An endocrine disease artificial intelligence auxiliary diagnosis and treatment system based on an internet hospital

Abstract

The present application relates to the technical field of medical data, in particular to an endocrine disease artificial intelligence auxiliary diagnosis and treatment system based on an internet hospital, comprising: a data acquisition module for accessing multi-source data of wearable devices, home detection devices, patient self-reporting, offline medical records and third-party platforms; a time sequence alignment module, in use, the present application is beneficial to breaking through the limitations of the prior art relying on static or single data sources by efficiently fusing and processing multi-channel, fragmented, real-time and cross-platform heterogeneous data, improving data timeliness, providing accurate data support for rapid judgment of emergency conditions, avoiding diagnosis and treatment lag, adjusting personalized treatment plans based on real-time data dynamic diagnosis and treatment, facilitating the change of static scheme recommendation mode, improving the response speed of remote diagnosis and treatment, facilitating the solution of the problem of lag in handling emergency conditions, breaking through the resource barrier through online and offline resource linkage, making online decision-making smoothly land, and being beneficial to improving the auxiliary diagnosis and treatment effect.

Description

Technical Field

[0001] This invention relates to the field of medical data technology, specifically to an artificial intelligence-assisted diagnosis and treatment system for endocrine diseases based on an internet hospital. Background Technology

[0002] Endocrine disorders are a group of complex diseases caused by abnormal secretion of hormones in the body (too much, too little, or resistance). They include diabetes, thyroid diseases, pituitary tumors, adrenal diseases, gonadal diseases, etc. Their diagnosis and treatment require a combination of multi-dimensional information such as symptoms, signs, laboratory tests (hormone levels), and imaging data. Moreover, there are great individual differences and significant dynamic changes in the course of the disease.

[0003] The patent with publication number CN111489820A describes in its specification that "This invention relates to an artificial intelligence-based assisted diagnosis and treatment system, comprising at least: a patient client (1), a database (3), a server (4), and a medical client (2), wherein the patient client (1) is capable of recording the patient's place of residence information, the database (3) is capable of storing medication data for a specific patient in a medication-related manner, and the server (4) is configured to: statistically analyze the medication data of the specific patient for the same disease type, thereby obtaining the frequency and / or quantity of the same or similar drugs repeatedly taken by the specific patient within a set time period; if the server (4) determines, based on the analysis performed by its analysis module, that the frequency and / or quantity of the same or similar drugs repeatedly taken by the specific patient is greater than a set threshold, then..." "The reminder is issued by the medical client (2) and / or the patient client (1)". Although the above technology achieves the effect of simultaneously solving the two major pain points of decreased efficacy due to lack of medication data at the grassroots level and difficulty in diagnosing regional diseases by combining medication frequency analysis with regional patient data, there are many types of endocrine diseases, such as diabetes, thyroid diseases, pituitary diseases, etc., and their diagnosis and treatment are complex. At present, the artificial intelligence-assisted diagnosis and treatment system of Internet hospitals mostly relies on static data or a single data source, which cannot effectively process the heterogeneous data from multiple channels, fragmented, real-time and cross-platform, such as wearable devices, home testing devices, patient self-reporting, offline medical records and third-party platforms in Internet hospitals. This leads to a lag in the judgment and handling of emergencies in assisted diagnosis and treatment. At the same time, the lack of linkage management of online and offline resources leads to the general effect of assisted diagnosis and treatment.

[0004] In conclusion, developing an AI-assisted diagnosis and treatment system for endocrine diseases based on internet hospitals remains a critical issue that urgently needs to be addressed in the field of medical data technology. Summary of the Invention

[0005] The purpose of this invention is to address the challenges of existing technologies in treating a wide variety of endocrine diseases, such as diabetes, thyroid diseases, and pituitary diseases, which are complex to diagnose and treat. Currently, AI-assisted diagnosis and treatment systems in internet hospitals often rely on static data or single data sources, failing to effectively handle the diverse, fragmented, real-time, and cross-platform heterogeneous data from multiple channels, including wearable devices, home testing devices, patient self-reports, offline medical records, and third-party platforms. This results in delays in assessing and handling emergencies during assisted diagnosis and treatment. Furthermore, the lack of coordinated management of online and offline resources leads to generally poor results in assisted diagnosis and treatment.

[0006] To achieve the above objectives, the present invention provides an artificial intelligence-assisted diagnosis and treatment system for endocrine diseases based on an internet hospital, comprising:

[0007] The data acquisition module is used to access multi-source data from wearable devices, home testing devices, patient self-reports, offline medical records, and third-party platforms;

[0008] The time-series alignment module is used to align multi-source data based on timestamps using a high-precision time-series alignment algorithm and output aligned data.

[0009] The intelligent learning module is used to identify the correlation of endocrine disease data in the aligned data through a deep learning model, generate a physiological digital twin of the patient and update it in real time;

[0010] The dynamic diagnosis and treatment module is used to output personalized treatment plans based on the physiological digital twin combined with the knowledge graph;

[0011] The doctor-patient collaboration module is used to link online and offline resources with the database, providing doctors with a data visualization dashboard and pushing interactive diagnosis and treatment suggestions to patients.

