AI-based traditional Chinese medicine expert remote medical assistance system

Abstract

The invention discloses a traditional Chinese medicine expert remote medical assistance system based on AI, and the method comprises the following steps: filing a doctor seeing patient at a patient server, and supplementing the original medical history information into a patient file; selecting information collection at an outdoor end or a hospital end according to the actual condition of the patient; the collected data is subjected to auxiliary analysis through a MobileNetV3 algorithm, and an existing patient is subjected to pre-diagnosis; and matching traditional Chinese medicine experts with appropriate qualification, inquiry direction and inquiry experience according to a pre-diagnosis result, and performing doctor-patient butt joint to complete inquiry. According to the method, multiple traditional Chinese medicine signs are collected and pre-diagnosed through the MobileNetV3 model, the corresponding traditional Chinese medicine specialist is matched according to the pre-diagnosis result, and after the traditional Chinese medicine specialist and the patient are successfully matched, medical record data of the patient and related information about inspection diagnosis, smelling diagnosis and cut diagnosis are directly obtained, so that the diagnosis efficiency is improved. And pre-diagnosis and drug recommendation given by the AI system can be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to an AI-based remote medical assistance system for traditional Chinese medicine experts. Background Art

[0002] Telemedicine originated in 1973 and is the product of the combination of medicine and information technology. In the early days, the remote system of traditional Chinese medicine was centered on video consultation. For example, the "observation system" proposed in 2011 observed the patient's facial expressions and appearance through high-definition video. After 2013, technologies such as NET architecture and SQL database further optimized the system functions and supported electronic prescriptions and medical record management. In 2022, the application of LoRa technology and 5G networks improved anti-interference and real-time performance. Traditional Chinese medicine relies on "observation, smell, questioning, and palpation", and traditionally requires face-to-face consultation. However, my country's medical resources are unevenly distributed, and it is difficult for remote areas to obtain high-quality services. Telemedicine makes up for the regional gap through video consultation and data sharing, becoming an important way to solve resource imbalances.

[0003] After searching, the invention patent with national patent publication number CN107506603A provides a remote TCM-assisted diagnosis and treatment system and method, which includes the steps of: receiving the user's account information and logging in after verifying that the user's account information is correct; sending a remote medical request to the remote medical server; establishing a communication connection between the smart terminal and the doctor workstation assigned by the remote medical server; receiving the diagnosis and treatment information sent by the doctor workstation; receiving the symptom information input by the user in the symptom information collection interface; receiving the user's TCM physical sign information collected by the TCM physical sign collection device; sending the user's symptom information and TCM physical sign information to the doctor workstation; receiving the user's diagnosis and treatment information sent by the doctor workstation; displaying the user's diagnosis and treatment information on the diagnosis and treatment result display interface. This invention realizes remote TCM diagnosis and treatment through the Internet.

[0004] The above patent realizes remote TCM diagnosis and treatment, but there is a problem of slow efficiency of TCM diagnosis and treatment. The development of current scientific and technological level makes remote transmission of information more convenient, but there is a contradiction between weak primary medical capacity and growing patient demand. Studies have shown that on the one hand, the remote system can improve the efficiency of primary doctor training, and on the other hand, patients' medical needs for consultation with TCM experts can be effectively alleviated, and resources can be promoted to sink through regional medical platforms. Summary of the invention

[0005] The purpose of the present invention is to solve the problem of slow efficiency of TCM diagnosis and treatment in the prior art and to propose an AI-based TCM expert remote medical assistance system.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a remote medical assistance system for Chinese medicine experts based on AI, comprising the following steps:

[0007] Step S1, the patient undergoes verification registration at the patient service end, establishes a patient file, and retrieves the patient's medical history information at the database end, and the medical history information is directly added to the patient file;

[0008] Step S2, TCM physical sign collection, is divided into outdoor end and hospital end. The hospital end has sufficient and standardized medical equipment, and the outdoor end will prompt patients to use appropriate digital products to assist in information collection;

[0009] Step S3, before matching with a TCM expert for consultation, collect information on inspection, auscultation, and palpation according to the TCM diagnostic standards of inspection, auscultation, auscultation, and palpation, collect information related to the patient's facial features using image acquisition, collect the patient's voice and perform acoustic feature analysis, and patients at the hospital end can also use existing medical equipment to collect pulse or other TCM physical signs information;

[0010] Step S4: The data analysis end first uses the MobileNetV3 algorithm to assist in analysis based on the TCM signs collected in advance, performs pre-diagnosis on existing patients, and evaluates the accuracy of the pre-diagnosis results;

[0011] Step S5, according to the results of the preliminary diagnosis, matching a TCM expert with appropriate qualifications, consultation direction, and consultation experience according to the preliminary diagnosis results and the accuracy of the preliminary diagnosis results, and the doctor-patient connection is improved.

