Telemedicine online consultation method, device, storage medium, and program product

Abstract

The disclosed embodiments disclose a telemedicine online consultation method, device, storage medium, and program product. The method includes: in response to a consultation request initiated by any participant, authenticating all participants, where the participants are patients or doctors; creating a consultation room for participants who have passed the identity authentication; recommending multiple available medical models to doctors through the consultation room and determining the designated model through the doctor's designated operation; displaying the input data required for the designated model in the consultation room; in response to the upload operation of the participant, collecting the input data, and obtaining a diagnosis report based on the input data and the designated model; displaying the diagnosis report in the consultation room for reference by the participants; and closing the consultation room in response to an end request initiated by any participant. The method can provide an environment for remote multi-party consultation by creating a virtual consultation room, and improve the convenience, accuracy, and efficiency of diagnosis through intelligent model recommendation and sharing of model diagnosis results.

Description

Technical Field

[0001] The present disclosure relates to the field of medical health technology, and in particular to a remote medical online consultation method, device, storage medium, and program product. Background Art

[0002] With the rapid advancement of Internet technology and the widespread use of smart mobile devices, the medical industry is facing an opportunity for rapid transformation. The traditional medical model usually requires patients to communicate with doctors face to face. This model brings many challenges to patients' access to high-quality medical services in areas with remote locations or insufficient medical resources. These areas often lack professional medical personnel and advanced medical equipment, making it difficult for patients to obtain timely and effective diagnosis and treatment. Summary of the invention

[0003] In view of this, the embodiments of the present disclosure provide a telemedicine online consultation method, device, storage medium, and program product, which can provide a remote multi-party consultation environment by creating a virtual consultation room, and improve the convenience, accuracy, and efficiency of diagnosis through intelligent model recommendation and sharing of model diagnosis results.

[0004] In a first aspect, the present disclosure provides a telemedicine online consultation method, which adopts the following technical solution:

[0005] In response to a consultation request initiated by any participant, identity authentication is performed on all participants, where the participant is a patient or a doctor;

[0006] Create a consultation room for participants who have passed identity authentication;

[0007] Recommend multiple available medical models to doctors through the consultation room;

[0008] In response to a doctor's designated operation among a plurality of available medical models, determining a designated model;

[0009] Presenting input data required for the specified model in the consultation room;

[0010] In response to an upload operation by a participant, the input data is collected, and a diagnosis report is obtained based on the input data and the specified model;

[0011] Displaying the diagnostic report in the consultation room for reference by the participants;

[0012] In response to a termination request initiated by any participant, the consultation room is closed.

[0013] Optionally, the consultation room recommends multiple available medical models to the doctor, including:

[0014] Determine whether the physician conducts a keyword search;

[0015] If yes, obtaining the keywords input by the doctor, and filtering out the multiple available medical models in the model library based on the keywords;

[0016] If not, obtaining the patient's medical record text, and filtering out the multiple available medical models in the model library based on the medical record text and the label library;

[0017] The plurality of available medical models are displayed in a doctor's consultation room.

[0018] Optionally, the label library includes a first label set, a second label set and a third label set for each medical model;

[0019] Preprocessing the medical record text;

[0020] Perform entity recognition and entity classification on the preprocessed medical record text to obtain the patient's basic information set, symptom information set, and diagnosis information set;

[0021] Obtaining a first matching degree between the basic information set and the first tag set;

[0022] Obtaining a second matching degree between the symptom information set and the second label set;

[0023] Obtaining a third matching degree between the diagnostic information set and the third tag set;

[0024] Performing weighted summation on the first matching degree, the second matching degree, and the third matching degree to obtain a degree of fit between the patient and each medical model;

[0025] The plurality of available medical models are selected based on the degree of fit.

[0026] Optionally, the calculation formula of the first matching degree is:

[0027] PAa=f1+f2;

[0028] Wherein, A is the number of the first label set; PAa is the first matching degree between the basic information set and the first label set numbered A; f1 is the first score; f2 is the second score;

[0029] The first label set includes applicable age groups, applicable genders, age scores, and gender scores of the medical model;

[0030] The basic information set includes the patient's real age and real gender;

[0031] If the real age is within the applicable age range, determining that the first score is equal to the age score;

[0032] If the real age is not within the applicable age range, the first score is determined to be equal to 0;

[0033] If the real gender is the applicable gender, determining that the second score is equal to the gender score;

[0034] If the actual gender is not the applicable gender, the second score is determined to be equal to 0.

[0035] Optionally, the second label set includes a plurality of symptom labels and a first label threshold corresponding to each symptom label;

[0036] Set the first parameter to be initially 0;

[0037] Traversing all disease symptom information in the symptom information set;

[0038] Calculate the similarity between the current disease symptom information and the symptom label;

[0039] If the similarity reaches the corresponding first tag threshold, the first parameter is increased by 1;

[0040] If the similarity does not reach the corresponding first tag threshold, the first parameter remains unchanged;

[0041] After all the disease symptom information is traversed, the second matching degree is obtained based on the first parameter and the total number of symptom labels in the second label set;

[0042] The calculation formula of the second matching degree is:

[0043] PBb=S1 / S2;

[0044] Among them, B is the number of the second label set; PBb is the second matching degree between the symptom information set and the second label set numbered B; S1 is the first parameter; S2 is the total number of symptom labels.

[0045] Optionally, the third label set includes a plurality of diagnostic labels and a second label threshold corresponding to each diagnostic label;

[0046] Set a second parameter initially set to 0, and traverse all the diagnosis result information in the diagnosis information set;

[0047] Calculating the similarity between the current diagnosis result information and the diagnosis label;

[0048] If the similarity reaches the corresponding second tag threshold, the second parameter is increased by 1;

[0049] If the similarity does not reach the corresponding second tag threshold, the second parameter remains unchanged;

[0050] After all the diagnosis result information is traversed, the third matching degree is obtained based on the second parameter and the total number of diagnosis tags in the third tag set;

[0051] The calculation formula of the third matching degree is:

[0052] PCc=S3 / S4;

[0053] Among them, C is the number of the second label set; PCc is the third matching degree between the diagnostic information set and the third label set numbered C; S3 is the second parameter; S4 is the total number of diagnostic labels.

[0054] Optionally, if the designated model is a residual visual Mamba layer model, the input data is a 3D medical image, and the output is a diagnostic report including a 3D spine segmentation result.

[0055] In a second aspect, the disclosed embodiment further provides a telemedicine online consultation system, which adopts the following technical solution:

[0056] An identity authentication module, used to perform identity authentication on all participants in response to a consultation request initiated by any participant, wherein the participant is a patient or a doctor;

[0057] The consultation room creation module is used to create a consultation room for participants who have passed identity authentication;

[0058] A model recommendation module is used to recommend multiple available medical models to doctors through the consultation room;

[0059] A model specifying module, for determining a specified model in response to a doctor's specifying operation among a plurality of available medical models;

[0060] A data display module, used for displaying the input data required by the specified model in the consultation room;

[0061] A data collection module, configured to collect the input data in response to an upload operation of a participant, and obtain a diagnosis report based on the input data and the specified model;

[0062] A report display module, used to display the diagnosis report in the consultation room for reference by the participants;

[0063] The consultation room closing module is used to close the consultation room in response to a termination request initiated by any participant.

