A processing method and apparatus for predicting health status based on multidimensional indicators

Abstract

This invention relates to a method and apparatus for predicting health status based on multidimensional indicators. The method includes: constructing a deep learning model for predicting health status based on multidimensional indicators, denoted as the health status prediction model; constructing a first dataset through volunteer data collection and expert group scoring; performing a first round of training on the health status prediction model based on the first dataset under the condition that the model input vector is unmasked, and then performing a second round of training on the health status prediction model based on the first dataset under the condition that the model input vector is masked; after the model training is completed, inputting the user-input organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 into the health status prediction model for prediction processing to obtain the corresponding health status prediction vector Y, which is then fed back to the current user. This invention can improve the model's prediction performance.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a processing method and apparatus for predicting health status based on multidimensional indicators. Background Technology

[0002] With the application and development of artificial intelligence technology in the healthcare field, many health status prediction models based on machine learning or deep learning models have been deployed in various application scenarios. However, research has revealed some problems with current conventional model solutions: 1) Existing models often rely on a single data source (such as using only Western medicine laboratory indicators or traditional Chinese medicine diagnostic data), lacking the ability to fuse multi-dimensional information, which prevents further improvement in prediction performance; 2) Existing models have stringent requirements for the completeness of input data and cannot effectively handle common real-world issues such as missing indicators. When some examination data is unavailable, prediction performance will significantly decrease. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, apparatus, electronic device, and computer-readable storage medium for predicting health status based on multidimensional indicators. This invention constructs a health status prediction model based on multidimensional indicators. This model can predict health status based on multidimensional characteristics of both Western and Traditional Chinese Medicine examinations (Western medicine's human tissue and organ examination characteristics + Western medicine's blood examination characteristics + Traditional Chinese Medicine's Qi and blood examination characteristics). A first dataset is constructed through volunteer data collection and expert group scoring. A two-stage training strategy is employed when training the model based on the first dataset: the first round trains the model under unmasked input vectors, and the second round trains the model under masked input vectors. This invention overcomes the limitations of a single data source by introducing multidimensional indicators to improve model prediction performance; this invention also improves the model's prediction performance when examination indicators are incomplete through a two-stage training strategy.

[0004] To achieve the above objectives, a first aspect of the present invention provides a method for predicting health status based on multidimensional indicators, the method comprising: A deep learning model for predicting health status based on multidimensional indicators is constructed, denoted as the health status prediction model. This model is used to perform health status prediction processing based on the input organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, and outputs the corresponding health status prediction vector Y. The organ feature vector X1 includes N1 examination indicators. , 1 ≤ index i ≤ N1, where N1 is the preset total number of first indicators, and each of the inspection indicators Corresponding to a class of Western medical examination indicators for human tissues and organs; the blood feature vector X2 includes N2 examination indicators. ,1≤indexj≤N2, where N2 is the preset total number of second indicators, and each of the inspection indicators Corresponding to a type of blood test index in Western medicine; the Qi and Blood Feature Vector X3 includes N3 test indicators. , 1 ≤ index k ≤ N3, where N3 is the preset total number of third indicators, and each of the inspection indicators Corresponding to a class of Qi and blood examination indicators in Traditional Chinese Medicine; the health status prediction vector Y consists of M health scores y m Composition, 1 ≤ index m ≤ M, where M is the preset total number of chronic diseases, and each of the health scores y m Corresponding to a type of chronic disease; the health score y m The score range is the integer score range from 0 to 10; The first dataset is constructed through volunteer data collection and expert panel scoring; the first dataset includes multiple first data records; the first data record includes the organ feature vector X1, the blood feature vector X2, the qi and blood feature vector X3, and the label vector R. * The label vector R * Score of M tags Composition; score of each of the aforementioned tags Corresponding to a type of chronic disease; the label score The score range is the real number range from 0 to 1; The health status prediction model is trained in the first round on the first dataset under the condition that the model input vector is unmasked, and then trained in the second round on the first dataset under the condition that the model input vector is masked. After the model training is completed, the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 input by the user are input into the health status prediction model for prediction processing, and the health status prediction vector Y output by the model is fed back to the current user.

[0005] Preferably, the first model input terminal of the health status prediction model is used to receive the organ feature vector X1, the second model input terminal is used to receive the blood feature vector X2, the third model input terminal is used to receive the qi and blood feature vector X3, and the model output terminal is used to output the corresponding health status prediction vector Y. The health status prediction model includes a first feature extraction network, a second feature extraction network, a third feature extraction network, a first feature mapping layer, a second feature mapping layer, a third feature mapping layer, a feature fusion layer, and a linear regression layer. The input terminals of the first feature extraction network, the second feature extraction network, and the third feature extraction network are respectively connected to the corresponding first model input terminal, the second model input terminal, and the third model input terminal; the output terminals of the first feature extraction network, the second feature extraction network, and the third feature extraction network are respectively connected to the corresponding first feature mapping layer, the second feature mapping layer, and the third feature mapping layer input terminals; the output terminals of the first feature mapping layer, the second feature mapping layer, and the third feature mapping layer are respectively connected to the first input terminal, the second input terminal, and the third input terminal of the feature fusion layer; the output terminal of the feature fusion layer is connected to the input terminal of the linear regression layer; the output terminal of the linear regression layer is connected to the model output terminal. The first feature extraction network is implemented based on an MLP model structure; the first feature extraction network is used to extract features from the organ feature vector X1 to generate a corresponding feature vector H1 and send it to the first feature mapping layer; The reasoning method for the feature vector H1 is as follows: ; MLP1() is the inference function of the MLP model corresponding to the first feature extraction network, and θ1 is the model parameter of the current inference function; the vector shape of the feature vector H1 is N1×D1, and D1 is the preset first feature dimension; The second feature extraction network is implemented based on an MLP model structure; the second feature extraction network is used to extract features from the blood feature vector X2 to generate a corresponding feature vector H2 and send it to the second feature mapping layer; The reasoning method for the feature vector H2 is as follows: ; MLP2() is the inference function of the MLP model corresponding to the second feature extraction network, and θ2 is the model parameter of the current inference function; the vector shape of the feature vector H2 is N2×D2, and D2 is the preset second feature dimension; The third feature extraction network is implemented based on the MLP model structure; the third feature extraction network is used to extract features from the Qi and Blood feature vector X3 to generate the corresponding feature vector H3 and send it to the third feature mapping layer; The reasoning method for the feature vector H3 is as follows: ; MLP3() is the inference function of the MLP model corresponding to the third feature extraction network, and θ3 is the model parameter of the current inference function; the vector shape of the feature vector H3 is N3×D3, and D3 is the preset third feature dimension; The first feature mapping layer is used to map the feature vector H1 to a preset shared vector space to obtain the corresponding feature vector Z1, which is then sent to the feature fusion layer. The mapping method of the feature vector Z1 is as follows: ; W1 and b1 are the weight matrix and bias matrix of the first feature mapping layer; the shape of the weight matrix W1 is D. s ×D1,D s The shared vector feature dimension is preset; the bias matrix B1 has a shape of N1×D. s It consists of N1 elements of shape 1×D s Furthermore, the bias vector b1 is composed of identical parameters; the shape of the feature vector Z1 is N1×D. s ; The second feature mapping layer is used to map the feature vector H2 to the shared vector space to obtain the corresponding feature vector Z2, which is then sent to the feature fusion layer. The mapping method of the feature vector Z2 is as follows: ; W2 and B2 are the weight matrix and bias matrix of the second feature mapping layer; the shape of the weight matrix W2 is D. s ×D2; The shape of the bias matrix B2 is N2×D s It consists of N2 elements of shape 1×D s Furthermore, the bias vector b2 is composed of identical parameters; the shape of the feature vector Z2 is N2×D. s ; The third feature mapping layer is used to map the feature vector H3 to the shared vector space to obtain the corresponding feature vector Z3, which is then sent to the feature fusion layer. The mapping method of the feature vector Z3 is as follows: ; W3 and B3 are the weight matrix and bias matrix of the third feature mapping layer; the shape of the weight matrix W3 is D. s ×D3, the shape of the bias matrix B3 is N3×D s It consists of N3 elements of shape 1×D s Furthermore, the bias vector Z3 is composed of identical parameters; the shape of the feature vector Z3 is N3×D. s ; The feature fusion layer is used to concatenate the feature vectors Z1, Z2, and Z3 to obtain a vector of shape (N1+N2+N3)×D. sThe feature vector Z4; then, each feature channel of the feature vector Z4 is pooled to obtain a 1×D feature vector. s The feature vector P is sent to the linear regression layer; The linear regression layer is used to perform a linear transformation on the feature vector P and activate the transformation result to obtain the corresponding activation vector R; and derives and outputs the corresponding health status prediction vector Y based on the activation vector R. The reasoning method for the activation vector R is as follows: ; W4 and b4 are the weight matrix and bias vector of the linear regression layer; the weight matrix W4 has an M×D shape. s The bias vector b4 has a shape of M×1; sigmoid() is the sigmoid activation function; the activation vector R has a shape of M×1, consisting of M vector data r m Composition, each of the vector data r m The range of values ​​for all of them is the range of real numbers between 0 and 1; The health scores y of the health status prediction vector Y m The derivation method is as follows: ; round() is the rounding function.