[0012] Furthermore, in step S1, the following sub-steps are also included:

[0013] S1-1, the patient service end is used for patient operation. The patient or the patient's family members register, verify and log in on the login interface of the patient service end. The independent file is established. Other relevant information of the patient can be independently entered into the file for supplementation, such as relevant medical records of other hospitals and clinics, usage records, etc.;

[0014] S1-2, the patient successfully logs in to the patient server, and the automatic extraction technology extracts the patient's original initial data in the system, including patient symptoms, disease diagnosis, hospitalization, laboratory examinations, surgical conditions, drug treatment and non-drug treatment, etc. Entity recognition and event extraction technology extract electronic medical record information, using the BERT-BiLSTM-CRF model.

[0015] Furthermore, in step S2, outdoor patients use digital products to collect information independently, take photos of the whole face, part of the tongue and lips according to the prompts, and take and upload picture information in a stable and appropriate environment according to the sound and image prompts of the patient service end, and upload the picture information of traditional Chinese medicine signs to the information processing end; patients on the hospital side will use professional and standardized medical instruments to collect information in the consultation room prepared by the hospital.

[0016] Furthermore, in step S3, the following sub-steps are also included:

[0017] S3-1, inspection, using electronic cameras to collect clear images of TCM signs, such as tongue and lip images;

[0018] S3-2, olfactory diagnosis, artificial intelligence-assisted olfactory diagnosis in traditional Chinese medicine focuses on auscultation, using a twenty-five-sound analyzer to analyze the patient's acoustic characteristics to provide a basis for clinical diagnosis. The patient emits stable vowels in a stable and noise-free environment to provide voice data, which is converted into a voice signal by the system. The voice signal is processed by audio data processing such as noise reduction, filtering, transformation and feature extraction as the input feature of the machine learning classifier to provide a basis for the patient's acoustic characteristics. The commonly used machine learning classifier in the field of auscultation is SVM. In addition, olfactory diagnosis may use an olfactory analyzer to collect information such as the patient's breath, and the final collected data is saved in the form of images;

[0019] S3-3, palpation, uses a pulse diagnostic instrument to collect information on pulse, breath and other TCM signs. The pulse diagnostic instrument uses piezoelectric or piezoresistive pressure sensors, Doppler ultrasonic sensors, photoelectric pulse sensors, etc. to convert the pulse into a series of digital signals and record them in a computer. In the future, it will be easier for artificial intelligence technology to process pulse diagnosis and convert digital signals into image signals for deep learning pre-diagnosis.

[0020] Furthermore, in step S4, the image collection of TCM physical signs is enhanced by transfer learning + lightweight CNN preprocessing, the MobileNetV3 model is selected, and the tongue image and lip image datasets are used to train the feature extraction layer parameters of CNN. The CNN used for questioning includes 4 steps:

[0021] S4-1, image data collection and preprocessing, including calibration of color, lighting, etc.;

[0022] S4-2, annotate the image to introduce prior information for the model to learn;

[0023] S4-3, design and optimization of model structure and network parameter training based on labeled image dataset;

[0024] S4-4, using the trained model to perform feature extraction and classification on the data images converted from inspection, auscultation and palpation.

[0025] Further, in step S5, after the patient's multiple objective TCM signs are collected and pre-diagnosed, the pre-diagnosis results are matched with the corresponding TCM experts. Based on the expert's historical consultation data, matrix decomposition or deep collaborative filtering is used to assist in calculating the matching degree. The expert characteristics recorded in big data are used to summarize different medical options for common diseases, and a mapping from pre-diagnosis to matching is constructed. It belongs to prior knowledge. The matching is related to the pre-diagnosis obtained by inspection, auscultation, and palpation. The pre-diagnosis obtained by the data processing end is included in the mapping range, and matching calculation is performed. Before the actual doctor-patient matching is determined, a mapping between pre-diagnosis and matching is established. The pre-diagnosis is expressed as a question-answer pair: z = (r, a), where z is the pre-diagnosis, r represents the results of different diagnostic points of the pre-diagnosis, and a represents the accuracy of the pre-diagnosis. For example:

[0026] The matching expression of tongue diagnosis (color) is z1 = (r, a) = (ruddy, clear)

[0027] Tongue diagnosis (tongue coating) is expressed as z2 = (r, a) = (clear, unclear)

[0028] Auscultation (full of qi in the middle) is expressed as z3 = (r, a) = (full of qi in the middle, clear) ......