[0012] Furthermore, the data acquisition module, used to access multi-source data from wearable devices, home testing devices, patient self-reports, offline medical records, and third-party platforms, includes the following operational procedures:

[0013] The data acquisition module employs an SSL / TLS encrypted transmission protocol and establishes a local caching mechanism, which creates a local cache pool. Each data stream is cached separately, and the cache scheduling strategy uses a priority-weighted sliding window mechanism, expressed as:

[0014]

[0015] In the formula, Indicates at time Time The status of the local cache pool corresponding to the path data. Indicates the size of the sliding window. Indicates the time from arrive Sum all the terms within this interval. Indicates at time Weighting coefficients for time data Indicates at time Time The raw data content of the road data Represents all moments Corresponding weight coefficients The sum is 1. It is a hyperparameter used to control the rate of weight decay. It is an exponentially decaying function. It is the normalization factor.

[0016] Furthermore, the time-series alignment module employs a high-precision time-series alignment algorithm to align multi-source data based on timestamps, and the operation process for outputting aligned data includes:

[0017] A high-precision time-series alignment algorithm is employed to address the issues present in the multi-source data. A heterogeneous data source device, data from different devices sampling frequency Generates a data sequence, expression:

[0018]

[0019] In the formula, Indicates the first The original data sequence from a heterogeneous data source, Indicates the first The device in the The timestamp of each sampling moment Indicates the first The device in the The observations at each sampling time point, Indicates the first The total number of samples from each device is used to reconstruct the data for each device using a spline function interpolation method with perturbation compensation, based on a globally unified timestamp. The expression is:

[0020]

[0021] In the formula, Indicates the first Each device at a globally unified time Interpolated reconstructed values ​​at the location, Represents a globally unified timestamp. Indicates the measurement of the original sampling points Global unified time The contribution of the reconstructed value, It is the Gaussian kernel part. It is the disturbance compensation coefficient. It is in a globally unified time The cubic spline function at the location.

[0022] Furthermore, the time-series alignment module employs a high-precision time-series alignment algorithm to align multi-source data based on timestamps, and the operation process for outputting aligned data includes:

[0023] The alignment process involves constructing a global alignment objective function, and aligning the data on a globally unified timestamp. Minimize the asynchronous cost of each interpolation sequence, expressed as:

[0024]

[0025] In the formula, It is the loss function for global alignment. This indicates traversing the globally unified timestamp set. Each globally unified timestamp , This indicates that all data pairs are iterated over. The weight representation of the data source pair is the first... and the The importance of a data source Indicates the first Data sources in time Interpolated data at the point, It is the square of the Euclidean distance, outputting aligned data. The expression is:

[0026]

[0027] In the formula, This represents the aligned data set. Represents a globally unified timestamp, indicating the first to the last timestamp. Data sources in time Interpolated data at the point, Indicates the total number of data sources. This indicates traversing the globally unified timestamp set. Each time point in .

[0028] Furthermore, the intelligent learning module, used to identify the correlation of endocrine disease data in the aligned data through a deep learning model, generates a physiological digital twin of the patient, and updates it in real time, includes the following operational procedures:

[0029] The deep learning model is based on an attention mechanism, expressed as:

[0030]

[0031] In the formula, express Time of the first Hidden state vectors of each channel / modality yes Activation function express Time of the first Raw input data for each channel / modality Represents the embedding weight matrix. This represents the embedding bias vector. express Time of the first Attention output vectors for each channel / modality This indicates that the splicing operation will The outputs of each attention head are spliced ​​together. Indicates the first The value weight matrix of each attention point will hide the state. Convert to a value vector. Indicates attention weights Weighted summation of value vectors, This indicates the total number of heads of attention. express Time of the first A position in the attention. With position Attention weights were assigned, and the deep learning model was trained using the Apache Hadoop and Spark frameworks, and then optimized using cross-validation.

[0032] Furthermore, the intelligent learning module, used to identify the correlation of endocrine disease data in the aligned data through a deep learning model, generates a physiological digital twin of the patient, and updates it in real time, includes the following operational procedures:

[0033] The process generates and updates the patient's physiological digital twin in real time. The state vector of the physiological digital twin includes, but is not limited to, core states such as hormone secretion level and metabolic rate, and is generated by hidden state mapping, expressed as:

[0034]

[0035] In the formula, express The state vector of the physiological digital twin at any given time. This represents the state mapping weight matrix. express The hidden state vector at time step 1. This represents the state-mapped bias vector, which, upon new data input, fuses historical states with new observations using a Kalman filter. The expression is:

[0036]

[0037] In the formula, express Time's up The predicted state value at time 10:00. Represents the state transition matrix. express The state vector of the physiological digital twin at any given time. Represents the control matrix. yes Control vector at time, express Time's up The state prediction error covariance matrix at time 1. express The error covariance matrix of the state at time step. Represents the state transition matrix The transpose of the matrix, Represents the process noise covariance matrix. express Kalman gain at time step Represents the observation matrix. It is the inverse covariance matrix for calculating the observation noise. It is the observation noise covariance matrix. Indicating the integration of new observations The state of the physiological digital twin after constant correction. express New observations at time [time] Indicates based on the predicted state The derived theoretical observations are used to compare with new observations. Compare and calculate the correction amount. Indicating the integration of new observations The error covariance matrix after time correction, The identity matrix is ​​used for identity transformations in matrix operations.