[0064] In a third aspect, the embodiments of the present disclosure further provide a computer device, which adopts the following technical solution:

[0065] The computer device comprises:

[0066] at least one processor; and,

[0067] a memory communicatively connected to the at least one processor; wherein,

[0068] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any of the above-mentioned remote medical online consultation methods.

[0069] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute any of the above-mentioned telemedicine online consultation methods.

[0070] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, including a computer program / instruction, which implements the steps of any of the above methods when executed by a processor.

[0071] The telemedicine online consultation method provided by the embodiment of the present disclosure can quickly respond to the consultation request initiated by any participant (patient or doctor), and authenticate the identity of all participants to ensure the security of the consultation process and the authenticity of the identity of the participants, and provide a basic guarantee for subsequent medical activities. After the identity authentication is passed, a consultation room is automatically created for the participants, providing a virtual but efficient communication space, making the consultation process more flexible and not restricted by physical locations. The consultation room recommends multiple available medical models to doctors. These models are professional diagnostic tools built based on technologies such as big data and artificial intelligence, so that doctors can have more auxiliary diagnostic means when facing complex cases, improving the accuracy and scientificity of diagnosis. Doctors can choose a suitable model from multiple available medical models for use according to actual needs. This personalized diagnosis process design enables each doctor to choose the most suitable medical model for the current case based on their professional knowledge and experience. After the model is specified, the consultation room displays the input data required for the specified model, responds to the upload operation of the participant, quickly collects these data, and quickly generates a diagnosis report based on these data and the specified model, greatly improving the efficiency of diagnosis. The diagnosis report is displayed in the consultation room for reference by the participants. Multiple doctors can discuss and analyze the diagnosis results together, realize real-time information sharing and multi-party collaboration, and help to formulate a more comprehensive and scientific diagnosis and treatment plan. When the consultation is over, any participant can initiate an end request, and then close the consultation room in time to release related resources and realize efficient management of system resources.

[0072] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0074] Figure 1 A flowchart of a telemedicine online consultation method provided by an embodiment of the present disclosure;

[0075] Figure 2 A flowchart of a medical model recommendation method provided in an embodiment of the present disclosure;

[0076] Figure 3 A flowchart of a medical model screening method based on medical record text provided in an embodiment of the present disclosure;

[0077] Figure 4 A block diagram of the principle of a telemedicine online consultation system provided by an embodiment of the present disclosure;

[0078] Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0079] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0080] It should be clear that the following embodiments of the present disclosure are described by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0081] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0082] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0083] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0084] Reference Figure 1 The present disclosure provides a telemedicine online consultation method, comprising the following steps:

[0085] S1: In response to a consultation request initiated by any participant, all participants are authenticated, and the participants are either patients or doctors;

[0086] S2: Create a consultation room for participants who have passed identity authentication;

[0087] S3: Recommend multiple available medical models to doctors through the consultation room;

[0088] S4: In response to the doctor's designated operation among multiple available medical models, determining a designated model;

[0089] S5: Present the input data required for the specified model in the consultation room;

[0090] S6: In response to the upload operation of the participant, collect input data, and obtain a diagnosis report based on the input data and the specified model;

[0091] S7: Display the diagnostic report in the consultation room for participants’ reference;

[0092] S8: In response to a termination request initiated by any participant, the consultation room is closed.

[0093] The telemedicine online consultation method disclosed in the present invention can quickly respond to the consultation request initiated by any participant (patient or doctor), and authenticate the identity of all participants to ensure the security of the consultation process and the authenticity of the identity of the participants, providing a basic guarantee for subsequent medical activities. After the identity authentication is passed, a consultation room is automatically created for the participants, providing a virtual but efficient communication space, making the consultation process more flexible and not restricted by physical locations. The consultation room recommends multiple available medical models to doctors. These models are professional diagnostic tools built based on technologies such as big data and artificial intelligence, so that doctors can have more auxiliary diagnostic means when facing complex cases, improving the accuracy and scientificity of diagnosis. Doctors can choose a suitable model from multiple available medical models according to actual needs. This personalized diagnosis process design enables each doctor to choose the most suitable medical model for the current case according to their professional knowledge and experience. After the model is specified, the consultation room displays the input data required for the specified model, responds to the upload operation of the participant, quickly collects these data, and quickly generates a diagnosis report based on these data and the specified model, greatly improving the efficiency of diagnosis. The diagnosis report is displayed in the consultation room for reference by the participants. Multiple doctors can discuss and analyze the diagnosis results together, realize real-time information sharing and multi-party collaboration, and help to formulate a more comprehensive and scientific diagnosis and treatment plan. When the consultation is over, any participant can initiate an end request, and then close the consultation room in time to release related resources and realize efficient management of system resources.

[0094] In summary, this telemedicine online consultation method has significantly improved the convenience, efficiency and treatment effect of telemedicine services through its beneficial effects in many aspects, such as efficient response, identity protection, flexible creation of consultation environment, intelligent model recommendation, personalized diagnosis process, data collection and efficient diagnosis, and convenient termination and resource management.

[0095] In S1, patients or doctors access the telemedicine platform (hereinafter referred to as the platform) through their personal terminal devices (such as smartphones, tablets, or computers) and initiate consultation requests.

[0096] For any participant who wants to participate in the consultation, they need to be authenticated. For registered users, they need to authenticate with their username and password. After successful login, they are further required to perform biometric verification, such as facial recognition or fingerprint recognition, to confirm their identity.

[0097] Participants who do not have a username and password, that is, new users, need to register first. Specifically, if the participant is a patient, a valid ID card or social security card is required for real-name authentication. After completing the real-name authentication, the face or fingerprint information is entered, and then the SMS verification code is used for secondary verification. After the verification is passed, the user can set the username and password, and the platform will record this information for subsequent use. If the participant is a doctor, he or she needs to upload documents such as a practicing certificate, photos of the front and back of the ID card, and the authorization certificate of the medical institution where he or she is located. The platform will digitally authenticate these documents to ensure the authenticity of the doctor's identity. After the identity verification is passed, the doctor can set the username and password, and the platform will also record this information for subsequent use.

[0098] In the above authentication process, JWT (JSON Web Tokens) can also be used for identity authentication and authorization to ensure the authenticity of the identity of doctors and patients participating in the consultation and the correctness of their access rights under authorization. The platform uses HTTPS encrypted transmission technology to protect user information and ensure that all transmission processes comply with network security standards. It also adopts a mechanism of continuous monitoring and regular re-verification to ensure the continued legitimacy of user identities.

[0099] Please note that there is usually only one patient, but the consultation room also supports multiple patients using it at the same time. There can be one or more doctors. The consultation room allows multiple doctors to conduct remote consultations with patients at the same time.

[0100] In S2, when a participant passes identity authentication, a consultation room is created on the participant's terminal device. After the participant enters the consultation room, he or she can conduct remote video and voice communication with other participants.

[0101] The consultation room is personalized according to the different needs of patients and doctors. For example, for patients, the consultation room interface highlights treatment plans and educational materials, allows patients to set privacy levels as needed, reminds patients what information can be shared and what information should be kept confidential, and provides barrier-free functions such as voice recognition, high contrast mode, font enlargement, etc. for patients. For patients, the consultation room interface focuses on medical record viewing and diagnostic tools.