[0030] Match m is expressed as a set of n pre-diagnostic sequences m = {z i |i=1,2,...,n}, including the specific information and relationship of inspection, auscultation and palpation. For example, the set consisting of three pre-diagnoses of TCM expert No. 1 is represented as follows: m={z i |i=1,2,3}, the matching order is to compare and analyze TCM signs in turn, and the TCM expert matching is determined. The final doctor-patient matching is determined by the matching degree between the pre-diagnosis set and the TCM expert's consultation ability. It is matched with all online TCM experts, and finally the TCM expert with the highest matching degree is selected.

[0031] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0032] The present invention trains the feature extraction layer parameters of CNN by using image data sets such as tongue images, lip images, acoustic images, and pulse images in the MobileNetV3 model. The trained model extracts and classifies information images, and adopts efficient activation functions to achieve automatic feature learning. The model construction has low complexity, a small number of weights, and can achieve weight sharing. It can also avoid complex manual feature extraction and data reconstruction processes, and has great advantages when applied to large-scale image recognition and classification.

[0033] The present invention uses extraction technology to extract the patient's initial medical record data originally in the system at the patient service end, and calculates the accuracy and recall rate of the medical record information, so that Chinese medicine experts can better understand the basic conditions of the patients and greatly reduce the consultation time. On the other hand, the information recorded in the database is more objective and can more intuitively and accurately reflect the basic conditions of the patients.

[0034] The method of the present invention can reduce the time that Chinese medicine experts spend on guiding patients at the medical service end. After the Chinese medicine experts and patients are successfully matched, the patient's medical records and relevant information about inspection, auscultation, and palpation can be directly obtained, and the pre-diagnosis and drug recommendations given by the AI ​​system can be obtained. Only the Chinese medicine experts are required to observe and conduct more flexible and complex consultations, which greatly reduces the consultation time for a single patient and is conducive to medical resources serving patients more comprehensively and fairly. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 A system flow chart provided for an embodiment of the present invention;

[0037] Figure 2 A mapping diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of an AI-based remote medical assistance system for Chinese medicine experts proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0039] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0040] The following examples are for illustrative purposes only and are not intended to limit the scope of the present invention.

[0041] The following is a detailed description of a specific solution of an AI-based TCM expert remote medical assistance system provided by the present invention in conjunction with the accompanying drawings.

[0042] Example

[0043] See also Figure 1 , which shows a method flow chart of an AI-based TCM expert remote medical assistance system provided by an embodiment of the present invention, the method comprising the following steps:

[0044] Step S1: The patient undergoes verification registration at the patient service end, establishes a patient file, and retrieves the patient's medical history information at the database end, and the medical history information is directly added to the patient file;

[0045] Wherein step S1 also includes the following sub-steps:

[0046] S1-1, the patient service end is used for patient operation. The patient or the patient's family members register, verify and log in on the login interface of the patient service end. The independent file is established. Other relevant information of the patient can be entered into the file for supplementation.

[0047] S1-1, the patient successfully logs in to the patient service end, and the automatic extraction technology extracts the patient's original initial data in the system, including patient symptoms, disease diagnosis, hospitalization, examination, surgery, drug treatment and non-drug treatment, etc. Entity recognition (NER) and event extraction technology extract electronic medical record information, using the BERT-BiLSTM-CRF model, the model's important parameters precision (Precision), recall rate (Recall) and F1 value The main parameters of the model are precision (Precision), recall rate (Recall) and F1 value. The technical formula is as follows:

[0048]

[0049] Precision P: The proportion of samples predicted by the model to be positive that are actually positive (to avoid false positives). The calculation method is the number of correctly predicted entities / the number of all entities predicted by the model.