[0038] Furthermore, the dynamic diagnosis and treatment module, used to output personalized treatment plans based on the physiological digital twin combined with a knowledge graph, includes the following operational procedures:

[0039] The dynamic diagnosis and treatment module employs a reinforcement learning model, loading the quarterly updated endocrine disease diagnosis and treatment knowledge graph in triplet form, performing semantic parsing and feature mapping, and generating a decision feature vector based on the physiological digital twin combined with the knowledge graph. The expression is:

[0040]

[0041] In the formula, express The decision feature vector generated at each time step, This indicates that the fusion weight matrix is ​​a learnable parameter used to perform a linear transformation on the concatenated vector. It is a knowledge graph feature fusion computation based on the attention mechanism. It is the total number of entities participating in the computation in the knowledge graph. The attention weight matrix is ​​a learnable parameter used to calculate the state vector of the physiological digital twin. With knowledge graph entity vectors The correlation weight between them Indicates the first Vectors of entities, It is the first time when traversing all entities. Entity vectors, Indicates calculation using an exponential function and The relevance score after attention matrix transformation This indicates that all entities and The correlation scores are summed for normalization. The attention weight representation is the first A knowledge graph entity relative to a physiological digital twin The importance of The vector concatenation operation represents concatenating the physiological digital twin's state vector with the feature vector obtained by fusing the knowledge graph through the attention mechanism. It is the learnable parameters of the fused bias vector and In conjunction with this, the vectors that have been spliced ​​and linearly transformed are offset and adjusted.

[0042] Furthermore, the dynamic diagnosis and treatment module, used to output personalized treatment plans based on the physiological digital twin combined with a knowledge graph, includes the following operational procedures:

[0043] The reinforcement learning model adopts an Actor-Critic architecture, consisting of a policy function and a value function. It outputs personalized treatment plans. The knowledge graph is updated quarterly, incorporating the latest research findings and clinical experience. The reward function of the reinforcement learning model focuses on treatment effectiveness indicators, including glycemic control achievement rate and complication rate. The expression is:

[0044]

[0045] In the formula, Indicates the first Instant rewards for each moment Weighting coefficient The reward weighting for achieving blood sugar control targets Control the penalty weighting for other adverse clinical events. The exponential function makes the reward exhibit a smooth decay / increase trend with blood sugar deviation. Indicates the first The actual blood glucose level at that moment, Indicates the target blood glucose level. A parameter representing tolerance to blood glucose fluctuations. This represents the square of the blood glucose deviation. Representing the Other adverse clinical events at the time include, but are not limited to, the risk of hypoglycemia and ketosis.

[0046] Furthermore, the doctor-patient collaboration module, which connects online and offline resources with the database, provides doctors with a data visualization dashboard and pushes interactive treatment suggestions to patients. The operational process includes:

[0047] The doctor-patient collaboration module includes a doctor's end and a patient's end. The doctor's end provides a data visualization dashboard that displays the physiological digital twin and treatment plan in real time, and supports manual adjustments by the doctor. (Expression:)

[0048]

[0049] In the formula, Indicates the current patient profile and the A portrait of a historical patient The similarity value between them This represents the current patient's profile vector. Indicates the first A vector image of a historical patient. Indicates to from arrive The items are accumulated. It is the total number of features in the patient profile. Indicates the first The weights of each feature, Indicates the current patient's number Features With historical patients No. Features The numerical product, Indicates the current patient profile The weighted norm, Indicates historical patients portrait The weighted norm.

[0050] Furthermore, the doctor-patient collaboration module, which connects online and offline resources with the database, provides doctors with a data visualization dashboard and pushes interactive treatment suggestions to patients. The operational process includes:

[0051] The interactive treatment suggestions pushed to the patient include, but are not limited to, animated demonstrations of insulin injection adjustment methods. The online and offline resource database is linked to calculate the fit with the patient profile, expressed as:

[0052]

[0053] In the formula, Indicates that for the first Individual treatment recommendations The fit calculation results Indicates the first One treatment recommendation to be evaluated. From arrive Summing up the following expressions, The total number of historical patient cases and related data used in the calculation. Indicates the current patient profile and the A portrait of a historical patient The similarity value between them This represents the current patient's profile vector. Indicates the first A vector image of a historical patient. The evaluation function represents the measure of treatment recommendations. Applied to the first A portrait corresponding to a historical patient The effect produced at that time, select The highest recommendation is pushed out in the form of animation. The online and offline resource linkage database includes, but is not limited to, the diagnosis and treatment resources, equipment, and expert information of offline medical institutions, as well as key information of doctor-patient communication recorded through natural language processing technology, which feeds back into the deep learning model.