[0102] In S3, refer to Figure 2 The flowchart of the medical model recommendation method shown in the figure recommends multiple available medical models to doctors through the consultation room, including the following steps:

[0103] S31: Determine whether the doctor performs a keyword search; if so, execute S32; if not, execute S33;

[0104] S32: Obtain keywords input by the doctor, and screen out multiple available medical models in the model library based on the keywords;

[0105] S33: Obtain the patient's medical record text, and filter out multiple available medical models in the model library based on the medical record text and the label library;

[0106] S34: Display multiple available medical models in a doctor's consultation room.

[0107] Specifically, if the doctor enters a keyword in the search bar of the consultation room to search, the system will provide the doctor with targeted model recommendations based on the keyword; if the doctor is not detected to be searching for a keyword, the system will obtain the medical record text uploaded by the patient and recommend a medical model suitable for the patient's condition based on the medical record text. The doctor can also choose to share the recommended medical model with other doctors or patients so that they can discuss and decide which medical model to use in the end.

[0108] In S32, a specific implementation scheme of “screening out multiple available medical models in a model library based on keywords” includes:

[0109] Vectorize several keywords entered by the doctor to generate a first vector sequence. The model library contains multiple medical models, which are concentrated in the platform. Each medical model has a corresponding dedicated description, including model name, indications, symptoms, diseases, recommended treatments, etc. The dedicated description of each medical model is vectorized to generate multiple second vector sequences. Calculate the cosine similarity between the first vector sequence and each second vector sequence, recorded as the first similarity, sort all medical models in descending order according to the first similarity, and select the top K medical models for recommendation.

[0110] Another specific implementation scheme of “screening out multiple available medical models in a model library based on keywords” includes:

[0111] All medical models in the model library are grouped according to different diseases (such as cardiovascular diseases, respiratory diseases, etc.) and / or symptoms (such as fever, cough, etc.). For each group of models, a general description is set, which includes symptoms, diseases, etc. that are common to all medical models in the group. Several keywords entered by the doctor are vectorized to generate a first vector sequence. The general description of each group of models is vectorized to generate multiple third vector sequences. The cosine similarity between the first vector sequence and each third vector sequence is calculated, recorded as the second similarity, and a group of models with the largest second similarity is selected as the optimal group. Each medical model also has a corresponding special description, including model name, indications, symptoms, diseases, recommended treatments, etc. The special description of the medical model in the optimal group is vectorized to generate multiple fourth vector sequences. The cosine similarity between the first vector sequence and each fourth vector sequence is calculated, recorded as the third similarity, and the medical models in the optimal group are sorted in descending order according to the third similarity, and the top K medical models are selected for recommendation.

[0112] In S33, the tag library includes a first tag set, a second tag set and a third tag set for each medical model, the first tag set includes the medical model adaptation, and an index is established between each medical model and the corresponding first tag set, second tag set and third tag set to facilitate search in subsequent calculations. Figure 3 The flowchart of the medical model screening method based on medical record text is shown. Based on the medical record text and the label library, multiple available medical models are screened out in the model library, including the following steps:

[0113] S331: Preprocess the medical record text.

[0114] Preprocessing includes text cleaning, standardization, sentence segmentation, and part-of-speech tagging. Text cleaning refers to removing noise to ensure that the medical record text has no irrelevant information, such as headers, footers, and comments; standardization refers to converting medical terms and abbreviations into standard forms to ensure the consistency of terminology; sentence segmentation refers to using natural language processing (NLP) tools such as NLTK or spaCy to segment the medical record text; part-of-speech tagging refers to using part-of-speech tagging tools (such as CRF, LSTM, and other models) to mark the part of speech of each word in order to better understand the context.

[0115] S332: Perform entity recognition and entity classification on the preprocessed medical record text to obtain the patient's basic information set, symptom information set and diagnosis information set.

[0116] Among them, the pre-trained NER model (such as BERT, BioBERT, etc.) is used to perform entity recognition on the pre-processed medical record text, and the recognized entities are classified (i.e. entity classification). The classification results include the basic information set, symptom information set and diagnosis information set. Among them, the basic information set includes the real age and real gender of the patient, the symptom information set is a collection of information on several kinds of symptoms of the patient, specifically including the name of each symptom (such as fever, cough, headache, etc.) and detailed description (such as intensity, duration, onset situation, etc.); the diagnosis information set is a collection of information on several kinds of diagnosis results for the patient, specifically including the name of the disease (such as diabetes, hypertension, etc.), the stage of the disease (such as early stage, terminal stage) and the diagnostic criteria (such as clinical examination results, laboratory test data, imaging results).

[0117] S333: Obtain a first matching degree between the basic information set and the first tag set.

[0118] Among them, the first label set includes the applicable age group, applicable gender (can be at least one of male and female), age score and gender score of the medical model. If the real age in the basic information set is within the applicable age group, the first score is determined to be equal to the age score; if the real age is not within the applicable age group, the first score is determined to be equal to 0. If the real gender in the basic information set is the applicable gender, the second score is determined to be equal to the gender score; if the real gender is not the applicable gender, the second score is determined to be equal to 0. The first score and the second score are weighted and summed to obtain the first matching degree. Specifically, the calculation formula of the first matching degree is:

[0119] PAa=f1+f2;

[0120] In the formula, A is the number of the first label set, and A corresponds to the number i of the medical model one by one; PAa is the first matching degree between the basic information set and the first label set numbered A; f1 is the first score; and f2 is the second score.

[0121] S334: Obtain a second matching degree between the symptom information set and the second label set.

[0122] In S334, the second label set includes several symptom labels and the first label threshold corresponding to each symptom label, which means that the second label set includes various labels of symptoms applicable to the medical model, such as the name, severity, symptom frequency, symptom duration, occurrence context, etc. of the applicable symptoms. For each label, a corresponding threshold is also set, which is recorded as the first label threshold, so as to facilitate setting targeted matching strength for different symptoms.

[0123] Set the first parameter to 0 initially, traverse all the disease symptom information in the symptom information set, record the disease symptom information being traversed as the current disease symptom information, calculate the similarity between the current disease symptom information and the symptom label (using the cosine similarity algorithm or Jaccard similarity), record it as the fourth similarity, if the fourth similarity reaches the corresponding first label threshold, then the first parameter is increased by 1, if the fourth similarity does not reach the corresponding first label threshold, then the first parameter remains unchanged. After all the disease symptom information is traversed, the second matching degree is obtained based on the final first parameter and the total number of symptom labels in the second label set. Specifically, the calculation formula for the second matching degree is:

[0124] PBb=S1 / S2;

[0125] In the formula, B is the number of the second label set, and B corresponds to the number i of the medical model one by one; PBb is the second matching degree between the symptom information set and the second label set numbered B; S1 is the first parameter; S2 is the total number of symptom labels.

[0126] S335: Obtain a third matching degree between the diagnosis information set and the third label set.

[0127] In S335, the third label set includes several diagnostic labels and the second label threshold corresponding to each diagnostic label, which means that the third label set includes various labels applicable to the diagnostic results of the medical model, such as the name of the disease, the stage (such as early stage, late stage), experimental data, etc. For each label, a corresponding threshold is also set, which is recorded as the second label threshold, so as to facilitate setting targeted matching strength for different diagnostic results.