[0050] Recall rate R: the proportion of samples that are actually positive that are correctly predicted by the model (avoiding missed samples), calculated as the number of correctly predicted entities / the number of all real entities;

[0051] F1 value: Combining the two, it is more valuable for reference in scenarios where the categories are unbalanced or false positives / missing negatives need to be balanced. The F1 value is the harmonic mean of precision and recall.

[0052] Let TP be the number of entities correctly predicted by the model, FP be the number of entities predicted by the model minus the number of other entities, and FN be the number of entities not predicted by the model. The technical formula can be improved to the following formula:

[0053]

[0054] Medical history data and report sheets will enter the information processing end, which will store the extracted symptoms, diseases and other entities in an Excel table according to a certain structure, so that the text information of the TCM electronic medical record expressed in natural language can be standardized and structured, making it easier to conduct in-depth mining and research on the TCM electronic medical record admission record information in the future.

[0055] Step S2, TCM physical sign collection, divided into outdoor end and hospital end. The hospital end has sufficient and standardized medical equipment, and the outdoor end will prompt patients to use appropriate digital products to assist in information collection;

[0056] In step S2, there are two cases: outdoor end and hospital end, where:

[0057] Patients at the outdoor end use digital products to collect information on their own. They take photos of their entire face, tongue, and lips as prompted. They also record specific audio in a stable and appropriate environment and upload it to the information processing end based on the sound and image prompts of the patient service end.

[0058] Patients in the hospital will use professional and standardized medical instruments to collect information in the consultation room prepared by the hospital. In addition to the relevant TCM vital signs information collected by outdoor patients, they can also use pulse diagnostic instruments, olfactory analyzers and other instruments to collect pulse, breath and other TCM vital signs information.

[0059] Step S3, before matching with a TCM expert for consultation, according to the TCM diagnostic standards of observation, auscultation, questioning and palpation, the patient's facial information is collected using image acquisition, and the patient's voice is collected and acoustic features are collected. Patients at the hospital can also use existing standardized medical equipment to collect pulse or other TCM physical signs information.

[0060] Wherein step S3 also includes the following sub-steps:

[0061] Observation diagnosis mainly involves the collection of picture information related to TCM physiognomy. According to the sound and image prompts of the patient's service end, specific files are photographed or recorded in a stable and appropriate environment and uploaded to the information processing end.

[0062] Auscultation, artificial intelligence-assisted auscultation in TCM diagnosis focuses on auscultation. Traditional machine learning methods are mainly used for artificial intelligence-assisted auscultation. A twenty-five-tone analyzer is used to analyze the patient's acoustic characteristics to provide a basis for clinical diagnosis. Sound data is collected from the patient. The patient makes stable vowels in a stable and noise-free environment to provide voice data, which is converted into voice signals by the system. The voice signals are processed through audio data processing such as noise reduction, filtering, transformation and feature extraction, and then used as input features of the machine learning classifier to provide a basis for the patient's acoustic characteristics. The commonly used machine learning classifier in the field of auscultation is SVM, which can distinguish between three types of people: lung qi deficiency, lung yin deficiency, and healthy people with an accuracy rate of over 85% under the condition of gender prior.

[0063] Palpation refers to the collection, analysis and diagnosis of pulse information. The four elements of pulse include pulse position, pulse rate, pulse shape and pulse strength. The development of digital sensor technology has made it possible to transform the cognition of pulse from human subjective perception into objective digital signals. Piezoelectric or piezoresistive pressure sensors, Doppler ultrasonic sensors, photoelectric pulse sensors, etc. can be used to convert pulse into a series of digital signals and record them in the computer. Artificial intelligence technology is used to process pulse diagnosis data and perform diagnosis. Artificial intelligence-assisted pulse diagnosis is standardized and accurate. Assisted pulse diagnosis generally consists of three stages: data collection and preprocessing, feature extraction, and pulse classification or differentiation.

[0064] In addition, it is necessary to modify the model structure and adjust parameters several times to improve the optimal performance of the data. The time domain features related to pulse classification and disease classification mainly include the main wave slope, main wave amplitude, phase, pulse width cycle ratio, etc., and the frequency domain features include wavelet coefficients, energy features, etc. Each learning algorithm has its own advantages in pulse classification. Learning-assisted pulse diagnosis can achieve an accuracy of 80% and an area under the curve (AUC) of 0.8.