[0054] Beneficial effects

[0055] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects:

[0056] When in use, this invention efficiently integrates and processes multi-channel, fragmented, real-time, and cross-platform heterogeneous data, which helps to overcome the limitations of existing technologies that rely on static or single data sources, improves data timeliness, provides accurate data support for rapid judgment of emergencies, avoids delays in diagnosis and treatment, and facilitates personalized treatment plan adjustments based on real-time data dynamic diagnosis and treatment. This makes it easier to change the static plan recommendation mode, improve the response speed of remote diagnosis and treatment, and solve the problem of delays in handling emergencies. By linking online and offline resources, it breaks down resource barriers, enables online decision-making to be implemented smoothly, and helps to improve the effectiveness of auxiliary diagnosis and treatment. Attached Figure Description

[0057] Figure 1 This is a system diagram of an artificial intelligence-assisted diagnosis and treatment system for endocrine diseases based on an internet hospital, according to the present invention. Detailed Implementation

[0058] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0059] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0060] The present invention will now be described in further detail with reference to the accompanying drawings:

[0061] Example:

[0062] like Figure 1 As shown, this invention provides an AI-assisted diagnosis and treatment system for endocrine diseases based on an internet hospital, comprising:

[0063] The data acquisition module is used to access multi-source data from wearable devices, home testing devices, patient self-reports, offline medical records, and third-party platforms;

[0064] Furthermore, the data acquisition module, used to access multi-source data from wearable devices, home testing devices, patient self-reports, offline medical records, and third-party platforms, includes the following operational procedures:

[0065] The data acquisition module employs an SSL / TLS encrypted transmission protocol and establishes a local caching mechanism, which creates a local cache pool. Each data stream is cached separately, and the cache scheduling strategy uses a priority-weighted sliding window mechanism, expressed as:

[0066]

[0067] In the formula, Indicates at time Time The status of the local cache pool corresponding to the path data. Indicates the size of the sliding window. Indicates the time from arrive Sum all the terms within this interval. Indicates at time Weighting coefficients for time data Indicates at time Time The raw data content of the road data Represents all moments Corresponding weight coefficients The sum is 1. It is a hyperparameter used to control the rate of weight decay. It is an exponentially decaying function. It is the normalization factor.

[0068] Specifically, the data acquisition module securely accesses multi-source health data and efficiently manages the data through a local caching mechanism. It employs SSL / TLS encryption protocols to ensure data transmission security and establishes a local cache pool for each data stream. The cache is scheduled based on a priority-weighted sliding window mechanism, which defines the data range through a sliding time window and assigns weights that decay over time to data at different times within the window (recent data has a higher weight). After normalization, the weighted sum is calculated to determine the cache pool status, prioritizing the retention of important data. This helps protect the privacy and compliance of data transmission throughout the process and mitigates data loss caused by network instability or device disconnection.

[0069] The time-series alignment module is used to align multi-source data based on timestamps using a high-precision time-series alignment algorithm and output aligned data.

[0070] Furthermore, the time-series alignment module employs a high-precision time-series alignment algorithm to align multi-source data based on timestamps, and the operation process for outputting aligned data includes:

[0071] A high-precision time-series alignment algorithm is employed to address the issues present in the multi-source data. A heterogeneous data source device, data from different devices sampling frequency Generates a data sequence, expression:

[0072]

[0073] In the formula, Indicates the first The original data sequence from a heterogeneous data source, Indicates the first The device in the The timestamp of each sampling moment Indicates the first The device in the The observations at each sampling time point, Indicates the first The total number of samples from each device is used to reconstruct the data for each device using a spline function interpolation method with perturbation compensation, based on a globally unified timestamp. The expression is:

[0074]

[0075] In the formula, Indicates the first Each device at a globally unified time Interpolated reconstructed values ​​at the location, Represents a globally unified timestamp. Indicates the measurement of the original sampling points Global unified time The contribution of the reconstructed value, It is the Gaussian kernel part. It is the disturbance compensation coefficient. It is in a globally unified time The cubic spline function at the location.

[0076] Furthermore, the time-series alignment module employs a high-precision time-series alignment algorithm to align multi-source data based on timestamps, and the operation process for outputting aligned data includes:

[0077] The alignment process involves constructing a global alignment objective function, and aligning the data on a globally unified timestamp. Minimize the asynchronous cost of each interpolation sequence, expressed as:

[0078]

[0079] In the formula, It is the loss function for global alignment. This indicates traversing the globally unified timestamp set. Each globally unified timestamp , This indicates that all data pairs are iterated over. The weight representation of the data source pair is the first... and the The importance of a data source Indicates the first Data sources in time Interpolated data at the point, It is the square of the Euclidean distance, outputting aligned data. The expression is:

[0080]

[0081] In the formula, This represents the aligned data set. Represents a globally unified timestamp, indicating the first to the last timestamp. Data sources in time Interpolated data at the point, Indicates the total number of data sources. This indicates traversing the globally unified timestamp set. Each time point in .