[0128] Set the second parameter to 0 initially, traverse all the diagnostic result information in the diagnostic information set, record the diagnostic result information being traversed as the current diagnostic result information, calculate the similarity between the current diagnostic result information and the diagnostic label (using the cosine similarity algorithm or Jaccard similarity), record it as the fifth similarity, if the fifth similarity reaches the corresponding second label threshold, then the second parameter is increased by 1, if the fifth similarity does not reach the corresponding second label threshold, then the second parameter remains unchanged. After all the diagnostic result information is traversed, the third matching degree is obtained based on the final second parameter and the total number of diagnostic labels in the third label set. Specifically, the calculation formula for the third matching degree is:

[0129] PCc=S3 / S4;

[0130] In the formula, C is the number of the second label set, and C corresponds to the number i of the medical model one by one; PCc is the third matching degree between the diagnosis information set and the third label set numbered C; S3 is the second parameter; S4 is the total number of diagnosis labels.

[0131] S336: Perform weighted summation on the first matching degree, the second matching degree, and the third matching degree to obtain the degree of fit between the patient and each medical model.

[0132] The calculation formula of the fitness is:

[0133] Pi=W1* PAa + W2*PBb + W3*PCc;

[0134] Where i is the number of the medical model; Pi is the degree of fit between the patient and the medical model numbered i; W1, W2 and W3 are all preset weights of the medical model numbered i.

[0135] S337: Select multiple available medical models based on the degree of fit.

[0136] In S337, all medical models are sorted in descending order of fitness, and the first K medical models are selected as available medical models.

[0137] In S4, multiple available medical models are displayed using charts, for example, the characteristics, required input data, advantages and applicable scenarios of multiple available medical models are displayed in a list, which is convenient for doctors to view and discuss. When the doctor decides which medical model to choose, he or she specifies it by clicking the mouse, and in response to the doctor's specified operation, the final specified model is determined from the multiple available medical models.

[0138] In S5, each medical model may have different requirements for input data. Therefore, after the specified model is determined, the input data required for the specified model is displayed in the consultation room. Specifically, a data requirement display area is set in the user interface of the consultation room. The input data required for the specified model is clearly listed in the data requirement display area, such as all required data fields and any optional data fields, or CT images, etc. The data requirement display area also provides an upload guide for each data type, including file format requirements, necessary metadata, and how to correctly fill in or select information, etc., to ensure the correctness and completeness of the input data upload.

[0139] In S6, participants upload relevant input data to the platform through their personal terminal devices. Once the upload is successful, the platform verifies the integrity and correctness of the uploaded data, including checking the file format, size, type, and data consistency and accuracy. If an error is found, the platform guides the participants to re-upload in the correct way. When all input data are collected, they are given to the designated model, which processes the input data and generates a diagnostic report.

[0140] Next, an example is given for the specified model:

[0141] If the specified model is the MiniGPT-Med model, the input data can be a variety of radiological images including X-rays, CT scans, and MRIs, and the output includes a diagnostic report of the probability of acute myocardial infarction. The MiniGPT-Med model includes a visual encoder, a linear projection layer, and a large language model. Among them, the visual encoder uses a pre-trained EVA visual encoder to convert the input radiological image into visual semantic features. During the training process, the parameters of the visual encoder remain unchanged (frozen state). The linear projection layer connects the features output by the visual encoder into a single visual tag and maps it to the feature space of the large language model. The large language model uses LLaMA2-chat as the main language model to process the projected visual tags and generate output in combination with text instructions to obtain a detailed diagnostic report. In addition to the probability of acute myocardial infarction, the diagnostic report also includes the location, size, and other related descriptions of the lesion.

[0142] If the specified model is an ischemic stroke diagnosis model based on LSTM multi-feature combination, the input data are the patient's ultrasound indicators, biochemical examination indicators and classification results (i.e., basic information set, symptom information set and diagnosis information set), and the output is a diagnosis report containing the results of stroke.

[0143] The ischemic stroke diagnosis model based on LSTM multi-feature combination includes an input layer, a feature extraction layer, a self-attention mechanism, a classification layer, and an output layer. The input layer consists of preprocessed ultrasound indicators, biochemical examination indicators, and classification results. These data are preprocessed, such as normalization, unique hot encoding, etc., to meet the input requirements of the model. The feature extraction layer learns feature information through three bidirectional LSTM feature extraction submodules built on the LSTM model. These submodules correspond to ultrasound feature extraction, biochemical examination feature extraction, and basic information feature extraction. Each submodule contains LSTM units, which can process sequence data and capture important patterns in time series. On the basis of the feature extraction layer, the self-attention mechanism is added to learn the correlation between features and assign weights. The self-attention mechanism can enhance the model's understanding of the relationship between features and improve classification performance. The classification layer classifies the feature vectors processed by the self-attention mechanism through one or two fully connected layers (Dense layers) and an activation function (such as Sigmoid or Softmax) to output the final diagnosis results later. The output layer is used to output the disease diagnosis results, that is, whether the patient suffers from ischemic stroke and the probability of suffering from the disease. This output can be a binary classification result (yes or no) or a probability value, which indicates the confidence of the model in the diagnosis result.

[0144] The operating principle of the ischemic stroke diagnosis model based on LSTM multi-feature combination is based on deep learning, especially the ability of LSTM network to process sequence data. LSTM network can capture long-term dependencies in time series through its unique gating mechanism (including forget gate, input gate and output gate). In this model, each LSTM unit updates its internal state according to the input data and the state of the previous unit. This state can be regarded as a "memory" that changes over time. In this way, the model can learn the changing patterns of different features over time and use these patterns for stroke risk assessment.

[0145] If the specified model is a trauma triage prediction model, the input data is vital signs data, including heart rate, blood pressure, respiratory rate, blood oxygen saturation, etc., and the output includes a diagnostic report of the severity of the injury.

[0146] The prediction model for triage of trauma patients is based on the gradient boosted decision tree (GBDT). The initialization of GBDT starts from the mean (for regression problems) or the majority of categories (for classification problems) of the target variable in the data set. The model uses a loss function (such as mean square error MSE or logarithmic loss) to measure the difference between the predicted value and the true value. In each iteration, GBDT calculates the negative gradient of the loss function with respect to the current predicted value, which is used as the training target for the next tree. The new tree aims to predict the residual of the previous round. GBDT trains decision trees iteratively, with each tree trained independently to correct the errors of the previous tree. The prediction results of each tree are accumulated to form the final output, usually by weighted averaging, where the weight of each tree is determined by its contribution to reducing the loss function. The GBDT model can also evaluate the contribution of each feature to the prediction, thereby providing insights for clinical decision-making. By integrating multiple trees, GBDT improves the generalization ability of the model, reduces the risk of overfitting, and finally outputs the prediction results of the injury classification of trauma patients, such as minor injuries, moderate injuries, severe injuries, and critical injuries.

[0147] If the specified model is a prediction model for pregnancy complications in critically ill pregnant women, the input data is clinical data, including age, body mass index (BMI), gestational age, blood routine, liver function, kidney function, etc., and the output is a diagnostic report containing the risk assessment results of pregnancy complications in critically ill pregnant women.