[0065] Step S4: The data analysis end first uses the deep learning CNN system based on the TCM signs collected in advance, adopts transfer learning, lightweight CNN and preprocessing enhancement, uses the MobileNetV3 model, and uses image data sets such as tongue images, lip images, acoustic images, and pulse images to train the feature extraction layer parameters of CNN. The CNN used for medical consultation includes 4 steps:

[0066] Image data collection and preprocessing, including calibration of color, lighting, etc.; S2-2-1,

[0067] Annotate the image to introduce prior information for the model to learn;

[0068] Design and optimization of model structure and network parameter training based on annotated tongue image dataset;

[0069] Use the trained model to extract features and classify information images.

[0070] MobileNetV3 is based on two core ideas:

[0071] Depthwise separable convolution decomposes the standard convolution into depthwise convolution (channel-by-channel convolution) and point-by-point convolution (1×1 convolution), which greatly reduces the amount of calculation; the calculation complexity comparison formula is as follows:

[0072] Standard convolution: C out ×K 2 ×C in ×H×W

[0073] ο Depthwise Separable Convolution: K 2 ×C in ×H×W+C out ×C in H×W

[0074] Where K is the convolution kernel size, C in / C out is the number of input / output channels, and H / W is the feature map size.

[0075] The Inverted Residuals structure, in contrast to the "wide-narrow-wide" structure of ResNet, adopts a "narrow-wide-narrow" design: first expand the number of channels through 1×1 convolution, then perform depth-separable convolution, and finally compress the channels. Combined with the linear activation function (Linear Bottleneck), information loss is avoided.

[0076] The main mechanisms of MobileNetV3 include:

[0077] Attention mechanism: MobileNetV3 integrates the SE (Squeeze-and-Excitation) module, which generates channel attention weights through global average pooling to enhance the feature response of important channels. Improved versions such as ECA (Efficient Channel Attention) further reduce parameters. Some studies replace SE with CBAM (combination of channel and spatial attention) to improve feature focusing capabilities.

[0078] Efficient activation function, using h-swish (Hard-Swish) activation function, approximate calculation of Swish function

[0079] (x·σ(βx))

[0080] Avoid high computational cost. Its segmented form is:

[0081]

[0082] Take into account both nonlinearity and speed;

[0083] Lightweight module optimization, introducing NAS (Neural Architecture Search) to automatically search for network structure, perform multi-objective optimization for latency and accuracy, and generate efficient layer configuration and channel number combinations.

[0084] To adapt to edge devices, MobileNetV3 is often combined with the following methods for further compression:

[0085] Structured pruning, through sparse regularization training to screen redundant filters, uses the product of the convolutional layer sparse value and the batch normalization (BN) scaling factor as the pruning basis, reducing the number of parameters by 44.5%.

[0086] Parameter quantization, converting 32-bit floating-point weights to low bit width (such as 8-bit integers), combined with mixed precision quantization (such as the AWQ method), assigning different bit widths according to the importance of the convolutional layer, balancing the calculation speed and accuracy loss (performance loss <10%).

[0087] Attention module replacement: Use ShiftViT module to replace SE module, which can reduce the number of parameters while maintaining accuracy; or introduce dilated convolution to expand the receptive field and reduce the number of layers.

[0088] The overall accuracy of the model can reach more than 90%, far exceeding traditional machine learning methods. Based on the robustness of CNN, the MobileNetV3 model can be easily migrated to images captured by other instruments with different lighting and different hardware equipment, which is more conducive to the collection of TCM physical signs of outdoor patients. In addition, compared with traditional recognition algorithms, convolutional neural networks can realize automatic feature learning, and their model construction has low complexity, a small number of weights, and can realize weight sharing. It can also avoid complex manual feature extraction and data reconstruction processes, and has great advantages when applied to large-scale image recognition and classification.

[0089] Step S5: According to the result of the pre-diagnosis, a TCM expert with appropriate qualifications, consultation direction, and consultation experience is matched according to the pre-diagnosis result and the accuracy of the pre-diagnosis result;

[0090] The representation and organization of patient matching involves the patient's pre-diagnosed symptoms and the candidate answers of the specific TCM experts. The TCM experts have accumulated clinical practice in diagnosis. Different experts have different authoritative fields and consultation methods. First, the characteristics of experts recorded in big data summarize different medical options for common symptoms and construct a mapping from pre-diagnosis to matching, which belongs to prior knowledge. It should be noted that matching is also related to the pre-diagnosis obtained by inspection, auscultation, and palpation, but such symptoms have been considered before the consultation, so only the symptoms obtained through the consultation are included in the mapping scope. The mapping diagram is as follows: Figure 2, make matching calculations and allocations.