[0082] Specifically, the time-series alignment module uses a high-precision time-series alignment algorithm to perform unified time-dimensional processing on multi-source heterogeneous data, outputting aligned data that can be directly used for subsequent analysis. For the original data sequences from multiple heterogeneous data sources, based on a globally unified timestamp, the data is reconstructed using a combination of Gaussian kernels and cubic spline functions with perturbation compensation. Then, the asynchronous cost of multi-source data is minimized through a global alignment objective function, ultimately generating a unified time-axis aligned data set. This facilitates the avoidance of information loss or distortion, ensuring optimal consistency and time-series synchronization of multi-source data at the same time scale. This allows for flexible application in various medical wearables, sensors, and smart devices, supporting the precise fusion and processing of medical big data.

[0083] The intelligent learning module is used to identify the correlation of endocrine disease data in the aligned data through a deep learning model, generate a physiological digital twin of the patient and update it in real time;

[0084] Furthermore, the intelligent learning module, used to identify the correlation of endocrine disease data in the aligned data through a deep learning model, generates a physiological digital twin of the patient, and updates it in real time, includes the following operational procedures:

[0085] The deep learning model is based on an attention mechanism, expressed as:

[0086]

[0087] In the formula, express Time of the first Hidden state vectors of each channel / modality yes Activation function express Time of the first The raw input data for each channel / modality Represents the embedding weight matrix. This represents the embedding bias vector. express Time of the first Attention output vectors for each channel / modality This indicates that the splicing operation will The outputs of each attention head are spliced ​​together. Indicates the first The value weight matrix of each attention point will hide the state. Convert to a value vector. Indicates attention weights Weighted summation of value vectors, This indicates the total number of heads of attention. express Time of the first A position in the attention. With position Attention weights were assigned, and the deep learning model was trained using the Apache Hadoop and Spark frameworks, and then optimized using cross-validation.

[0088] Furthermore, the intelligent learning module, used to identify the correlation of endocrine disease data in the aligned data through a deep learning model, generates a physiological digital twin of the patient, and updates it in real time, includes the following operational procedures:

[0089] The process generates and updates the patient's physiological digital twin in real time. The state vector of the physiological digital twin includes, but is not limited to, core states such as hormone secretion level and metabolic rate, and is generated by hidden state mapping, expressed as:

[0090]

[0091] In the formula, express The state vector of the physiological digital twin at any given time. This represents the state mapping weight matrix. express The hidden state vector at time step 1. This represents the state-mapped bias vector, which, upon new data input, fuses historical states with new observations using a Kalman filter. The expression is:

[0092]

[0093] In the formula, express Time's up The predicted state value at time 10:00. Represents the state transition matrix. express The state vector of the physiological digital twin at any given time. Represents the control matrix. yes Control vector at time, express Time's up The state prediction error covariance matrix at time 1. express The error covariance matrix of the state at time step. Represents the state transition matrix The transpose of the matrix, Represents the process noise covariance matrix. express Kalman gain at time step Represents the observation matrix. It is the inverse covariance matrix for calculating the observation noise. It is the observation noise covariance matrix. Indicating the integration of new observations The state of the physiological digital twin after constant correction. express New observations at time [time] Indicates based on the predicted state The derived theoretical observations are used to compare with new observations. Compare and calculate the correction amount. Indicating the integration of new observations The error covariance matrix after time correction, The identity matrix is ​​used for identity transformations in matrix operations.

[0094] Specifically, the intelligent learning module uses an attention-based deep learning model to identify data associations of endocrine diseases from multi-source aligned data, generating and updating the patient's physiological digital twin in real time. It uses an embedding layer and attention mechanism to process multimodal data, such as monitoring data from wearable devices and home monitoring instruments, to generate hidden states. These hidden states are then converted into digital twin states through a mapping matrix. Distributed training is performed using the Apache Hadoop and Spark frameworks, and the deep learning model is optimized through cross-validation to facilitate adaptation to massive amounts of medical data. When new data is input, Kalman filtering is used to fuse historical states with new observations, dynamically correcting the twin and improving real-time performance. This is beneficial for improving the monitoring and prediction accuracy of disease progression.

[0095] The dynamic diagnosis and treatment module is used to output personalized treatment plans based on the physiological digital twin combined with the knowledge graph;

[0096] Furthermore, the dynamic diagnosis and treatment module, used to output personalized treatment plans based on the physiological digital twin combined with a knowledge graph, includes the following operational procedures:

[0097] The dynamic diagnosis and treatment module employs a reinforcement learning model, loading the quarterly updated endocrine disease diagnosis and treatment knowledge graph in triplet form, performing semantic parsing and feature mapping, and generating a decision feature vector based on the physiological digital twin combined with the knowledge graph. The expression is:

[0098]

[0099] In the formula, express The decision feature vector generated at each time step, This indicates that the fusion weight matrix is ​​a learnable parameter used to perform a linear transformation on the concatenated vector. It is a knowledge graph feature fusion computation based on the attention mechanism. It is the total number of entities participating in the computation in the knowledge graph. The attention weight matrix is ​​a learnable parameter used to calculate the state vector of the physiological digital twin. With knowledge graph entity vectors The correlation weight between them Indicates the first Vectors of entities, It is the first time when traversing all entities. Entity vectors, Indicates calculation using an exponential function and The relevance score after attention matrix transformation This indicates that all entities and The correlation scores are summed for normalization. The attention weight representation is the first A knowledge graph entity relative to a physiological digital twin The importance of The vector concatenation operation represents concatenating the physiological digital twin's state vector with the feature vector obtained by fusing the knowledge graph through the attention mechanism. It is the learnable parameters of the fused bias vector and In conjunction with this, the vectors that have been spliced ​​and linearly transformed are offset and adjusted.