[0148] The prediction model for pregnancy complications in critically ill pregnant women includes a data preprocessing layer, a feature extraction layer, and a prediction layer. The data preprocessing layer cleans, standardizes, and selects features for the input data, handles missing values ​​and outliers, and ensures data quality. The feature extraction layer uses statistical methods or machine learning algorithms to identify and extract variables that best represent data features. Specifically, principal component analysis (PCA) can be used for dimensionality reduction to reduce the number of features and extract the most influential principal components for classification. At the same time, statistical features (such as mean, standard deviation) and text features (such as TF-IDF) can be used to fully capture the feature information of the data. The prediction layer uses random forests or support vector machines (SVMs). Random forests improve the accuracy and robustness of predictions by constructing multiple decision trees and voting, while SVMs achieve classification by finding support vectors that maximize the margin. These models can be used alone or in combination to improve prediction performance. For example, you can first use random forest for feature selection and extraction, and then use SVM for classification, or combine the prediction results of the two to obtain a more accurate prediction. Through this machine learning method, the prediction layer uses the features obtained from the feature extraction layer to obtain the possible pregnancy complications and their probability of occurrence for critically ill pregnant women, and organize them into a diagnostic report output to provide doctors with scientific diagnostic support and treatment recommendations.

[0149] If the specified model is a neonatal cold diagnosis model, the input data is the patient's medical record text, and the output contains a diagnostic report of the neonatal cold stage assessment results. The neonatal cold diagnosis model specifically uses the Transformer model, including an input embedding layer, a positional encoding layer, a multi-head self-attention mechanism, a feed-forward neural network, residual connections and layer normalization, and a decoder. Among them, the input embedding layer converts the input medical record text into a vector representation (word embedding), and uses a pre-trained word embedding model (such as Word2Vec, GloVe) to map each word to a vector sequence of fixed dimensions. The position encoding layer generates position encoding using sine and cosine functions, and adds position encoding to each position in the vector sequence, so that the model can understand the order of words. The multi-head self-attention mechanism passes the input sequence through multiple attention heads, each head independently calculates the attention weight, and then splices and linearly transforms the outputs of these heads, allowing the model to consider the information of the entire input sequence when processing each word. The feedforward neural network consists of two linear transformation layers with a ReLU activation function in the middle to perform a nonlinear transformation on the output of the self-attention layer. Residual connections and layer normalization add input to the output of each sublayer (self-attention and feedforward network), followed by layer normalization, to help the model train deeper networks and prevent gradients from vanishing or exploding. The decoder is similar to the encoder, but uses self-attention and encoder-decoder attention mechanisms to generate each word and generate an output text sequence (diagnosis report), which contains an assessment of the patient's cold stage, such as early stage, acute stage, peak stage, remission stage, and recovery stage.

[0150] If the specified model is a residual vision Mamba layer model, the input data is a 3D medical image (such as a CT scan or MRI scan), and the output is a diagnosis report containing 3D spine segmentation results. The residual vision Mamba layer model includes a residual vision Mamba layer and a shape prior module.

[0151] The residual visual Mamba layer includes residual connection, Mamba layer and visual feature extraction unit. The residual connection is used to allow information to directly skip the preset layer, which can be one or more, and is helpful for training deeper networks. For example, when processing spinal images, the residual connection can retain detailed information, such as the fine structure of the vertebra. By directly passing these detailed information to the deeper layers of the network, the residual connection helps to alleviate the problem of gradient disappearance, so that the network can be more effectively back-propagated and trained; the Mamba layer is a new type of sequence modeling architecture, which combines the spatial modeling capabilities of CNN and the temporal modeling capabilities of RNN. It is used to process sequence data and capture the relationship between 3D medical image slices, that is, in 3D medical images, it can be regarded as processing the relationship between different slices. For example, when processing spinal images, the Mamba layer can capture the spatial relationship between vertebrae, such as the gradual change from cervical to lumbar spine; the visual feature extraction unit is used to extract spatial features, that is, to extract spatial features in 3D medical images using operations such as 3D convolution.

[0152] The shape prior module includes a statistical shape model, an anatomical constraint unit, and a feature enhancement unit. The statistical shape model is a model that describes the shape change of the spine. By collecting a large amount of medical image data, these images are accurately annotated, and the anatomical structures of interest in the images are marked, such as the vertebrae of the spine. Then the structures in the annotated images are converted into a mathematically processable form, usually a set of coordinate points, which define the contour or surface of the structure. These coordinate points are used to calculate the relative position and distance between them, and a shape vector is constructed. The shape vector is statistically analyzed to find the main patterns and characteristics of the shape change. The main shape change patterns are extracted from the shape data using principal component analysis (PCA) or other statistical methods. Based on this, a statistical shape model is constructed to describe the natural range and trend of the spinal shape. When performing image segmentation or other image processing tasks, the statistical shape model is used as prior knowledge to help the residual visual Mamba layer model better understand and predict the shape of the target structure, thereby improving the accuracy and robustness of the segmentation. In addition, when the image data is incomplete or of low quality, the statistical shape model can be used to predict and restore missing or blurred structural parts. The anatomical constraint unit is used to integrate anatomical knowledge into the residual vision Mamba layer model, that is, to encode anatomical prior knowledge, such as the number and position of vertebrae, into the parameters of the residual vision Mamba layer model so that the model can take this knowledge into account during segmentation.

[0153] The feature enhancement unit is used to add specific network layers or loss functions to the residual vision Mamba layer model, encouraging the model to learn representations that conform to anatomical features. This enables the residual vision Mamba layer model to not only identify the anatomical structures in the images when processing 3D medical images, but also to accurately segment and classify these structures, which helps to improve the performance and accuracy of the model in medical image analysis tasks.

[0154] Based on the above, the residual vision Mamba layer model is a deep learning architecture for 3D spine segmentation, which combines the advantages of residual network (ResNet) and Mamba model, as well as shape prior technology to improve the accuracy and efficiency of segmentation. Among them, the design of residual vision combined with Mamba layer can integrate spatial and temporal information processing capabilities. In the code, the Mamba layer with residual connection is implemented using the `ResidualMambaBlock` class, which helps the model learn deeper features. This advanced AI model can improve the diagnostic ability of the platform, especially when dealing with complex 3D medical images, and the sequence modeling ability of the Mamba layer helps to reduce the interference of image noise. On this basis, the shape prior is integrated through the `shape_prior_loss` function, which can encode the anatomical features of the spine into the model, and use the pre-trained shape model to constrain the segmentation results. The shape prior can help the model maintain the anatomical features of the spine during the segmentation process, improve the accuracy of segmentation, and significantly improve the diagnostic accuracy of the platform, especially when dealing with organs with complex shapes, so that the model can better handle the anatomical differences of different patients.

[0155] When using the residual vision Mamba layer model, its input data, that is, the 3D medical image, needs to be preprocessed, including standardization and enhancement processing. Standardization refers to adjusting the brightness of the 3D medical image to a uniform range (such as between 0 and 1), resampling the 3D medical image to a uniform voxel size (such as a resolution of 1mm x 1mm x 1mm), and achieving brightness normalization and size unification; enhancement processing refers to using techniques such as histogram equalization to enhance contrast, and using methods such as Gaussian filtering to reduce noise. The preprocessed input data is input into the residual vision Mamba layer model to obtain the 3D spine segmentation result, which is then organized into a diagnosis report and output.