[0006] Preferably, the construction of the first dataset through volunteer data collection and expert panel scoring specifically includes: Using the M-type chronic diseases corresponding to the health status prediction vector Y as the recruitment population feature, multiple volunteers are recruited to form a volunteer group; and multiple medical experts are recruited to form an expert group; the total number of volunteer types in the volunteer group is M+1, consisting of healthy volunteers and volunteers with M-type chronic diseases; the expert group includes multiple specialists or general practitioners in the field of Western medicine, multiple specialists or general practitioners in the field of traditional Chinese medicine, and multiple specialists or general practitioners in the field of integrated traditional Chinese and Western medicine. Each volunteer in the volunteer group is designated as the current volunteer. With authorization from the current volunteer or their guardian, data is collected on the current volunteer's organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3. The expert group then assigns scores to the M tags corresponding to the M types of chronic diseases based on the current volunteer's organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, according to an expert consensus scoring method. A score assessment is performed; and the obtained M tag scores are used to evaluate the scores. The corresponding label vector R is formed. *; and composed of the organ feature vector X1, the blood feature vector X2, the gas and blood feature vector X3, and the label vector R corresponding to the current volunteer. * Form a corresponding first data record; The first dataset is composed of all the first data records obtained.

[0007] Preferably, the step of performing a first round of training on the health status prediction model based on the first dataset under the condition that the model input vector is unmasked, and then performing a second round of training on the health status prediction model based on the first dataset under the condition that the model input vector is masked, specifically includes: Step 41: Initialize the training round counter to 1; Step 42: Based on a preset first segmentation ratio, the first dataset is randomly divided into two subsets, denoted as the first training set and the first evaluation set. Both the first training set and the first evaluation set consist of multiple first data records; the total number of records in the first training set is denoted as N. tr The total number of records in the first evaluation set is denoted as N. av The ratio N of the total number of records in the first training set to the total number of records in the first evaluation set. tr :N av Satisfying the first segmentation ratio; the label vector R of each of the first data records in the first training set. * Record as the corresponding The label vector R * The scores of each of the aforementioned tags Record as the corresponding 1 ≤ index u ≤ N tr The label vector R of each of the first data records in the first evaluation set * Record as the corresponding The label vector R * The scores of each of the aforementioned tags Record as the corresponding 1 ≤ index v ≤ N tr ; Step 43: Take each of the first data records in the first training set as the current training record. When the training round counter is 1, directly input the three feature vectors of the current training record into the model for prediction processing. When the training round counter is 2, first randomly mask the three feature vectors of the current training record according to a random masking percentage, and then input the masked three feature vectors into the model for prediction processing. The activation vector R generated during this process is then used for prediction processing. u Its corresponding label vector Form a corresponding first prediction-label pair (R) u , ); Specifically: each of the first data records in the first training set is used as the current training record; and the training round counter is identified; if the training round counter is 1, the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 of the current training record are input into the health status prediction model for prediction processing; if the training round counter is 2, a percentage is randomly generated between 10% and a preset percentage α1. A percentage is randomly generated between 10% and a preset percentage α2. A percentage is randomly generated between 10% and a preset percentage α3. And calculate the corresponding three random occlusion numbers. , , And any of the organ feature vectors X1 recorded in the current training record The aforementioned inspection indicators Set as an invalid indicator, any one of the blood feature vectors X2 The aforementioned inspection indicators Set as an invalid indicator, any one of the Qi and Blood feature vectors X3 The aforementioned inspection indicators The organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3, which have been set as invalid indicators, are input into the health status prediction model for prediction processing; and the activation vector R generated during this processing is recorded as the corresponding activation vector R. u And the vector data r of the current activation vector R. m Let r be the corresponding r u,m And the activation vector R obtained this time u and the corresponding label vector Form a corresponding first prediction-label pair (R) u , ); in, , , int() is the integer function; Step 44, obtain N tr The first predicted-label pair (R) u , Substitute the value into the preset first model loss function L1 to calculate the corresponding first loss value; Wherein, the loss function L1 of the first model is: ; Step 45: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 46; if it does not, modulate the model parameters of the health state prediction model in one round based on the preset first model optimizer in the direction of minimizing the first model loss function L1, and return to step 43 when the first round of modulation ends. The first model optimizer includes the Adam optimizer and the SGD optimizer. Step 46: Take each of the first data records in the first evaluation set as the current evaluation record, and when the training round counter is 1, directly input the three feature vectors of the current evaluation record into the model for prediction processing. When the training round counter is 2, first randomly mask the three feature vectors of the current evaluation record according to the maximum masking percentage, and then input the masked three feature vectors into the model for prediction processing. The activation vector R generated in this process is then used for prediction processing. v Its corresponding label vector Form a corresponding second prediction-label pair (R) v , ); Specifically: each of the first data records in the first evaluation set is taken as the current evaluation record; and the training round counter is identified; if the training round counter is 1, the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 of the current evaluation record are input into the health status prediction model for prediction processing; if the training round counter is 2, the corresponding three random occlusion numbers are calculated. , , and any of the organ feature vectors X1 recorded in the current evaluation record The aforementioned inspection indicators Set as an invalid indicator, any one of the blood feature vectors X2 The aforementioned inspection indicators Set as an invalid indicator, any one of the Qi and Blood feature vectors X3 The aforementioned inspection indicators The organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3, after the invalidation of the indicator, are set as invalid indicators and input into the health status prediction model for prediction processing; the activation vector R generated during this processing is recorded as the corresponding activation vector R. v And the vector data r of the current activation vector R. mLet r be the corresponding r v,m And the activation vector R obtained this time v and the corresponding label vector Form a corresponding second prediction-label pair (R) v , ); Step 47, obtain N av The second prediction-label pair (R) v , Substitute the value into the preset first model evaluation function L2 to calculate the corresponding first evaluation value; Wherein, the first model evaluation function L2 is: ; Step 48: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 42; if it does, identify the training round counter; if the training round counter is 1, confirm the end of the first round of training, increment the training round counter by 1, and return to step 42; if the training round counter is 2, stop training and confirm the end of the second round of training.

