory library (built-in 1100 cases of standard syndrome feature data set verified by clinical data in third-grade TCM hospitals) to calculate the matching degree - the features that determine the syndrome conclusion (such as tongue edge tooth marks, pulse diagnosis features associated with stomach distention) are set to 1.7, and the features that have a minor impact on the syndrome conclusion (such as facial subtle luster differences) are set to 0.7, combined with the actual extraction value of each core feature and the standard threshold range of the corresponding feature in the theory library, the final matching degree is 0.65.

[0172] According to the information depth coefficient calculation formula (the basic information depth coefficient Q0=1, the highest information depth coefficient Q1=5 when the matching degree is 1), the current round of inquiry information depth coefficient Q is calculated as 3.8, and the multi-round inquiry content is generated accordingly: the first round of inquiry "stomach distention occurrence time (such as no fixed time before or after meals) and duration", Mr. Zhang feedback "usually appears about 1 hour after meals, lasts about 30 minutes"; the second round of inquiry "whether distention is accompanied by acid reflux, belching, loss of appetite", Mr. Zhang feedback "occasionally belching, no obvious change in appetite"; at the same time, Mr. Zhang manually inputs the symptom keyword "occasional diarrhea after meals" not covered, the system automatically retrieves and supplements the inquiry "diarrhea frequency (such as 1-2 times 3-4 times per week) and stool consistency (such as loose and unformed water-like stool)", Mr. Zhang feedback "2-3 times per week, stool is not formed".

[0173] V. Multi-dimensional data fusion and syndrome differentiation analysis

[0174] After collecting the inquiry feedback information, the system starts the multi-dimensional data fusion module: first, set the initial weight of each dimension - tongue feature 0.3, facial feature 0.2, pulse diagnosis data 0.3, inquiry information 0.2; then calculate the reliability index of each dimension: tongue image data effective feature proportion 93%, noise proportion 7%, reliability index 0.91; facial feature data effective feature proportion 89%, noise proportion 11%, reliability index 0.82; pulse diagnosis data effective feature proportion 91%, noise proportion 9%, reliability index 0.87; inquiry information because Mr. Zhang's expression is clear and has no contradictions, effective feature proportion 96%, noise proportion 4%, reliability index 0.93.

[0175] The average of each dimension reliability index is 0.8825, the maximum is 0.93, and the minimum is 0.82. Since the difference is small, the adjustment coefficient β is taken as 0.25, and the adjusted weight is calculated according to the weight dynamic adjustment formula: tongue features 0.305, facial features 0.195, pulse diagnosis data 0.3, and inquiry information 0.2. The fused data is converted into syndrome scores through the TCM syndrome element mapping matrix (such as "stomach distension + postprandial onset + belching + loose stools" and "spleen deficiency and dampness" syndrome correlation strength 0.8, "tongue edge tooth marks" and "spleen deficiency" correlation strength 0.75). For the "stomach distension + postprandial onset" feature (closely related to the syndrome), set the specificity adjustment coefficient δ = 0.24, and the standardized fusion feature value is 0.78. For the "tongue edge tooth marks" feature, set δ = 0.22, and the standardized fusion feature value is 0.75. Finally, the "spleen deficiency and dampness" syndrome score is 0.73 (> 0.6, determined to be established), and the "mild stomach failure and descent" syndrome score is 0.64 (> 0.6, determined to be established).

[0176] The AI model cross-validated and logically inferred the fused data, combined with the core feature vector and the syndrome score, and finally determined that Mr. Zhang's TCM constitution type was "phlegm-dampness tendency", and the specific syndrome type was "spleen deficiency and dampness with mild stomach failure and descent". At the same time, the functions of the viscera were comprehensively analyzed: the spleen transportation and transformation function was weakened (corresponding to loose stools and tongue edge tooth marks), the stomach descending function was slightly poor (corresponding to postprandial distension, belching), and the functions of the remaining viscera were basically normal.

[0177] Six, electronic health report generation and storage

[0178] The system generates Mr. Zhang's electronic health report based on the AI model output results. The report is automatically associated with Mr. Zhang's exclusive diagnosis file. Mr. Zhang can directly electronically retrieve the report through the main device of the service station, or scan the exclusive two-dimensional code generated by the system on the mobile phone to view it in real time. The staff also prints a paper report for Mr. Zhang, making it convenient for him to refer to at home.