[0100] Furthermore, the dynamic diagnosis and treatment module, used to output personalized treatment plans based on the physiological digital twin combined with a knowledge graph, includes the following operational procedures:

[0101] The reinforcement learning model adopts an Actor-Critic architecture, consisting of a policy function and a value function. It outputs personalized treatment plans. The knowledge graph is updated quarterly, incorporating the latest research findings and clinical experience. The reward function of the reinforcement learning model focuses on treatment effectiveness indicators, including glycemic control achievement rate and complication rate. The expression is:

[0102]

[0103] In the formula, Indicates the first Instant rewards for each moment Weighting coefficient The reward weighting for achieving blood sugar control targets Control the penalty weighting for other adverse clinical events. The exponential function makes the reward exhibit a smooth decay / increase trend with blood sugar deviation. Indicates the first The actual blood glucose level at that moment, Indicates the target blood glucose level. A parameter representing tolerance to blood glucose fluctuations. This represents the square of the blood glucose deviation. Representing the Other adverse clinical events at the time include, but are not limited to, the risk of hypoglycemia and ketosis.

[0104] Specifically, by dynamically integrating the patient's digital twin with the knowledge graph, it is easy to adjust the treatment plan in real time according to the patient's status. The reward mechanism for strengthening learning focuses on the blood glucose target achievement rate and the incidence of complications. The plan evolves dynamically, taking into account both control effect and risk prevention. The knowledge graph and deep learning model evolve together in quarterly updates, so that personalized treatment plans are always based on the latest clinical and scientific research results, improving efficacy and cutting-edge technology.

[0105] The doctor-patient collaboration module is used to link online and offline resources with the database, providing doctors with a data visualization dashboard and pushing interactive diagnosis and treatment suggestions to patients;

[0106] Furthermore, the doctor-patient collaboration module, which connects online and offline resources with the database, provides doctors with a data visualization dashboard and pushes interactive treatment suggestions to patients. The operational process includes:

[0107] The doctor-patient collaboration module includes a doctor's end and a patient's end. The doctor's end provides a data visualization dashboard that displays the physiological digital twin and treatment plan in real time, and supports manual adjustments by the doctor. (Expression:)

[0108]

[0109] In the formula, Indicates the current patient profile and the A portrait of a historical patient The similarity value between them This represents the current patient's profile vector. Indicates the first A vector image of a historical patient. Indicates to from arrive The items are accumulated. It is the total number of features in the patient profile. Indicates the first The weights of each feature, Indicates the current patient's number Features With historical patients No. Features The numerical product, Indicates the current patient profile The weighted norm, Indicates historical patients portrait The weighted norm.

[0110] Furthermore, the doctor-patient collaboration module, which connects online and offline resources with the database, provides doctors with a data visualization dashboard and pushes interactive treatment suggestions to patients. The operational process includes:

[0111] The interactive treatment suggestions pushed to the patient include, but are not limited to, animated demonstrations of insulin injection adjustment methods. The online and offline resource database is linked to calculate the fit with the patient profile, expressed as:

[0112]

[0113] In the formula, Indicates that for the first Individual treatment recommendations The fit calculation results Indicates the first One treatment recommendation to be evaluated. From arrive Summing up the following expressions, The total number of historical patient cases and related data used in the calculation. Indicates the current patient profile and the A portrait of a historical patient The similarity value between them This represents the current patient's profile vector. Indicates the first A vector image of a historical patient. The evaluation function represents the measure of treatment recommendations. Applied to the first A portrait corresponding to a historical patient The effect produced at that time, select The highest recommendation is pushed out in the form of animation. The online and offline resource linkage database includes, but is not limited to, the diagnosis and treatment resources, equipment, and expert information of offline medical institutions, as well as key information of doctor-patient communication recorded through natural language processing technology, which feeds back into the deep learning model.

[0114] Specifically, the doctor-patient collaboration module integrates online and offline resources and links databases to build an efficient collaborative mechanism between doctors and patients. On the doctor's end, a data visualization dashboard allows real-time viewing of the patient's physiological digital twin, such as dynamic blood glucose curves and thyroid hormone metabolism models, as well as AI-generated treatment plans, which can be manually adjusted based on clinical experience. On the patient's end, interactive treatment suggestions customized based on their profile are received, such as animated demonstrations of insulin injection site rotation and dosage adjustment. Combining the application effects of treatment suggestions in similar cases, the most suitable suggestions are pushed to the patient. At the same time, the database records key information from doctor-patient communication, such as adverse drug reactions and changes in lifestyle habits, which feeds back into the deep learning model for continuous optimization. Combining doctor experience and machine intelligence, it proposes treatment recommendations that are more suitable for the patient's condition. Through animated demonstrations and interactive suggestions, the understanding threshold of complex treatment operations for patients is reduced, improving compliance and treatment effectiveness. Natural language processing records doctor-patient communication, forming a closed loop of diagnosis-feedback-learning-optimization, which is conducive to improving the intelligence level and clinical practicality of this system.