[0008] A second aspect of the present invention provides an apparatus for implementing the processing method for predicting health status based on multidimensional indicators as described in the first aspect above. The apparatus includes: a model building module, a data acquisition module, a model training module, and a model application module. The model building module is used to construct a deep learning model for predicting health status based on multi-dimensional indicators, denoted as the health status prediction model. The health status prediction model is used to perform health status prediction processing based on the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 input to the model, and output the corresponding health status prediction vector Y. The organ feature vector X1 includes N1 examination indicators. , 1 ≤ index i ≤ N1, where N1 is the preset total number of first indicators, and each of the inspection indicators Corresponding to a class of Western medical examination indicators for human tissues and organs; the blood feature vector X2 includes N2 examination indicators. ,1≤indexj≤N2, where N2 is the preset total number of second indicators, and each of the inspection indicators Corresponding to a type of blood test index in Western medicine; the Qi and Blood Feature Vector X3 includes N3 test indicators. , 1 ≤ index k ≤ N3, where N3 is the preset total number of third indicators, and each of the inspection indicators Corresponding to a class of Qi and blood examination indicators in Traditional Chinese Medicine; the health status prediction vector Y consists of M health scores y mComposition, 1 ≤ index m ≤ M, where M is the preset total number of chronic diseases, and each of the health scores y m Corresponding to a type of chronic disease; the health score y m The score range is the integer score range from 0 to 10; The data acquisition module is used to construct a first dataset through volunteer data collection and expert panel scoring; the first dataset includes multiple first data records; the first data record includes the organ feature vector X1, the blood feature vector X2, the qi and blood feature vector X3, and the label vector R. * The label vector R * Score of M tags Composition; score of each of the aforementioned tags Corresponding to a type of chronic disease; the label score The score range is the real number range from 0 to 1; The model training module is used to perform a first round of training on the health status prediction model based on the first dataset under the condition that the model input vector is unmasked, and then perform a second round of training on the health status prediction model based on the first dataset under the condition that the model input vector is masked. The model application module is used to input the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 input by the user into the health status prediction model for prediction processing after the model training is completed, and to feed back the health status prediction vector Y output by the model to the current user.

[0009] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver; The processor is used to couple with the memory, read and execute instructions in the memory to implement the steps of the method described in the first aspect above; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0010] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions described in the first aspect.

[0011] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting health status based on multidimensional indicators. As described above, this invention constructs a health status prediction model based on multidimensional indicators. This model can predict health status based on multidimensional characteristics of both Western and Traditional Chinese Medicine examinations (Western medicine's human tissue and organ examination characteristics + Western medicine's blood examination characteristics + Traditional Chinese Medicine's Qi and blood examination characteristics). A first dataset is constructed through volunteer data collection and expert group scoring. A two-stage training strategy is employed when training the model based on the first dataset: the first round trains the model under unmasked input vectors, and the second round trains the model under masked input vectors. This invention improves the model's predictive performance by introducing multidimensional indicators and enhances its predictive performance when examination indicators are incomplete through a two-stage training strategy, thereby improving the model's generalization ability. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of a method for predicting health status based on multidimensional indicators provided in Embodiment 1 of the present invention; Figure 2 This is a model structure diagram of a health status prediction model provided in Embodiment 1 of the present invention; Figure 3 This is a module structure diagram of a processing device for predicting health status based on multidimensional indicators, provided in Embodiment 2 of the present invention. Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0014] Embodiment 1 of the present invention provides a processing method for predicting health status based on multidimensional indicators, such as... Figure 1 The schematic diagram shows a method for predicting health status based on multidimensional indicators provided in Embodiment 1 of the present invention. The method mainly includes the following steps: Step 1: Construct a deep learning model for predicting health status based on multidimensional indicators, denoted as the health status prediction model.

[0015] Here, the health status prediction model of this invention is used to perform health status prediction processing based on the organ feature vector X1, blood feature vector X2 and qi and blood feature vector X3 input by the model and output the corresponding health status prediction vector Y.

[0016] The organ feature vector X1 in this embodiment of the invention includes N1 examination indicators. , 1 ≤ index i ≤ N1, where N1 is the preset total number of first indicators, and each inspection indicator It corresponds to a type of Western medical examination indicator for human tissues and organs.

[0017] The set of human organ examination indicators corresponding to organ feature vector X1 consists of organ indicator set, system indicator set, and human structure indicator set. The organ indicator set comprises subsets of heart, lung, liver, spleen, kidney, and stomach indicators. The system indicator set comprises subsets of digestive system, respiratory system, endocrine system, urinary and reproductive system, circulatory system, nervous system, musculoskeletal system, and immune system indicators. The human structure indicator set includes subsets of head, neck, thoracic cavity, abdominal cavity, pelvic cavity, and limb indicators. Each indicator subset consists of one or more corresponding examination items. Each examination indicator in organ feature vector X1... It refers to the index value of a specific examination item within the set of examination indicators for human tissues and organs.

[0018] It should be noted that each inspection indicator When an indicator is valid, it is a percentage value ranging from 0 to 100; when an indicator is invalid, it is a preset invalid indicator value. The invalid indicator value in this embodiment can be set to 0 or a negative integer. This embodiment allows multiple invalid indicators in the organ feature vector X1, but the percentage of invalid indicators cannot exceed a preset upper limit; that is, the maximum invalid indicator percentage corresponding to organ feature vector X1 is a preset percentage α1, where 10% < α1 < 100%. Furthermore, this embodiment can further examine the indicators in organ feature vector X1. Configure mandatory attribute settings. These mandatory attributes include two states: yes and no. If a certain inspection indicator... If the required attribute is yes, the current indicator cannot be set as an invalid indicator; otherwise, if the required attribute is no, the current indicator can be set as invalid. Correspondingly, the examination indicators in the organ feature vector X1... When mandatory attributes are present, the maximum total number of invalid indicators in the organ feature vector X1 should be less than the total number of optional indicators. The maximum total number of invalid indicators in the organ feature vector X1 = int(N1×α1), where int() is the floor function. The total number of optional indicators in the organ feature vector X1 is the check indicator for whether the mandatory attribute is present or not in the vector. The total number.