[0156] During the training phase of the residual visual Mamba layer model, a large number of 3D spine CT or MRI image samples are collected, preprocessed, and then the sample diversity is increased through operations such as rotation, flipping, and scaling. These image samples are annotated by professional doctors, and an image dataset is constructed based on the annotated image samples. Combined with segmentation accuracy and anatomical rationality, a multi-objective loss function is set, and hyperparameters such as learning rate scheduling and batch size are preset. The residual visual Mamba layer model is trained based on the multi-objective loss function, hyperparameters, and image dataset, and the cross-validation method is used to evaluate the model performance and optimize the model. For example, an image dataset is constructed using spinal CT scans of 1,000 different patients. Each scan has detailed segmentation annotations of the vertebrae and intervertebral discs. The model will repeatedly learn these data and gradually improve its segmentation accuracy.

[0157] The post-processing method is used to continue to optimize the residual visual Mamba layer model to refine the segmentation results. Specifically, the statistical shape model is used to adjust the unreasonable segmentation boundaries, complete the shape correction, add topological constraints in the image segmentation process to ensure the connectivity and integrity of the segmentation results, complete the topological correction, and use the smoothing algorithm to remove the jagged artifacts of the segmentation edge to achieve smooth processing. For example, if the original segmentation result shows that a vertebra is abnormally large or small, the introduction of the post-processing method can adjust it to a more reasonable size, and at the same time, ensure that all vertebrae are continuous, without unexpected "gaps", so that the model can more accurately identify the spinal structure.

[0158] By using the residual vision Mamba layer model, the quality of remote orthopedic or neurosurgery consultations can be significantly improved. Doctors can discuss based on more accurate 3D spine segmentation results, providing objective and accurate auxiliary information for doctors to make more accurate diagnosis and treatment decisions, thereby improving the efficiency and accuracy of diagnosis. The introduction of this residual vision Mamba layer model not only forms an organic combination with the platform described above, further improving the platform's diagnostic capabilities and medical service quality, but also enables similar technologies to be applied to other complex anatomical structure segmentations to meet more complex and professional medical needs.

[0159] Optionally, multiple feature extraction models based on weakly supervised learning mechanisms can be set up for different pathological diagnoses. Specifically, pre-trained large-scale visual models are used as base models, such as ResNet, VisionTransformer (ViT) or CLIP, which are pre-trained on a large number of pathological slice images and have strong feature extraction capabilities. In order to adapt to the characteristics of pathological slice images, the main structure of the base model will be retained through transfer learning, and the top layer will be fine-tuned or an adaptation layer will be added.

[0160] During the training process, a weakly supervised learning mechanism is used to use limited, coarse-grained annotation information, such as the diagnosis results in the pathology report, as training labels without the need for precise pixel-level annotation. In addition, a multiple instance learning (MIL) strategy is used to treat the entire pathology slice image as a "bag" containing multiple small image blocks (instances), and the feature extraction model needs to learn to extract key features from these instances to make an overall diagnosis.

[0161] In order to improve the accuracy of feature extraction, an attention mechanism is added to the feature extraction model to help the feature extraction model focus on the key areas in the pathological slice image. At the same time, a feature aggregation method is used to synthesize the features extracted from multiple image blocks into a global representation for final classification or diagnosis. In order to enhance the interpretability of the feature extraction model, explainable AI techniques such as class activation mapping (CAM) are used to visualize the areas that the feature extraction model focuses on and assist doctors in understanding the basis for AI decisions. This design enables the AI-assisted diagnosis system to not only provide accurate diagnosis results, but also help doctors understand the decision-making process of AI, thereby improving the reliability of diagnosis and the trust of doctors.

[0162] If the specified model is a breast cancer diagnosis model, the input data is a breast tissue pathology slice image, and the output is a diagnosis report containing a breast pathology heat map, a benign diagnosis result, or a malignant diagnosis result.

[0163] To build a breast cancer diagnosis model, we first prepare the data by collecting a large number of breast tissue pathology slice images and annotating each image with a benign or malignant diagnosis result without detailed region annotation. Then we build the model, select the pre-trained Vision Transformer (ViT) as the base model of the breast cancer diagnosis model, and add a multi-head attention layer and a classification head on top of it to enhance the model's ability to extract pathological image features. During the training process, each annotated breast tissue pathology slice image is cut into multiple small blocks (e.g., 256x256 pixels), and a multi-instance learning method is used to train these small blocks as a bag, so that the model can learn to extract key features from these small blocks and focus on important areas in the image through the attention mechanism. During the inference process, a new breast tissue pathology slice image is input, and the breast cancer diagnosis model divides it into multiple image blocks, then extracts the features of all image blocks and aggregates them through the attention mechanism, and finally outputs a benign or malignant diagnosis result. At the same time, a breast pathology heat map is generated. The heat map shows areas with high diagnostic value, that is, areas that the model believes may contain lesions. This area may show the presence of tumors, abnormal cell proliferation or other pathological changes, which helps doctors quickly identify suspicious areas.

[0164] The above technology makes full use of limited annotation information and combines it with a powerful pre-trained model to provide telemedicine platforms with efficient and accurate pathology image analysis capabilities. By integrating the breast cancer diagnosis model into the platform, multiple remote pathologists can not only view the model analysis results and original images at the same time when making relevant diagnoses, but the generated breast pathology heat map can also guide doctors to quickly pay attention to suspicious areas, thereby improving diagnostic efficiency and accuracy, and realizing the automation and intelligence of pathology image analysis. The doctor's feedback and correction suggestions are recorded for regular updates and optimization of the model, and the continuous learning mechanism is deployed at any time, and the performance of the model will continue to improve.

[0165] If the specified model is a lung cancer detection model, the input data is a lung CT scan image, and the output is a diagnosis report containing a lung thermal map, a positive diagnosis result, or a negative diagnosis result.

[0166] To build a lung cancer detection model, a certain number (e.g., 1,000) of lung CT scan images are first collected, including confirmed lung cancer cases and normal controls. These images are preprocessed, including grayscale, denoising, size standardization, etc., for subsequent processing. The preprocessed CT scan images are transformed at multiple levels (e.g., 3-level or 4-level decomposition) using 2D discrete wavelet transform (DWT), while retaining the global structural information and local details of the image. Specifically, Daubechies or Haar wavelets can be used to retain the edge and texture features of the image, which is crucial for identifying lung lesions, especially distinguishing normal lung tissue from potential cancer lesions. Through wavelet transform, the image can be decomposed into components of different frequencies, and the extracted wavelet coefficients (i.e., approximate, horizontal, vertical, and diagonal detail coefficients) contain the multi-scale features of the image. Statistical features such as mean, variance, skewness, kurtosis, etc. are further extracted from these wavelet coefficients. These statistical features will be used as input to the neural network to form a high-dimensional feature space to enhance detail and texture information.