[0179] To sum up, in the implementation process of the method in the above embodiments, the exclusive diagnosis archives are constructed by associating the user identity with the physiological parameters, ensuring the pertinence and continuity of the data, combining the color card correction and the accurate shooting parameter to collect the tongue surface, tongue bottom and face images, eliminating the environmental light interference and improving the image definition, starting the pulse diagnosis component to collect the left and right hand pulse data and ensuring the effectiveness of the pulse diagnosis data through the length verification, interference identification and elimination and the signal-to-noise ratio secondary verification, inputting the collected data into the AI analysis model to generate the multi-round inquiry contents, supporting the user to manually input the symptom keywords to supplement the inquiry options, multi-dimensionally fusing various data and dynamically adjusting the weight, converting the syndrome score through the Chinese medicine syndrome element mapping matrix, generating the personalized health report containing multi-dimension information based on the analysis result and associating the archives to support the multi-mode acquisition, and overall realizing the accuracy of the Chinese medicine four diagnosis data collection, the intelligence of the analysis and the individualization of the diagnosis, effectively improving the Chinese medicine four diagnosis efficiency and accuracy, and providing the user with the comprehensive and adaptive health evaluation.

[0180] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for integrated AI analysis of comprehensive diagnosis of four examinations of traditional Chinese medicine, characterized in that, The application comprises the following steps: acquiring user identity information and basic physiological parameters including age, gender, height, and weight, associating and binding the two to build a user-specific diagnostic profile; collecting tongue surface, tongue bottom, and facial images through a preset camera combined with a color card correction mechanism, automatically adjusting the shooting angle and focusing parameters during the collection process, and eliminating environmental light interference through the color card correction mechanism; starting the pulse diagnosis component, synchronously establishing its connection with the host device through wireless communication, sequentially collecting left and right hand pulse data, and performing length verification on the pulse data after collection is completed, determining that the pulse data is valid when the data length reaches a preset threshold; inputting the collected tongue image, facial image, and valid pulse diagnosis data into an AI analysis model, which generates multiple rounds of inquiry content for the user based on traditional Chinese medicine syndrome differentiation theory; collecting user feedback information from multiple rounds of inquiry, performing multi-dimensional fusion on tongue image, facial feature data, valid pulse diagnosis data, and inquiry feedback information, cross-verification and logical reasoning of the fusion data by the AI analysis model to determine the user's traditional Chinese medicine constitution type and specific syndrome type, and simultaneously performing comprehensive analysis of the fusion information data; based on the constitution type, syndrome type, and comprehensive analysis results output by the AI analysis model, synchronously generating an electronic health report, and associating the report to the user-specific diagnostic profile to enable electronic retrieval, two-dimensional code viewing, and paper printing of the report; the camera resolution is ≥4K, the tongue surface shooting angle range is 30°-45°, the tongue bottom shooting angle range is 15°-25°, the facial shooting angle is ≤±5°, the automatic focusing accuracy is ≤0.1mm, and the focusing response time is ≤0.3s; the color card correction mechanism uses a 24-color standard color card, which includes 6 gray blocks and 18 color blocks, and the correction process is as follows: color correction is performed on