[0019] It should also be noted that the examination indicators for various subsets of the human tissue and organ examination indicator set are selected and defined based on the opinions of medical and health experts. This invention does not specifically limit the range of indicators. For example, the cardiac indicator subset can include examination indicators such as heart rate, systolic blood pressure, and diastolic blood pressure based on expert opinions. Furthermore, because each effective indicator is standardized to a percentage range [0, 100], when preparing the organ feature vector X1, it is necessary to convert the actual examination values ​​of various examination indicators to corresponding percentage values ​​to obtain the corresponding examination indicators. The specific conversion method can be customized based on application requirements. This invention does not impose specific limitations on the conversion method. For example, the true inspection value can be normalized first according to the average value and standard deviation of similar indicators using the z-core method, and then the normalized value can be converted into a percentage value.

[0020] The blood feature vector X2 in this embodiment of the invention includes N2 examination indicators. , 1 ≤ index j ≤ N2, where N2 is the preset total number of second indicators, and each inspection indicator It corresponds to a type of blood test indicator in Western medicine.

[0021] The set of blood test indicators corresponding to the blood feature vector X2 consists of subsets of lipid metabolism indicators, glucose metabolism indicators, hormone metabolism indicators, immune metabolism indicators, electrolyte indicators, pulmonary function indicators, liver function indicators, kidney function indicators, gastrointestinal function indicators, hypertension risk indicators, thrombosis risk indicators, inflammation risk indicators, and tumor risk indicators. Each subset of indicators is further composed of one or more corresponding test items. Each test indicator in the blood feature vector X2... It refers to the index value of a specific test item in the set of blood test indicators.

[0022] It should be noted that each inspection indicator When an indicator is valid, it is a percentage value ranging from 0 to 100; when an indicator is invalid, it is an invalid indicator value. This embodiment of the invention allows multiple invalid indicators in the blood feature vector X2, but the percentage of invalid indicators cannot exceed a preset upper limit. That is, the maximum percentage of invalid indicators corresponding to the blood feature vector X2 is a preset percentage α2, where 10% < α2 < 100%. Furthermore, this embodiment of the invention can further examine the indicators in the blood feature vector X2. Configure mandatory attribute settings. These mandatory attributes include two states: yes and no. If a certain inspection indicator... If the required attribute is yes, the current indicator cannot be set as an invalid indicator; otherwise, if the required attribute is no, the current indicator can be set as invalid. Correspondingly, the examination indicators in the blood feature vector X2... When mandatory attributes are present, the maximum total number of invalid indicators in the blood feature vector X2 should be less than the total number of optional indicators. The maximum total number of invalid indicators in the blood feature vector X2 = int(N1×α2). The total number of optional indicators in the blood feature vector X2 is the check indicators for whether the mandatory attribute is present or not in the vector. The total number.

[0023] It should also be noted that the test items in the various subsets of the blood test indicator set are selected and defined based on the opinions of medical and health experts. This invention does not specifically limit the range of indicators. For example, the lipid metabolism indicator subset can include test items such as total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, and triglycerides, based on expert opinions. Furthermore, because each effective indicator is standardized to a percentage range [0,100], when preparing the blood feature vector X2, it is necessary to convert the actual test values ​​of various test items into corresponding percentage values ​​to obtain the corresponding test indicators. The specific conversion method can be customized based on application requirements. This invention does not impose specific limitations on the conversion method. For example, the true inspection value can be normalized first according to the average value and standard deviation of similar indicators using the z-core method, and then the normalized value can be converted into a percentage value.

[0024] The Qi and Blood Feature Vector X3 of this embodiment includes N3 examination indicators. , 1 ≤ index k ≤ N3, where N3 is the total number of preset third indicators, and each inspection indicator It corresponds to a type of Qi and blood examination indicator in traditional Chinese medicine.

[0025] The set of Qi and Blood examination indicators corresponding to the Qi and Blood feature vector X3 consists of Qi and Blood sufficiency indicators, Qi and Blood patency indicators, pulse pressure indicators, and four diagnostic indicators. The pulse pressure indicator set includes mean arterial pressure, arterial elasticity, arterial pulse rate rhythm, left upper limb pulse pressure difference, left lower limb pulse pressure difference, right upper limb pulse pressure difference, and right lower limb pulse pressure difference. The four diagnostic indicators set consists of a set of corresponding subsets of inspection, auscultation, inquiry, and palpation indicators; each of these subsets further comprises multiple corresponding binary indicators. Each examination indicator in the Qi and Blood feature vector X3... It refers to the value of a specific indicator within the set of indicators for blood and qi examination.

[0026] It should be noted that the examination indicators correspond to the various indicators in the sets of Qi and Blood sufficiency, Qi and Blood Flow, and Pulse Pressure indicators. When an indicator is valid, it is a percentage value ranging from 0 to 100; when an indicator is invalid, it is an invalid indicator value. The examination indicators correspond to the various binary categories of the four diagnostic indicators set. When an indicator is valid, it is a binary value of 0 or 1, where 0 represents none and 1 represents present; when an indicator is invalid, it is -1. This embodiment of the invention allows multiple invalid indicators in the Qi and Blood feature vector X3, but the percentage of invalid indicators cannot exceed a preset upper limit. That is, the maximum percentage of invalid indicators corresponding to the Qi and Blood feature vector X3 is a preset percentage α3, where 10% < α3 < 100%. Furthermore, this embodiment of the invention can further examine the indicators in the Qi and Blood feature vector X3. Configure mandatory attribute settings. These mandatory attributes include two states: yes and no. If a certain inspection indicator... If the required attribute is yes, the current indicator cannot be set as an invalid indicator; otherwise, if the required attribute is no, the current indicator can be set as an invalid indicator. Correspondingly, the check indicator in the Qi and Blood feature vector X3... When mandatory attributes are present, the maximum total number of invalid indicators in the Qi and Blood feature vector X3 should be less than the total number of optional indicators. The maximum total number of invalid indicators in the Qi and Blood feature vector X3 = int(N1×α3). The total number of optional indicators in the Qi and Blood feature vector X3 is the check indicator for whether the mandatory attribute is present or not in the vector. The total number.

[0027] It should also be noted that the indicators for Qi and Blood sufficiency, Qi and Blood patency, and pulse pressure in the Qi and Blood examination indicator set are selected and defined based on the opinions of experts in the medical and health field. This invention does not specifically limit the range of indicators. For example, the Qi and Blood sufficiency and Qi and Blood patency indicators can be used to reflect the examination results of whether Qi and Blood are sufficient and unobstructed, based on the theory of Qi and Blood in Traditional Chinese Medicine and in conjunction with the opinions of TCM experts, through two corresponding percentage indicators. Another example is that the arterial elasticity index in the pulse pressure indicator set can be identified through its inverse linear relationship with the pulse wave velocity. Furthermore, because the Qi and Blood sufficiency, Qi and Blood patency, and various effective indicators in the pulse pressure indicator set are all standardized to a percentage range [0, 100], when preparing the Qi and Blood feature vector X3, it is necessary to convert the actual examination values ​​of the Qi and Blood sufficiency, Qi and Blood patency, and various indicators in the pulse pressure indicator set into corresponding percentage values ​​to obtain the corresponding examination indicators. The specific conversion method can be customized based on application requirements. This invention does not impose specific limitations on the conversion method. For example, the true inspection value can be normalized first according to the average value and standard deviation of similar indicators using the z-core method, and then the normalized value can be converted into a percentage value.