[0115] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An Internet hospital-based endocrine disease artificial intelligence assisted diagnosis and treatment system, characterized in that, Comprise: a data acquisition module for accessing multi-source data of wearable devices, home detection devices, patient self-reporting, offline medical records, and third-party platforms; a time sequence alignment module for aligning multi-source data according to timestamps using high-precision time sequence alignment algorithms and outputting aligned data; an intelligent learning module for identifying endocrine disease data associations in the aligned data through a deep learning model, generating a patient's physiological digital twin, and updating it in real time; a dynamic diagnosis and treatment module for outputting personalized treatment plans based on the physiological digital twin and a knowledge graph; a doctor-patient linkage module for providing doctors with a data visualization dashboard and pushing interactive diagnosis and treatment recommendations to patients through an online and offline resource linkage database; The operation process of the data acquisition module for accessing multi-source data of wearable devices, home detection devices, patient self-reporting, offline medical records, and third-party platforms includes: The data collection module adopts SSL / TLS encryption transmission protocol, and establishes a local cache mechanism Each data is cached separately, and the cache scheduling strategy adopts a priority weighted sliding window mechanism, and the expression is: In the formula, Indicates at time Time The status of the local cache pool corresponding to the path data. Indicates the size of the sliding window. Indicates the time from arrive Sum all the terms within this interval. Indicates at time Weighting coefficients for time data Indicates at time Time The raw data content of the road data Represents all moments Corresponding weight coefficients The sum is 1. It is a hyperparameter used to control the rate of weight decay. It is an exponentially decaying function. It is the normalization factor. 2.The endocrine disease artificial intelligence auxiliary diagnosis and treatment system based on an Internet hospital of claim 1, wherein The operation process of the time sequence alignment module for aligning multi-source data according to timestamps using high-precision time sequence alignment algorithms and outputting aligned data includes: Adopt high-precision time sequence alignment algorithm, aiming at the multiple source data has One heterogeneous data source equipment, different equipment data Collecting frequency , produce data sequence, expression: In the formula, represents the original data sequence of the isomeric data source, represents the time stamp of the device at the sampling time, represents the observation value of the device at the sampling time, represents the total number of samples of the device, and the spline function interpolation method with disturbance compensation is adopted according to the global unified time stamp to reconstruct each device data, and the expression is: wherein represents the interpolated reconstruction value of the device at the globally uniform time , represents the globally uniform time stamp, represents the contribution of the original sampling point to the reconstruction value at the globally uniform time , is the Gaussian kernel part, is the perturbation compensation coefficient, is the cubic spline function at the globally uniform time . 3.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital of claim 2, characterized in that, The operation process of the time sequence alignment module for aligning multi-source data according to timestamps using high-precision time sequence alignment algorithms and outputting aligned data includes: The alignment process involves constructing a global alignment objective function, and aligning the data on a globally unified timestamp. Minimize the asynchronous cost of each interpolation sequence, expressed as: wherein, is a global alignment loss function, denotes traversing each global uniform timestamp in the global uniform timestamp set , denotes traversing all data pairs, is a weight of a data source pair indicating the importance of the and the data source, denotes interpolated data of the data source at time , is the square of the Euclidean distance, output aligned data, expression: wherein, represents the aligned data set, represents the global uniform time stamp, represents the interpolated data of the 1st to the nth data source at time . represents the total number of data sources, represents the traversal of each time point in the global uniform time stamp set . 4.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital of claim 3, characterized in that, The operation process of the intelligent learning module for identifying endocrine disease data associations in the aligned data through a deep learning model, generating a patient's physiological digital twin, and updating it in real time includes: The deep learning model is based on an attention mechanism, and the expression is: In the formula, represents the hidden state vector of the i-th channel / modal at time t, is an activation function, represents the original input data of the i-th channel / modal at time t, represents an embedding weight matrix, represents an embedding bias vector, represents the attention output vector of the i-th channel / modal at time t, represents that the outputs of the i attention heads are spliced together by a splicing operation, represents that the value weight matrix of the i-th attention head converts the hidden state into a value vector, represents that the value vector is weighted and summed with attention weights represents the total number of attention heads, represents the attention weight of position in the i-th attention head at time t, and is trained by Apache Hadoop, Spark framework, and the deep learning model is optimized by cross-validation method.​​​​​​​ 5.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital of claim 4, characterized in that, The operation process of the intelligent learning module for identifying endocrine disease data associations in the aligned data through a deep learning model, generating a patient's physiological digital twin, and updating it in real time includes: The generation of the patient's physiological digital twin and real-time updates, the state vector of the physiological digital twin includes hormone secretion levels, metabolic rate core states, and is generated by hidden state mapping, and the expression is: wherein denotes a state vector of the physiological digital twin at time instant denotes a state mapping weight matrix, denotes a hidden state vector at time instant denotes a state mapping bias vector, which is fused with the historical state and the new observation by Kalman filtering when new data is input, expression: In the formula, express Time's up The predicted state value at time 10:

00. Represents the state transition matrix. express The state vector of the physiological digital twin at any given time. Represents the control matrix. yes Control vector at time, express Time's up The state prediction error covariance matrix at time 1. express The error covariance matrix of the state at time step. Represents the state transition matrix The transpose of the matrix, Represents the process noise covariance matrix. express Kalman gain at time step Represents the observation matrix. It is the inverse covariance matrix for calculating the observation noise. It is the observation noise covariance matrix. Indicating the integration of new observations The state of the physiological digital twin after constant correction. express New observations at time [time] Indicates based on the predicted state The derived theoretical observations are used to compare with new observations. Compare and calculate the correction amount. Indicating the integration of new observations The error covariance matrix after time correction, The identity matrix is ​​used for identity transformations in matrix operations. 6.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital of claim 5, wherein, The operation process of the dynamic diagnosis and treatment module for outputting personalized treatment plans based on the physiological digital twin and a knowledge graph includes: The dynamic diagnosis and treatment module uses a reinforcement learning model to load the quarterly updated endocrine disease diagnosis and treatment knowledge graph in the form of triples, perform semantic analysis and feature mapping, generate a decision feature vector based on the physiological digital twin and the knowledge graph, and the expression is: In the formula, represents the decision feature vector generated at the moment, represents that the fusion weight matrix is a learnable parameter for linear transformation on the spliced vector, is knowledge graph feature fusion calculation based on attention mechanism, is the total number of entities participating in the calculation in the knowledge graph, is the attention weight matrix learnable parameter for calculating the physiological digital twin state vector and the association weight between the knowledge graph entity vector , represents the vector of the th entity, is the th entity vector when traversing all entities, represents the association degree score calculated by the exponential function and after the attention matrix transformation, represents the sum of the association degree scores of all entities and for normalization, is the attention weight indicating the importance of the th knowledge graph entity relative to the physiological digital twin , is the vector splicing operation indicating that the physiological digital twin state vector and the feature vector obtained by fusing the knowledge graph through the attention mechanism are spliced together, is a fusion bias vector learnable parameter and cooperates to offset the vector after splicing and linear transformation. 7.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital according to claim 6, characterized in that, The operation process of the dynamic diagnosis and treatment module for outputting personalized treatment plans based on the physiological digital twin and a knowledge graph includes: The reinforcement learning model uses an Actor-Critic architecture, consisting of a policy function and a value function, the output of the personalized treatment plan, the knowledge graph is updated every quarter, incorporating the latest research findings and clinical experience, and the reward function of the reinforcement learning model is centered on diagnosis and treatment effect indicators, including blood glucose control compliance rate and complication incidence, and the expression is: wherein, represents the instant reward at time represents the instant reward at time is a weight coefficient controls the reward weight for glycemic control controls the penalty weight for other adverse clinical events, is an exponential function to let the reward present a smooth decay / growth trend with glycemic deviation, represents the actual glycemic value at time represents the actual glycemic value at time represents the target glycemic value, represents the tolerance parameter of glycemic fluctuation, represents the square of glycemic deviation, represents other adverse clinical events at time represents other adverse clinical events at time 8.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital of claim 7, wherein, The operation process of the doctor-patient linkage module for providing doctors with a data visualization dashboard and pushing interactive diagnosis and treatment recommendations to patients through an online and offline resource linkage database includes: The doctor-patient linkage module includes a doctor end and a patient end. The doctor end provides a data visualization dashboard, which displays the physiological digital twin and treatment plan in real time, supports manual adjustment by doctors, and expresses the following: wherein denotes a similarity value between a current patient profile and a historical patient profile , denotes a profile vector of a current patient, denotes a profile vector of a historical patient, denotes a summation over the terms from to , is the total number of features of the patient profile, denotes a weight of the th feature, denotes the numerical product of the th feature of the current patient and the th feature of the historical patient, denotes a weighted norm of the current patient profile , denotes a weighted norm of the historical patient profile . 9.The endocrine disease AI-assisted diagnosis and treatment system based on an Internet hospital of claim 8, wherein, The doctor-patient linkage module is used to provide doctors with a data visualization dashboard and push interactive diagnosis and treatment suggestions to patients through an online and offline resource linkage database. The operation process includes: The patient end pushes interactive diagnosis and treatment suggestions, including an animation demonstration of insulin injection adjustment methods. The online and offline resource linkage database calculates the adaptation degree with the patient portrait, and expresses the following: In the formula, Indicates that for the first Individual treatment recommendations The fit calculation results Indicates the first One treatment recommendation to be evaluated. From arrive Summing up the following expressions, The total number of historical patient cases included in the calculation. Indicates the current patient profile and the A portrait of a historical patient The similarity value between them This represents the current patient's profile vector. Indicates the first A vector image of a historical patient. The evaluation function represents the measure of treatment recommendations. Applied to the first A portrait corresponding to a historical patient The effect produced at that time, select The highest recommendation is pushed out in the form of animation. The online and offline resource linkage database includes the diagnosis and treatment resources, equipment, and expert information of offline medical institutions, as well as key information of doctor-patient communication recorded through natural language processing technology, which feeds back into the deep learning model.

Images