[0167] Construct an MLP model with multiple (e.g. 3 to 5) hidden layers. The number of nodes in the input layer is determined by the number of extracted wavelet coefficient features. The output layer is a binary classification (cancer / normal). In each hidden layer, a nonlinear activation function, such as ReLU (Rectified Linear Unit) or LeakyReLU, is introduced to increase the expressive power of the model. The nonlinear activation function can introduce the mapping of complex functions and help the network learn more complex feature representations. The number of neurons in these hidden layers is gradually reduced from 512 to 64, and the last is a binary classification output layer. To prevent overfitting, a Dropout layer is added after each hidden layer. The Dropout layer randomly "discards" some neurons during training. The dropout rate can be set between 0.3 and 0.5 to enhance the generalization ability of the model. In this way, a multi-layer perceptron (MLP) architecture is constructed. The swarm intelligence optimization algorithm (Dragonfly algorithm, DA) is combined with the MLP to simulate the group behavior of dragonflies, including separation (avoiding collision), alignment (synchronizing with neighboring individuals), cohesion (moving to the center of the group) and foraging (moving to the food source). DA is used to optimize the hyperparameters of MLP, including the number of layers, the number of neurons in each layer, and the learning rate. For example, the number of layers is [2-6], the number of neurons in each layer is [32-256], and the learning rate is [0.0001-0.1]. Model-based hyperparameter optimization methods, such as Bayesian optimization, are used to find the optimal network structure and hyperparameter configuration. During the optimization process, each "dragonfly" represents a set of hyperparameter configurations. The DA algorithm iteratively updates these configurations and gradually improves network performance. For example, DA can simultaneously optimize the learning rate (ranging from 1e-5 to 1e-2), the number of hidden layers (2 to 10 layers), the number of neurons in each layer (32 to 1024), etc. The optimization process usually runs hundreds of iterations, each time evaluating different hyperparameter combinations, and finally selecting the configuration that performs best on the validation set.

[0168] The DA algorithm is used to assist in feature selection, and the optimization ability of DA is used to determine the most relevant wavelet coefficients. This process can be viewed as a binary optimization problem, in which each feature will be selected or excluded. The DA algorithm is used to search the high-dimensional feature space, evaluate the performance of different feature combinations, and then filter features through preset thresholds, such as selecting the top 20% features in performance. This not only reduces the input dimension, but also improves the interpretability of the model, making it easier to analyze which types of wavelet features (such as details of a specific scale or direction) are most important for lung cancer detection.

[0169] By extracting the features of CT scan images through the above wavelet transform and optimizing the features using the DA algorithm, multiple WMLP (wavelet multilayer perceptron) models can be constructed. Each WMLP model focuses on a different subset of image features. The purpose of this is to analyze images from multiple angles and improve the model's ability to identify lung cancer. In order to further improve the robustness and accuracy of the model, ensemble learning technology is used for training. Specifically, multiple WMLP models are trained, and each model uses 80% of the randomly selected training data and a slightly different feature subset. This method imitates the idea of ​​bagging, and reduces overfitting by combining multiple models and improves the generalization ability of the model. In the prediction stage, a voting mechanism is used to integrate the prediction results of these models. Specifically, soft voting can be used, that is, the final result is obtained by averaging the prediction probability output by each model; or hard voting can be used, that is, the final classification decision is determined by majority voting. In addition to the simple voting mechanism, the Stacking method can also be used to learn how to best combine the predictions of each basic model by training a meta-classifier. This meta-classifier is trained on the prediction results of all basic models and can learn the advantages and disadvantages of the predictions of each model, thereby improving the overall prediction performance. During the training process, k-fold cross-validation is used to evaluate the performance of the model, with accuracy, sensitivity, and specificity as the main evaluation indicators, which ensures that the model not only performs well on the training set, but also maintains a high accuracy on unseen data.

[0170] The model's raw predictions are post-processed to improve detection accuracy, including applying morphological operations such as opening and closing operations to smooth the predicted lung cancer region boundaries and remove small isolated prediction regions. A conditional random field (CRF) model is used to consider the spatial relationship between pixels, ensuring that the predictions are more coherent and biologically meaningful in the image.

[0171] In post-processing, clinical knowledge is combined to filter out anatomically impossible false positive results. For example, predictions located in areas such as major blood vessels or trachea are excluded because lung cancer is unlikely to occur in these places based on anatomical knowledge. A minimum volume threshold (such as 50mm³) is set to filter out detection results that are too small because these are usually false positives. In order to better understand the model's predictions, SHAP (SHapley Additive exPlanations) values ​​are used to analyze the contribution of each feature to the prediction, which helps to identify which wavelet coefficients or other image features are most important to the model's decision. In addition, a lung thermogram is generated to intuitively display the suspicious areas considered by the model, which helps doctors quickly locate potential lung cancer areas and improves the transparency and credibility of the model.

[0172] The model's predictions were compared with radiologists' diagnoses for clinical validation. This step not only verified the model's accuracy, but also helped analyze the model's strengths, such as rapid screening capabilities, and limitations, such as its ability to identify certain rare lesions. Through this comparative analysis, the model was further adjusted and optimized to improve its effectiveness in actual clinical applications.

[0173] Through these post-processing and verification steps, the high performance and clinical applicability of the lung cancer detection model in lung cancer detection are ensured, providing doctors with a powerful auxiliary diagnostic tool to identify potential lung cancer lesions more quickly and accurately.

[0174] In S7, the diagnostic report is clearly displayed in the user interface of the doctor's consultation room in the form of charts, lists or summaries. After reviewing and verifying the diagnostic report, the doctor can choose to share the diagnostic report with other doctors or patients.

[0175] In S8, one or more doctors are allowed to fill in the checklist template according to the results of the consultation discussion. After filling in, the doctor electronically signs the checklist template to generate a consultation checklist to confirm the accuracy and legality of its content. The checklist template is provided by the consultation room or uploaded by any participant. The consultation checklist includes prescription data, examination results, treatment plans, etc.

[0176] Both the diagnostic report and the consultation list are available for download by patients or provided to them via email or a dedicated website.

[0177] When any participant initiates an end request, the platform closes the corresponding consultation room. The platform notifies other participants that someone has initiated an end request and provides options for other participants to choose whether to agree to end the consultation. If no one agrees, only the consultation room of the party that initiated the end request and the party that agrees to end the consultation will be closed. When other participants agree or all participants exit the consultation room, the platform closes the consultation room and the consultation ends.

[0178] Under different network conditions, doctors and patients may encounter network delays and instability when participating in remote consultations. Especially when doctors or patients with poor network environments join, the quality of video calls may decline, affecting the continuity of consultations. This will not only reduce the consultation experience and communication efficiency, but may also lead to misdiagnosis or reduced diagnostic efficiency, ultimately affecting the treatment effect of patients. In addition, frequent consultation interruptions or quality declines will also reduce the trust of doctors and medical institutions in the platform, affecting the user satisfaction of the platform.

[0179] To solve network problems, WebRTC (Web Real-Time Communication) technology is used to achieve low-latency, high-quality audio and video transmission to ensure the smoothness of real-time high-definition video calls between participants. At the same time, CDN (Content Distribution Network) technology is integrated to optimize network transmission and ensure the stability of cross-regional consultations. WebSocket (Web Real-Time Communication, real-time message push) technology is used to ensure instant synchronization of discussion content, and cloud storage technology such as Alibaba Cloud OSS is used to save consultation records and diagnostic opinions to support subsequent review and analysis. In this way, multiple participants can participate in the consultation, record the discussion content, and form collaborative diagnostic opinions.

[0180] The intelligent routing algorithm is used to select the optimal data transmission path according to the real-time network conditions, reduce the occurrence of delays and packet loss, improve the stability and efficiency of data transmission, and reduce call interruptions and image and sound quality degradation caused by network changes. The platform can automatically detect changes in network bandwidth and dynamically adjust the bit rate and resolution of video encoding to adapt to current network conditions, reduce video quality when bandwidth is insufficient, ensure call continuity, and improve video quality when bandwidth is restored to enhance user experience. The integrated network quality monitoring tool monitors key indicators such as network delay and packet loss rate in real time, and provides fault diagnosis functions to help quickly locate network problems and provide data support for network optimization.