[0028] It should also be noted that the various binary indicators in the four diagnostic methods of traditional Chinese medicine (TCM) for the Qi and Blood examination indicator set are selected and defined based on the opinions of experts in the medical and health field. This invention does not specifically limit the scope of the indicators. For example, the visual diagnosis indicator subset can include various binary indicators for judging tongue color (such as red, crimson, purple, etc.) based on the TCM four diagnostic methods theory and the opinions of TCM experts. It can also include various binary indicators for judging tongue quality (such as aged tongue quality, tender tongue quality, swollen tongue quality, etc.). Furthermore, it can include indicators for judging the quality and color of the tongue coating. Multiple binary classification indicators (such as indicators for thicker tongue coating, thinner tongue coating, moderately thick tongue coating, dry tongue coating, sticky tongue coating, moist tongue coating, white tongue coating, yellow tongue coating, black tongue coating, etc.) are included. For example, the palpation indicator subset can incorporate various binary indicators used to judge pulse characteristics (such as floating pulse, deep pulse, rapid pulse, slow pulse, slippery pulse, hesitant pulse, wiry pulse, tense pulse, weak pulse, strong pulse, surging pulse, hurried pulse, knotted pulse, intermittent pulse, etc.) based on the four diagnostic methods of traditional Chinese medicine and the opinions of TCM experts. Furthermore, because various effective binary classification indicators are standardized into a binary value of 0 or 1, when preparing the Qi and Blood feature vector X3, it is necessary to perform corresponding 0 or 1 binary value conversions on the actual examination results of each binary classification indicator to obtain the corresponding examination indicator. .

[0029] The health status prediction vector Y in this embodiment of the invention consists of M health scores y m Composition, 1 ≤ index m ≤ M, where M is the preset total number of chronic diseases, and each health score y m Corresponding to a type of chronic disease; health score y m The score range is an integer range from 0 to 10.

[0030] It should be noted that the types of chronic diseases in the chronic disease set corresponding to the health status prediction vector Y are selected and defined based on the opinions of experts in the medical and health field. This invention does not specifically limit the scope of diseases. For example, based on expert opinions, some or all of the following diseases may be included in the chronic disease set corresponding to the health status prediction vector Y: coronary heart disease, stroke, arteriosclerosis obliterans, thromboangiitis obliterans, deep vein thrombosis, varicose veins, diabetic complications, COPD, asthma, depression, anxiety, bipolar disorder, schizophrenia, obsessive-compulsive disorder, etc.

[0031] It should also be noted that the health score y m The score is used to predict and assess the likelihood that the current examinee is suffering from the corresponding type m chronic disease. The higher the score, the lower the likelihood that the current examinee is suffering from the type m chronic disease and the better the current examinee's health is relative to the type m chronic disease. Conversely, the lower the score, the higher the likelihood that the current examinee is suffering from the type m chronic disease and the worse the current examinee's health is relative to the type m chronic disease.

[0032] like Figure 2 As shown in the model structure diagram of a health status prediction model provided in Embodiment 1 of the present invention, the first model input terminal of the health status prediction model is used to receive organ feature vector X1, the second model input terminal is used to receive blood feature vector X2, the third model input terminal is used to receive qi and blood feature vector X3, and the model output terminal is used to output the corresponding health status prediction vector.

[0033] like Figure 2 As shown, the model components of the health status prediction model include: a first feature extraction network, a second feature extraction network, a third feature extraction network, a first feature mapping layer, a second feature mapping layer, a third feature mapping layer, a feature fusion layer, and a linear regression layer.

[0034] like Figure 2As shown, the connection relationships of the model components in the health status prediction model are as follows: the inputs of the first feature extraction network, the second feature extraction network, and the third feature extraction network are connected to the corresponding first model input, the second model input, and the third model input, respectively; the outputs of the first feature extraction network, the second feature extraction network, and the third feature extraction network are connected to the corresponding first feature mapping layer, the second feature mapping layer, and the third feature mapping layer, respectively; the outputs of the first feature mapping layer, the second feature mapping layer, and the third feature mapping layer are connected to the first, second, and third inputs of the feature fusion layer, respectively; the output of the feature fusion layer is connected to the input of the linear regression layer; and the output of the linear regression layer is connected to the model output.

[0035] The functionalities of the model components in the health status prediction model are shown below.

[0036] 1) First, second, and third feature extraction networks: The first feature extraction network in this embodiment of the invention is implemented based on an MLP model structure. The first feature extraction network is used to extract features from the organ feature vector X1 to generate a corresponding feature vector H1, which is then sent to the first feature mapping layer.

[0037] Here, the reasoning method for feature vector H1 in this embodiment of the invention is as follows: ; Where MLP1() is the inference function of the MLP model corresponding to the first feature extraction network, θ1 is the model parameter of the current inference function; the vector shape of the feature vector H1 is N1×D1, and D1 is the preset first feature dimension.

[0038] The second feature extraction network in this embodiment of the invention is implemented based on an MLP model structure. The second feature extraction network is used to extract features from the blood feature vector X2 to generate a corresponding feature vector H2, which is then sent to the second feature mapping layer.

[0039] Here, the reasoning method for feature vector H2 in this embodiment of the invention is as follows: ; Where MLP2() is the inference function of the MLP model corresponding to the second feature extraction network, θ2 is the model parameter of the current inference function; the vector shape of feature vector H2 is N2×D2, and D2 is the preset second feature dimension.

[0040] The third feature extraction network in this embodiment of the invention is implemented based on an MLP model structure. The third feature extraction network is used to extract features from the Qi and Blood feature vector X3 to generate the corresponding feature vector H3, which is then sent to the third feature mapping layer.

[0041] Here, the reasoning method for feature vector H3 in this embodiment of the invention is as follows: ; Where MLP3() is the inference function of the MLP model corresponding to the third feature extraction network, θ3 is the model parameter of the current inference function; the vector shape of feature vector H3 is N3×D3, and D3 is the preset third feature dimension.

[0042] 2) First, second, and third feature mapping layers: The first feature mapping layer in this embodiment of the invention is used to map the feature vector H1 to a preset shared vector space to obtain the corresponding feature vector Z1, which is then sent to the feature fusion layer.

[0043] Here, the mapping method of feature vector Z1 in this embodiment of the invention is as follows: ; Where W1 and b1 are the weight matrix and bias matrix of the first feature mapping layer; the shape of the weight matrix W1 is D. s ×D1,D s The shared vector feature dimension is preset; the bias matrix B1 has a shape of N1×D. s It consists of N1 elements of shape 1×D s Furthermore, the parameters are consistent, consisting of bias vector b1; the eigenvector Z1 has a vector shape of N1×D. s .

[0044] The second feature mapping layer in this embodiment of the invention is used to map the feature vector H2 to the shared vector space to obtain the corresponding feature vector Z2, which is then sent to the feature fusion layer.

[0045] Here, the mapping method of feature vector Z2 in this embodiment of the invention is as follows: ; Where W2 and B2 are the weight matrix and bias matrix of the second feature mapping layer; the shape of the weight matrix W2 is D. s ×D2; The shape of the bias matrix B2 is N2×D s It consists of N2 elements of shape 1×D s Furthermore, the bias vector b2 is composed of consistent parameters; the shape of the eigenvector Z2 is N2×D. s .

[0046] The third feature mapping layer in this embodiment of the invention is used to map the feature vector H3 to the shared vector space to obtain the corresponding feature vector Z3, which is then sent to the feature fusion layer.