[0091] In the process of determining the actual doctor-patient match, a mapping between pre-diagnosis and matching is established. The pre-diagnosis is expressed as a question-answer pair: z = (r, a), where z is the pre-diagnosis, r represents the results of different diagnostic points of the pre-diagnosis, and a represents the accuracy of the pre-diagnosis, for example:

[0092] The matching expression of tongue diagnosis (color) is z1 = (r, a) = (ruddy, clear)

[0093] Tongue diagnosis (tongue coating) is expressed as z2 = (r, a) = (clear, unclear)

[0094] Auscultation (full of qi in the middle) is expressed as z3 = (r, a) = (full of qi in the middle, clear) ......

[0096] Match m is expressed as a set of n pre-diagnostic sequences m = {z i |i=1,2,...,n}, including the specific information and relationship of inspection, auscultation and palpation. For example, the set consisting of three pre-diagnoses of TCM expert No. 1 is represented as follows: m={z i |i=1,2,3},

[0097] The matching order is to compare and analyze TCM signs in turn. The matching of TCM experts is determined by the matching degree between the pre-diagnosis set and the TCM expert's diagnosis ability. It is matched with all online TCM experts, and finally the TCM expert with the highest matching degree is selected.

[0098] The complexity and authority of TCM consultation make it impossible to have authority by relying solely on the system for diagnosis. Therefore, pre-diagnosis and system matching are only used as auxiliary means to greatly reduce the time that TCM experts spend guiding patients on the medical service side. After the TCM experts and patients successfully complete the matching, they can directly obtain the patient's medical records and relevant information about inspection, auscultation, and palpation, and can obtain pre-diagnosis and drug recommendations from the AI ​​system. TCM experts only need to observe and ask more flexible questions to make a diagnosis. The system greatly reduces the consultation time of a single patient, which is conducive to medical resources serving patients more comprehensively and fairly.

[0099] In this way, an AI-based remote medical assistance system for TCM experts can be realized.

[0100] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An AI-based remote medical assistance system for Chinese medicine experts, characterized in that: The following steps are involved: Step S1, the patient undergoes verification registration at the patient service end, establishes a patient file, and retrieves the patient's medical history information at the database end, and the medical history information is directly added to the patient file; Step S2, TCM physical sign collection, is divided into outdoor end and hospital end. The hospital end has sufficient and standardized medical equipment, and the outdoor end will prompt patients to use appropriate digital products to assist in information collection; Step S3, before matching with a TCM expert for consultation, collect information on inspection, auscultation, and palpation according to the TCM diagnostic standards of inspection, auscultation, auscultation, and palpation, collect information related to the patient's facial features using image acquisition, collect the patient's voice and perform acoustic feature analysis, and patients at the hospital end can also use existing medical equipment to collect pulse or other TCM physical signs information; Step S4: The data analysis end first uses the MobileNetV3 algorithm to assist in analysis based on the TCM signs collected in advance, performs pre-diagnosis on existing patients, and evaluates the accuracy of the pre-diagnosis results; Step S5, according to the results of the preliminary diagnosis, matching a TCM expert with appropriate qualifications, consultation direction, and consultation experience according to the preliminary diagnosis results and the accuracy of the preliminary diagnosis results, and the doctor-patient connection is improved.

2. The AI-based TCM expert remote medical assistance system as claimed in claim 1, characterized in that: Wherein step S1 also includes the following sub-steps: S1-1, the patient service end is used for patient operation. The patient or the patient's family members register, verify and log in on the login interface of the patient service end. The independent file is established. Other relevant information of the patient can be independently entered into the file for supplementation, such as relevant medical records of other hospitals and clinics, usage records, etc.; S1-2, the patient successfully logs in to the patient server, and the automatic extraction technology extracts the patient's original initial data in the system, including patient symptoms, disease diagnosis, hospitalization, laboratory examinations, surgical conditions, drug treatment and non-drug treatment, etc. Entity recognition and event extraction technology extract electronic medical record information, using the BERT-BiLSTM-CRF model.