[0181] Reference Figure 4 The present disclosure provides a telemedicine online consultation system, comprising:

[0182] The identity authentication module 101 is used to perform identity authentication on all participants in response to a consultation request initiated by any participant, where the participant is a patient or a doctor;

[0183] A consultation room creation module 102 is used to create a consultation room for a participant who has passed identity authentication;

[0184] A model recommendation module 103, for recommending a plurality of available medical models to doctors through a consultation room;

[0185] A model specifying module 104, for determining a specified model in response to a doctor's specifying operation among a plurality of available medical models;

[0186] A data display module 105, for displaying the input data required for the specified model in the consultation room;

[0187] The data collection module 106 is used to collect input data in response to the upload operation of the participant, and obtain a diagnostic report based on the input data and a specified model;

[0188] A report display module 107, used to display the diagnosis report in the consultation room for reference by the participants;

[0189] The consultation room closing module 108 is used to close the consultation room in response to a termination request initiated by any participant.

[0190] The various variations and specific examples of the telemedicine online consultation method provided above are also applicable to the telemedicine online consultation system provided in the present disclosure. Through the above detailed description of the telemedicine online consultation method, those skilled in the art can clearly know the implementation method of the telemedicine online consultation system. For the sake of brevity of the specification, it will not be described in detail here.

[0191] The computer device according to the embodiment of the present disclosure includes a memory and a processor. The memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.

[0192] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory, so that the computer device performs all or part of the steps of the telemedicine online consultation method of each embodiment of the present disclosure.

[0193] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0194] like Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure is shown, which is a schematic diagram of the structure of a computer device suitable for implementing the embodiment of the present disclosure. Figure 5 The computer device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0195] like Figure 5As shown, the computer device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer device are also stored. The processor, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0196] Typically, the following devices can be connected to the I / O interface: input devices such as sensors or visual information acquisition devices; output devices such as display screens; storage devices such as tapes, hard disks, etc.; and communication devices. The communication device can allow the computer device to communicate with other devices (such as edge computing devices) wirelessly or by wire to exchange data. Figure 5 A computer device having various devices is shown, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0197] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the telemedicine online consultation method of the embodiment of the present disclosure are executed.

[0198] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0199] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the telemedicine online consultation method of each embodiment of the present disclosure are executed.

[0200] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0201] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0202] The basic principles of the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present disclosure. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit the present disclosure to the necessity of adopting the above specific details to be implemented.

[0203] In the present disclosure, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0204] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0205] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0206] Various changes, substitutions, and modifications of the techniques described herein may be made without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of the present disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and actions described above. Currently existing or later to be developed processes, machines, manufactures, compositions of events, means, methods, or actions that perform substantially the same functions or achieve substantially the same results as the corresponding aspects described herein may be utilized. Thus, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or actions within their scope.

[0207] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0208] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A telemedicine online consultation method, characterized in that: include: In response to a consultation request initiated by any participant, identity authentication is performed on all participants, including patients and doctors; Create a consultation room for participants who have passed identity authentication; Recommend multiple available medical models to doctors through the consultation room; In response to a doctor's designated operation among a plurality of available medical models, determining a designated model; Presenting input data required for the specified model in the consultation room; In response to an upload operation by a participant, the input data is collected, and a diagnosis report is obtained based on the input data and the specified model; Displaying the diagnostic report in the consultation room for reference by the participants; In response to a termination request initiated by any participant, the consultation room is closed.

2. The telemedicine online consultation method according to claim 1, characterized in that: The consultation room recommends multiple available medical models to doctors, including: Determine whether the physician conducts a keyword search; If yes, obtaining the keywords input by the doctor, and filtering out the multiple available medical models in the model library based on the keywords; If not, obtaining the patient's medical record text, and filtering out the multiple available medical models in the model library based on the medical record text and the label library; The plurality of available medical models are displayed in a doctor's consultation room.

3. The telemedicine online consultation method according to claim 2, characterized in that: The label library includes a first label set, a second label set and a third label set for each medical model; Preprocessing the medical record text; Perform entity recognition and entity classification on the preprocessed medical record text to obtain the patient's basic information set, symptom information set, and diagnosis information set; Obtaining a first matching degree between the basic information set and the first tag set; Obtaining a second matching degree between the symptom information set and the second label set; Obtaining a third matching degree between the diagnostic information set and the third tag set; Performing weighted summation on the first matching degree, the second matching degree, and the third matching degree to obtain a degree of fit between the patient and each medical model; The plurality of available medical models are selected based on the degree of fit.

4. The telemedicine online consultation method according to claim 3, characterized in that: The calculation formula of the first matching degree is: PAa=f1+f2; Wherein, A is the number of the first label set; PAa is the first matching degree between the basic information set and the first label set numbered A; f1 is the first score; f2 is the second score; The first label set includes applicable age groups, applicable genders, age scores, and gender scores of the medical model; The basic information set includes the patient's real age and real gender; If the real age is within the applicable age range, determining that the first score is equal to the age score; If the real age is not within the applicable age range, the first score is determined to be equal to 0; If the real gender is the applicable gender, determining that the second score is equal to the gender score; If the actual gender is not the applicable gender, the second score is determined to be equal to 0.

5. The telemedicine online consultation method according to claim 3, characterized in that: The second label set includes a plurality of symptom labels and a first label threshold corresponding to each symptom label; Set the first parameter to be initially 0; Traversing all disease symptom information in the symptom information set; Calculate the similarity between the current disease symptom information and the symptom label; If the similarity reaches the corresponding first tag threshold, the first parameter is increased by 1; If the similarity does not reach the corresponding first tag threshold, the first parameter remains unchanged; After all the disease symptom information is traversed, the second matching degree is obtained based on the first parameter and the total number of symptom labels in the second label set; The calculation formula of the second matching degree is: PBb = S1 / S2; Among them, B is the number of the second label set; PBb is the second matching degree between the symptom information set and the second label set numbered B; S1 is the first parameter; S2 is the total number of symptom labels.

6. The telemedicine online consultation method according to claim 3, characterized in that: The third label set includes a plurality of diagnostic labels and a second label threshold corresponding to each diagnostic label; Set a second parameter initially set to 0, and traverse all the diagnosis result information in the diagnosis information set; Calculating the similarity between the current diagnosis result information and the diagnosis label; If the similarity reaches the corresponding second tag threshold, the second parameter is increased by 1; If the similarity does not reach the corresponding second tag threshold, the second parameter remains unchanged; After all the diagnosis result information is traversed, the third matching degree is obtained based on the second parameter and the total number of diagnosis tags in the third tag set; The calculation formula of the third matching degree is: PCc=S3 / S4; Among them, C is the number of the second label set; PCc is the third matching degree between the diagnostic information set and the third label set numbered C; S3 is the second parameter; S4 is the total number of diagnostic labels.

7. The telemedicine online consultation method according to claim 1, characterized in that: If the designated model is the residual visual Mamba layer model, the input data is a 3D medical image, and the output is a diagnostic report including a 3D spine segmentation result.

8. A computer device, characterized in that: The computer device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the remote medical online consultation method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the remote medical online consultation method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images