[0047] Here, the mapping method of feature vector Z3 in this embodiment of the invention is as follows: ; Where W3 and B3 are the weight matrix and bias matrix of the third feature mapping layer; the shape of the weight matrix W3 is D. s ×D3, the shape of the bias matrix B3 is N3×D s It consists of N3 elements of shape 1×D s Furthermore, the parameters are consistent, and the bias vector b3 is composed of the vector; the shape of the eigenvector Z3 is N3×D. s .

[0048] 3) Feature fusion layer: The feature fusion layer in this embodiment of the invention is used to concatenate feature vectors Z1, Z2, and Z3 to obtain a vector of shape (N1+N2+N3)×D. s The feature vector Z4 is then pooled for each feature channel of the feature vector Z4 to obtain a 1×D feature vector. s The feature vector P is sent to the linear regression layer.

[0049] 4) Linear regression layer: In this embodiment of the invention, the linear regression layer is used to perform a linear transformation on the feature vector P and activate the transformation result to obtain the corresponding activation vector R; and the corresponding health state prediction vector Y is derived and output based on the activation vector R.

[0050] Here, the reasoning method for the activation vector R in this embodiment of the invention is as follows: ; Where W4 and b4 are the weight matrix and bias vector of the linear regression layer; the weight matrix W4 has an M×D shape. s The bias vector b4 has a shape of M×1; sigmoid() is the sigmoid activation function; the activation vector R has a shape of M×1, consisting of M vector data r. m Composition, each vector data r m The range of values ​​for all of them is the range of real numbers between 0 and 1.

[0051] Health scores y of the health status prediction vector Y m The derivation method is as follows: ; The round() function is used for rounding.

[0052] Step 2: Construct the first dataset by collecting data from volunteers and scoring by an expert panel.

[0053] The first dataset comprises multiple first data records. Each first data record includes an organ feature vector X1, a blood feature vector X2, a gas and blood feature vector X3, and a label vector R. * Label vector R *Score of M tags Composition; Score for each tag Corresponding to a type of chronic disease; label score The score range is the real number range from 0 to 1.

[0054] Step 2 specifically includes: Step 21: Using the M-type chronic diseases corresponding to the health status prediction vector Y as the characteristics of the recruited population, recruit multiple volunteers to form a volunteer group; and recruit multiple medical experts to form an expert group.

[0055] Here, in this embodiment of the invention, the total number of volunteer types in the volunteer group is M+1, consisting of healthy volunteers and volunteers with M types of chronic diseases. For example, if the M types of chronic diseases corresponding to the health status prediction vector Y include coronary heart disease, stroke, arteriosclerosis obliterans, thromboangiitis obliterans, deep vein thrombosis, varicose veins, diabetic complications, COPD, asthma, depression, anxiety, bipolar disorder, schizophrenia, and obsessive-compulsive disorder, then the volunteer types in the volunteer group should include healthy volunteers, coronary heart disease volunteers, stroke volunteers, arteriosclerosis obliterans volunteers, thromboangiitis obliterans volunteers, deep vein thrombosis volunteers, varicose vein volunteers, diabetic complications volunteers, COPD volunteers, asthma volunteers, depression volunteers, anxiety volunteers, bipolar disorder volunteers, schizophrenia volunteers, and obsessive-compulsive disorder volunteers. Furthermore, when recruiting volunteers with chronic diseases, it is necessary not only to recruit volunteers with a single chronic disease but also volunteers suffering from multiple chronic diseases simultaneously.

[0056] The expert group in this invention embodiment includes specialists or general practitioners in multiple Western medicine fields, specialists or general practitioners in multiple traditional Chinese medicine fields, and specialists or general practitioners in multiple integrated traditional Chinese and Western medicine fields.

[0057] Step 22: Each volunteer in the volunteer group is designated as the current volunteer. With authorization from the current volunteer or their guardian, data on the current volunteer's organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 are collected. An expert panel then assigns scores to the M labels corresponding to the M types of chronic diseases based on the current volunteer's organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, according to an expert consensus scoring method. The scores are determined by assigning values ​​to the M tags. Form the corresponding label vector R * ; and is composed of the organ feature vector X1, blood feature vector X2, gas and blood feature vector X3, and label vector R corresponding to the current volunteer. * This forms a corresponding first data record.

[0058] Step 23, and the first dataset is composed of all the first data records obtained.

[0059] Step 3: Train the health status prediction model for the first time on the first dataset under the condition that the model input vector is unmasked, and then train the health status prediction model for the second time on the first dataset under the condition that the model input vector is masked.

[0060] Specifically, it includes: Step 31: Initialize the training round counter to 1.

[0061] Step 32: Based on the preset first segmentation ratio, the first dataset is randomly divided into two sub-datasets, denoted as the first training set and the first evaluation set.

[0062] Here, the first division ratio in this embodiment of the invention is a pre-set ratio parameter, such as 8:2.

[0063] In this embodiment of the invention, both the first training set and the first evaluation set consist of multiple first data records; the total number of records in the first training set is denoted as N. tr The total number of records in the first evaluation set is denoted as N. av The ratio N of the total number of records in the first training set to the total number of records in the first evaluation set. tr :N av Satisfying the first segmentation ratio. The label vector R of each first data record in the first training set. * Record as the corresponding Label vector R * Each tag score Record as the corresponding 1 ≤ index u ≤ N tr The label vector R of each first data record in the first evaluation set * Record as the corresponding Label vector R * Each tag score Record as the corresponding 1 ≤ index v ≤ N tr .

[0064] Step 33: Take each first data record of the first training set as the current training record. When the training round counter is 1, directly input the three feature vectors of the current training record into the model for prediction. When the training round counter is 2, first randomly mask the three feature vectors of the current training record according to a random masking percentage, and then input the masked three feature vectors into the model for prediction. The activation vector R generated during this process is used to predict the result. u Its corresponding label vector Form a corresponding first prediction-label pair (R) u, ).

[0065] Specifically, it includes: Step 331: Take each of the first data records in the first training set as the current training record.

[0066] Step 332, and identify the training round counter.

[0067] Step 333: If the training round counter is 1, then input the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 recorded in the current training into the health status prediction model for prediction processing.

[0068] Step 334: If the training round counter is 2, then randomly generate a percentage between 10% and the preset percentage α1. A percentage is randomly generated between 10% and a preset percentage α2. A percentage is randomly generated between 10% and a preset percentage α3. And calculate the corresponding three random occlusion numbers. , , And any of the organ feature vectors X1 recorded in the current training record Inspection indicators Let the invalid index be any one of the blood feature vectors X2. Inspection indicators Let any of the following be an invalid indicator or a feature vector X3: Inspection indicators The invalid indicators are set as the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, which are set as invalid indicators, and then input into the health status prediction model for prediction processing.

[0069] in, , , int() is the integer function.