each pixel point of the target image to obtain the corrected image, and the correction formula is as follows: Synchronously acquiring the original pixel data of the color card region and the target image region, extracting the average original pixel value of each color block in the color card, denoted as I raw_card (c); Based on the standard reflectance value of the color card factory calibration, the ambient light influence coefficient is calculated where c represents the serial number of the 1st to 6th gray blocks and the 7th to 24th color blocks. where: S(m) is the standard reflectance value of the mth gray scale patch; I raw_card (m) is the average raw pixel value of the mth gray scale patch; I corr (i,j,C) is the pixel value of pixel point (i,j) in the corrected image in color channel C; S(c) is the standard reflectance value of the cth color block of the 24-color standard color card calibrated when the standard color card is manufactured; I raw (i,j,C) is the original pixel value of pixel point (i,j) in the target image in color channel C; The identity information is obtained through user identification or user self-input; 2. The integrated AI analysis-based comprehensive diagnosis method of four traditional Chinese medical examinations according to claim 1, characterized in that, The user identity information includes the user's ID number and biological feature identification, which is collected by the optical sensor built-in the host device and stored after irreversible hash processing; The basic physiological parameters also include body mass index (BMI), resting heart rate, preset historical period weight fluctuation value, and basic medical history; The user-specific diagnostic profile is stored in a hierarchical structure of basic information + dynamic data, the basic information layer stores the identity and fixed physiological parameters, i.e. age and gender, and the dynamic data layer updates the variable physiological parameters in real time, with the update trigger condition being a physiological parameter change amplitude exceeding a preset value or a preset period automatic update; The biological feature identification includes fingerprint or face feature code, and the variable physiological parameters include weight, resting heart rate, and weight fluctuation value. The AI analysis model architecture is: syndrome differentiation feature extraction-inquiry logic generation-feedback adaptation; 3. The integrated AI analysis-based comprehensive diagnosis method of four traditional Chinese medical examinations according to claim 1, characterized in that, The syndrome differentiation feature extraction layer maps the tongue image, facial image, and valid pulse diagnosis data to output the core feature vector of the traditional Chinese medicine syndrome differentiation dimension; ​ The inquiry logic generation layer dynamically generates multiple rounds of inquiry content based on the matching degree of the core feature vector and the TCM syndrome differentiation theory library, and the calculation of the matching degree and inquiry parameters is subject to: In the formula, M is the matching degree of the core feature vector and the standard feature vector of the syndrome differentiation theory library, M ∈ [0.1]; n is the dimension number of the core feature vector; w k is the syndrome differentiation weight of the kth core feature, w k ∈ [0.5, 2]; F k is the actual extraction value of the kth feature in the core feature vector; T k is the standard threshold range median corresponding to the kth feature in the syndrome differentiation theory library; Q is the information depth coefficient of the current round of inquiry content; Q0 is the basic information depth coefficient when the matching degree is zero; Q1 is the highest information depth coefficient when the matching degree is 1. wherein w k The greater the value corresponding to the core feature that plays a decisive role in the syndrome differentiation conclusion, the smaller the value corresponding to the core feature that plays a minor role in the syndrome differentiation conclusion. Q0 and Q1 are preset values, and the values are 1 and 5, respectively. The inquiry parameters are denoted by Q.