3. The AI-based TCM expert remote medical assistance system as claimed in claim 1, characterized in that: In step S2, outdoor patients use digital products to collect information independently, take photos of the whole face, part of the tongue and lips according to the prompts, and take and upload picture information in a stable and appropriate environment according to the sound and image prompts of the patient service end, and upload the picture information of traditional Chinese medicine signs to the information processing end; patients on the hospital side will use professional and standardized medical instruments to collect information in the consultation room prepared by the hospital.

4. The AI-based TCM expert remote medical assistance system according to claim 1, characterized in that: Wherein step S3 also includes the following sub-steps: S3-1, inspection, using electronic cameras to collect clear images of TCM signs, such as tongue and lip images; S3-2, olfactory diagnosis, artificial intelligence-assisted olfactory diagnosis in traditional Chinese medicine focuses on auscultation, using a twenty-five-sound analyzer to analyze the patient's acoustic characteristics to provide a basis for clinical diagnosis. The patient emits stable vowels in a stable and noise-free environment to provide voice data, which is converted into a voice signal by the system. The voice signal is processed by audio data processing such as noise reduction, filtering, transformation and feature extraction as the input feature of the machine learning classifier to provide a basis for the patient's acoustic characteristics. The commonly used machine learning classifier in the field of auscultation is SVM. In addition, olfactory diagnosis may use an olfactory analyzer to collect information such as the patient's breath, and the final collected data is saved in the form of images; S3-3, palpation, uses a pulse diagnostic instrument to collect information on pulse, breath and other TCM signs. The pulse diagnostic instrument uses piezoelectric or piezoresistive pressure sensors, Doppler ultrasonic sensors, photoelectric pulse sensors, etc. to convert the pulse into a series of digital signals and record them in a computer. In the future, it will be easier for artificial intelligence technology to process pulse diagnosis and convert digital signals into image signals for deep learning pre-diagnosis.

5. The AI-based TCM expert remote medical assistance system as claimed in claim 2, characterized in that: include: In step S4, the image collection of TCM physical signs is enhanced by transfer learning + lightweight CNN preprocessing, the MobileNetV3 model is selected, and the tongue and lip image datasets are used to train the feature extraction layer parameters of CNN. The CNN used for questioning includes 4 steps: S4-1, image data collection and preprocessing, including calibration of color, lighting, etc.; S4-2, annotate the image to introduce prior information for the model to learn; S4-3, design and optimization of model structure and network parameter training based on labeled image dataset; S4-4, using the trained model to perform feature extraction and classification on the data images converted from inspection, auscultation and palpation.

6. The AI-based TCM expert remote medical assistance system according to claim 1, characterized in that: In step S5, after the patient's multiple objective TCM signs are collected and pre-diagnosed, the pre-diagnosis results are matched with the corresponding TCM experts. Based on the expert's historical consultation data, matrix decomposition or deep collaborative filtering is used to assist in calculating the matching degree. The expert characteristics recorded in big data are used to summarize different medical options for common diseases, and a mapping from pre-diagnosis to matching is constructed. It belongs to prior knowledge. The matching is related to the pre-diagnosis obtained by inspection, auscultation, and palpation. The pre-diagnosis obtained by the data processing end is included in the mapping range and the matching calculation is performed. Before the actual doctor-patient matching is determined, the mapping between pre-diagnosis and matching is established. The pre-diagnosis is expressed as a question-answer pair: z = (r, a), where z is the pre-diagnosis, r represents the results of different diagnostic points of the pre-diagnosis, and a represents the accuracy of the pre-diagnosis. For example: The matching expression of tongue diagnosis (color) is z1 = (r, a) = (ruddy, clear) Tongue diagnosis (tongue coating) is expressed as z2 = (r, a) = (clear, unclear) Auscultation (full of qi in the middle) is expressed as z3 = (r, a) = (full of qi in the middle, clear) ...... Match m is expressed as a set of n pre-diagnostic sequences m = {z i |i=1,2,...,n}, including the specific information and relationship of inspection, auscultation and palpation. For example, the set consisting of three pre-diagnoses of TCM expert No. 1 is represented as follows: m={z i |i=1,2,3}, the matching order is to compare and analyze TCM signs in turn, and the TCM expert matching is determined. The final doctor-patient matching is determined by the matching degree between the pre-diagnosis set and the TCM expert's consultation ability. It is matched with all online TCM experts, and finally the TCM expert with the highest matching degree is selected.

Images