[0070] It should also be noted here that the examination indicators in organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3... , , All of them have mandatory attributes, so, in the current training record of any organ feature vector X1, Inspection indicators Let the invalid index be any one of the blood feature vectors X2. Inspection indicators Let any of the following be an invalid indicator or a feature vector X3: Inspection indicators When setting an indicator as invalid: First, cluster the inspection indicator types in organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 where the mandatory attribute is no to obtain the corresponding optional indicator categories G1, G2, and G3. Then, select any organ feature vector X1 that belongs to optional indicator category G1. Inspection indicators Set as invalid indicators, and select any blood feature vector X2 that falls within the optional indicator category G2. Inspection indicators Set as invalid indicators, and select any indicator from the Qi and Blood feature vector X3 that falls within the optional indicator category G3. Inspection indicators Set as invalid indicator.

[0071] Step 335, and denote the activation vector R generated during this process as the corresponding activation vector R0. u And the vector data r of the current activation vector R. m Let r be the corresponding r u,m And the activation vector R obtained this time u and its corresponding label vector Form a corresponding first prediction-label pair (R) u , ).

[0072] Step 34, obtain N tr The first prediction-label pair (R u , Substitute the first model loss function L1 into the preset value to calculate the corresponding first loss value.

[0073] Here, the first model loss function L1 in this embodiment of the invention is: .

[0074] Step 35: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 36; if it does not, modulate the model parameters of the health status prediction model in one round based on the preset first model optimizer in the direction of minimizing the first model loss function L1, and return to step 33 when the first round of modulation ends.

[0075] Here, the first loss value range in this embodiment of the invention is a pre-set numerical range. The first model optimizer includes the Adam optimizer and the SGD optimizer.

[0076] Step 36: Take each first data record of the first evaluation set as the current evaluation record. When the training round counter is 1, directly input the three feature vectors of the current evaluation record into the model for prediction processing. When the training round counter is 2, first randomly mask the three feature vectors of the current evaluation record according to the maximum masking percentage, and then input the three masked feature vectors into the model for prediction processing. The activation vector R generated in this process is used for prediction processing. v Its corresponding label vector Form a corresponding second prediction-label pair (R) v , ).

[0077] Specifically, it includes: Step 361: Take each of the first data records in the first evaluation set as the current evaluation record.

[0078] Step 362, and identify the training round counter.

[0079] Step 363: If the training round counter is 1, then input the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 of the current evaluation record into the health status prediction model for prediction processing.

[0080] Step 364: If the training round counter is 2, then calculate the corresponding three random occlusion numbers. , , And any of the organ feature vectors X1 recorded in the current assessment will be used. Inspection indicators Let the invalid index be any one of the blood feature vectors X2. Inspection indicators Let any of the following be an invalid indicator or a feature vector X3: Inspection indicators The invalid indicators are set as the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, which are set as invalid indicators, and then input into the health status prediction model for prediction processing.

[0081] It should also be noted here that the examination indicators in organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3... , , All of them have mandatory attributes, so, in the current evaluation record of any organ feature vector X1, Inspection indicators Let the invalid index be any one of the blood feature vectors X2. Inspection indicators Let any of the following be an invalid indicator or a feature vector X3: Inspection indicators When an indicator is set as invalid: any indicator in the organ feature vector X1 that falls within the optional indicator category G1 should be excluded. Inspection indicators Set as invalid indicators, and select any blood feature vector X2 that falls within the optional indicator category G2. Inspection indicators Set as invalid indicators, and select any indicator from the Qi and Blood feature vector X3 that falls within the optional indicator category G3. Inspection indicators Set as invalid indicator.

[0082] Step 365, and denote the activation vector R generated during this process as the corresponding activation vector R0. v And the vector data r of the current activation vector R. m Let r be the corresponding r v,m And the activation vector R obtained this time v and its corresponding label vector Form a corresponding second prediction-label pair (R) v , ).

[0083] Step 37, obtain N av The second prediction-label pair (R) v , Substitute the values ​​into the preset first model evaluation function L2 to calculate the corresponding first evaluation value.

[0084] Here, the first model evaluation function L2 in this embodiment of the invention is: .

[0085] Step 38: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 32; if it does, identify the training round counter; if the training round counter is 1, confirm the end of the first round of training, increment the training round counter by 1, and return to step 32; if the training round counter is 2, stop training and confirm the end of the second round of training.

[0086] Here, the first evaluation value range of this embodiment of the invention is a pre-set numerical range.

[0087] Step 4: After the model training is completed, the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 input by the user are input into the health status prediction model for prediction processing, and the health status prediction vector Y output by the model is fed back to the current user.

[0088] Figure 3This is a module structure diagram of a processing device for predicting health status based on multidimensional indicators, provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the aforementioned method embodiments, or it can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiments. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a model building module 201, a data acquisition module 202, a model training module 203, and a model application module 204.

[0089] Model building module 201 is used to build a deep learning model for predicting health status based on multi-dimensional indicators, denoted as the health status prediction model. The health status prediction model is used to perform health status prediction processing based on the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 input to the model and output the corresponding health status prediction vector Y. The organ feature vector X1 includes N1 examination indicators. , 1 ≤ index i ≤ N1, where N1 is the preset total number of first indicators, and each inspection indicator Corresponding to a class of Western medical examination indicators for human tissues and organs; the blood feature vector X2 includes N2 examination indicators. , 1 ≤ index j ≤ N2, where N2 is the preset total number of second indicators, and each inspection indicator Corresponding to a type of blood test index in Western medicine; the Qi and Blood Feature Vector X3 includes N3 test indicators. , 1 ≤ index k ≤ N3, where N3 is the total number of preset third indicators, and each inspection indicator Corresponding to a class of Qi and blood examination indicators in Traditional Chinese Medicine; the health status prediction vector Y consists of M health scores y m Composition, 1 ≤ index m ≤ M, where M is the preset total number of chronic diseases, and each health score y m Corresponding to a type of chronic disease; health score y m The score range is an integer range from 0 to 10.

[0090] Data acquisition module 202 is used to construct a first dataset through volunteer data collection and expert panel scoring; the first dataset includes multiple first data records; the first data records include organ feature vector X1, blood feature vector X2, gas and blood feature vector X3, and label vector R. * Label vector R * Score of M tags Composition; Score for each tag Corresponding to a type of chronic disease; label score The score range is the real number range from 0 to 1.

[0091] The model training module 203 is used to perform a first round of training on the health status prediction model based on the first dataset under the condition that the model input vector is unmasked, and then to perform a second round of training on the health status prediction model based on the first dataset under the condition that the model input vector is masked.

[0092] The model application module 204 is used to input the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 input by the user into the health status prediction model for prediction processing after the model training is completed, and to feed back the health status prediction vector Y output by the model to the current user.

[0093] The present invention provides a processing device for predicting health status based on multidimensional indicators, which can execute the method steps in the above method embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0094] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing elements; they can be fully implemented in hardware; or some modules can be implemented by processing elements calling software, while others are implemented in hardware. For example, the model building module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0095] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-on-a-Chip (SOC).

[0096] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The aforementioned computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the aforementioned computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The aforementioned computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0097] Figure 4 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be a terminal device or server implementing the methods of the aforementioned embodiments, or it can be a terminal device or server connected to the aforementioned terminal device or server implementing the methods of the aforementioned embodiments. Figure 4As shown, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for performing various processing functions and implementing the processing steps described in the foregoing embodiments. Preferably, the electronic device involved in the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0098] exist Figure 4 The system bus 305 mentioned can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include Non-Volatile Memory, such as at least one disk storage device.