4. The integrated AI analysis-based comprehensive diagnosis method of four traditional Chinese medical examinations according to claim 1, characterized in that, During the inquiry stage, the user can manually input keywords of symptoms not covered, and the host device automatically matches and supplements corresponding inquiry options based on the input keywords; The host device's processing of valid pulse diagnosis data also includes preprocessing and feature quantization steps: Feature quantization involves converting the extracted time domain and frequency domain features into pulse condition quantization values: In the formula, F is the float-and-sink amount; A max , A min is the maximum and minimum amplitudes of the denoised pulse signal; A ref is a preset standard pulse amplitude reference value; C' is the late number; T avg is the average period of the de-noised 5 consecutive pulse periods; S is the slippery amount; E main is the frequency band energy corresponding to the main frequency; E sub is the frequency band energy corresponding to the secondary main frequency; E total is the total energy in the 0-20Hz frequency domain range of the entire pulse signal.

5. The integrated AI analysis-based comprehensive diagnosis method of four traditional Chinese medical examinations according to claim 1, characterized in that, When the pulse diagnosis component collects pulse data from the user's left and right hands, it simultaneously performs signal interference recognition and elimination processing on the pulse data: Collecting three-axis acceleration data A of the pulse diagnosis assembly x , A y , A z , synchronously calculating the interference intensity score: A0 represents a preset acceleration reference value; Set the interference judgment threshold [a, b], a < b. When I < a, it is determined that there is no significant interference, and the pulse data is collected normally. When a ≤ I < b, it is determined that there is slight interference, and the preset vibration feedback unit on the pulse diagnosis component is started to prompt the user to keep their wrist stable through 50Hz low-frequency micro-vibration, and the collected pulse signal is simultaneously subjected to adaptive filtering processing. When I ≥ b, it is determined that there is serious interference, and the pulse collection is immediately paused, and the user is prompted to adjust their wrist posture through the host device's voice. After I < a, the collection is restarted. For the pulse signal after adaptive filtering, a secondary check is performed by calculating the signal-to-noise ratio (SNR): E s represents the energy value of the main frequency band of pulse signal, i.e. 0.5-3 Hz, E n represents the energy value of the noise frequency band in the signal, i.e. >10 Hz; When SNR ≥ 15dB, it is determined that the filtered signal is valid and is included in the pulse diagnosis valid data. When SNR < 15dB, it is determined that the filtered signal is still invalid, triggering a re-collection process with a maximum of 2 re-collections. If after 2 re-collections, SNR ≥ 15dB is still not met, record the text information of the pulse diagnosis collection anomaly in the user's diagnosis file, and during subsequent AI analysis model reasoning, increase the initial weight of tongue and inquiry information by 0.

1.

6. The integrated AI analysis-based comprehensive diagnosis method of four traditional Chinese medical examinations according to claim 1, characterized in that, The multi-dimensional fusion is subject to: The initial weights of each dimension of data are set as follows: tongue features 0.3, facial features 0.2, pulse diagnosis data 0.3, and inquiry information 0.

2. The weight dynamic adjustment formula is wherein: W i is the adjusted weight; W i is the original weight; β is an adjustment factor; R i is the reliability index of the i-th dimension data; is the average of the reliability index of each dimension; σ(R) is the global standard deviation of all the reliability indexes; ε is a minimum value, which is defined by the system end user and initially set as 0.05; The fused data is converted into syndrome score through a preset TCM syndrome element mapping matrix, and the matrix element M s,e represents the association strength between the s th fused feature and the e th syndrome element, and the syndrome score S e = M s,e × F s ; In the formula, F s is the standardized fusion feature value; When S e >0.6, it is determined that the syndrome element is established. wherein the adjustment coefficient β ∈ [0.2, 0.6], the greater the difference between the reliability indexes of the dimensions, the greater the value, and the smaller the difference between the reliability indexes of the dimensions, the smaller the value of the reliability index R of the i-th dimension data i The difference between the proportion of effective features and the proportion of noise signals in the dimension data is calculated, and is obtained after normalization processing.

7. The integrated AI analysis-based comprehensive diagnosis method of four traditional Chinese medical examinations according to claim 6, characterized in that, the standardized fusion feature value F s is calculated by the following equation: wherein: F s,raw is the raw detection value of the s-th fusion feature; δ s is the feature-specific adjustment coefficient; F s,max , F s,min is the maximum and minimum value of the s-th fusion feature in the healthy population; wherein, δ s According to the close degree of the feature and the syndrome in the theory of traditional Chinese medicine, the value range is 0.1≤δ s ≤0.25, the higher the correlation degree is, the larger the value is, and vice versa. When F s,raw <F s,min + δ s × (F s,max - F s,min ), F s is forced to 0; When F s,raw > F s,max - δ s × (F s,max - F s,min ), F s is forced to 1.

8. The integrated AI analysis-based comprehensive diagnosis method of the four examinations of traditional Chinese medicine according to claim 1 or 3 or 4, characterized in that, The camera is an RGB multispectral imaging camera, and the environmental light intensity for image acquisition is controlled at 500-800 lux. The TCM syndrome differentiation theory library is a preset database, and the TCM syndrome differentiation theory library contains at least 1000 standard syndrome feature data sets verified clinically by the latest 1000 cases of top-level TCM institutions, and the number of dimensions of the core feature vector is within the range of 12-20. During the pulse data preprocessing, the wavelet threshold denoising algorithm is used to complete the preprocessing, and the time domain features include pulse wave conduction velocity, pulse width, and pulse amplitude variability, and the frequency domain features include main frequency, secondary main frequency, and main frequency band energy proportion.

9. An AI traditional Chinese four-examination instrument device, characterized in that, The device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, which, when executed by the processor, implements the execution steps of the integrated AI analysis TCM four-diagnosis comprehensive diagnosis method according to any one of claims 1 to 8.

Images