[0099] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), graphics processing units (GPUs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0100] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to perform the methods and processes provided in the above embodiments.

[0101] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting health status based on multidimensional indicators. As described above, this invention constructs a health status prediction model based on multidimensional indicators. This model can predict health status based on multidimensional characteristics of both Western and Traditional Chinese Medicine examinations (Western medicine's human tissue and organ examination characteristics + Western medicine's blood examination characteristics + Traditional Chinese Medicine's Qi and blood examination characteristics). A first dataset is constructed through volunteer data collection and expert group scoring. A two-stage training strategy is employed when training the model based on the first dataset: the first round trains the model under unmasked input vectors, and the second round trains the model under masked input vectors. This invention improves the model's predictive performance by introducing multidimensional indicators and enhances its predictive performance when examination indicators are incomplete through a two-stage training strategy, thereby improving the model's generalization ability.

[0102] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0103] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting health status based on multidimensional indicators, characterized in that, The method includes: A deep learning model for predicting health status based on multidimensional indicators is constructed, denoted as the health status prediction model. This model is used to perform health status prediction processing based on the input organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, and outputs the corresponding health status prediction vector Y. The organ feature vector X1 includes N1 examination indicators. , 1 ≤ index i ≤ N1, where N1 is the preset total number of first indicators, and each of the inspection indicators Corresponding to a class of Western medical examination indicators for human tissues and organs; the blood feature vector X2 includes N2 examination indicators. ,1≤indexj≤N2, where N2 is the preset total number of second indicators, and each of the inspection indicators Corresponding to a type of blood test index in Western medicine; the Qi and Blood Feature Vector X3 includes N3 test indicators. , 1 ≤ index k ≤ N3, where N3 is the preset total number of third indicators, and each of the inspection indicators Corresponding to a class of Qi and blood examination indicators in Traditional Chinese Medicine; the health status prediction vector Y consists of M health scores y m Composition, 1 ≤ index m ≤ M, where M is the preset total number of chronic diseases, and each of the health scores y m Corresponding to a type of chronic disease; the health score y m The score range is the integer score range from 0 to 10; The first dataset is constructed through volunteer data collection and expert panel scoring; the first dataset includes multiple first data records; the first data record includes the organ feature vector X1, the blood feature vector X2, the qi and blood feature vector X3, and the label vector R. * The label vector R * Score of M tags Composition; score of each of the aforementioned tags Corresponding to a type of chronic disease; the label score The score range is the real number range from 0 to 1; The health status prediction model is trained in the first round on the first dataset under the condition that the model input vector is unmasked, and then trained in the second round on the first dataset under the condition that the model input vector is masked. After the model training is completed, the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 input by the user are input into the health status prediction model for prediction processing, and the health status prediction vector Y output by the model is fed back to the current user.

2. The processing method for predicting health status based on multidimensional indicators according to claim 1, characterized in that, The first model input terminal of the health status prediction model is used to receive the organ feature vector X1, the second model input terminal is used to receive the blood feature vector X2, and the third model input terminal is used to receive the qi and blood feature vector X3. The model output terminal is used to output the corresponding health status prediction vector Y. The health status prediction model includes a first feature extraction network, a second feature extraction network, a third feature extraction network, a first feature mapping layer, a second feature mapping layer, a third feature mapping layer, a feature fusion layer, and a linear regression layer. The input terminals of the first feature extraction network, the second feature extraction network, and the third feature extraction network are respectively connected to the corresponding first model input terminal, the second model input terminal, and the third model input terminal; the output terminals of the first feature extraction network, the second feature extraction network, and the third feature extraction network are respectively connected to the corresponding first feature mapping layer, the second feature mapping layer, and the third feature mapping layer input terminals; the output terminals of the first feature mapping layer, the second feature mapping layer, and the third feature mapping layer are respectively connected to the first input terminal, the second input terminal, and the third input terminal of the feature fusion layer; the output terminal of the feature fusion layer is connected to the input terminal of the linear regression layer; the output terminal of the linear regression layer is connected to the model output terminal. The first feature extraction network is implemented based on an MLP model structure; the first feature extraction network is used to extract features from the organ feature vector X1 to generate a corresponding feature vector H1 and send it to the first feature mapping layer; The reasoning method for the feature vector H1 is as follows: ； MLP1() is the inference function of the MLP model corresponding to the first feature extraction network, and θ1 is the model parameter of the current inference function; the vector shape of the feature vector H1 is N1×D1, and D1 is the preset first feature dimension; The second feature extraction network is implemented based on an MLP model structure; the second feature extraction network is used to extract features from the blood feature vector X2 to generate a corresponding feature vector H2 and send it to the second feature mapping layer; The reasoning method for the feature vector H2 is as follows: ； MLP2() is the inference function of the MLP model corresponding to the second feature extraction network, and θ2 is the model parameter of the current inference function; the vector shape of the feature vector H2 is N2×D2, and D2 is the preset second feature dimension; The third feature extraction network is implemented based on the MLP model structure; the third feature extraction network is used to extract features from the Qi and Blood feature vector X3 to generate the corresponding feature vector H3 and send it to the third feature mapping layer; The reasoning method for the feature vector H3 is as follows: ； MLP3() is the inference function of the MLP model corresponding to the third feature extraction network, and θ3 is the model parameter of the current inference function; the vector shape of the feature vector H3 is N3×D3, and D3 is the preset third feature dimension; The first feature mapping layer is used to map the feature vector H1 to a preset shared vector space to obtain the corresponding feature vector Z1, which is then sent to the feature fusion layer. The mapping method of the feature vector Z1 is as follows: ； W1 and b1 are the weight matrix and bias matrix of the first feature mapping layer; the shape of the weight matrix W1 is D. s ×D1,D s The shared vector feature dimension is preset; the bias matrix B1 has a shape of N1×D. s It consists of N1 elements of shape 1×D s Furthermore, the bias vector b1 is composed of identical parameters; the shape of the feature vector Z1 is N1×D. s ; The second feature mapping layer is used to map the feature vector H2 to the shared vector space to obtain the corresponding feature vector Z2, which is then sent to the feature fusion layer. The mapping method of the feature vector Z2 is as follows: ； W2 and B2 are the weight matrix and bias matrix of the second feature mapping layer; the shape of the weight matrix W2 is D. s ×D2; The shape of the bias matrix B2 is N2×D s It consists of N2 elements of shape 1×D s Furthermore, the bias vector b2 is composed of identical parameters; the shape of the feature vector Z2 is N2×D. s ; The third feature mapping layer is used to map the feature vector H3 to the shared vector space to obtain the corresponding feature vector Z3, which is then sent to the feature fusion layer. The mapping method of the feature vector Z3 is as follows: ； W3 and B3 are the weight matrix and bias matrix of the third feature mapping layer; the shape of the weight matrix W3 is D. s ×D3, the shape of the bias matrix B3 is N3×D s It consists of N3 elements of shape 1×D s Furthermore, the bias vector Z3 is composed of identical parameters; the shape of the feature vector Z3 is N3×D. s ; The feature fusion layer is used to concatenate the feature vectors Z1, Z2, and Z3 to obtain a vector of shape (N1+N2+N3)×D. s The feature vector Z4; then, each feature channel of the feature vector Z4 is pooled to obtain a 1×D feature vector. s The feature vector P is sent to the linear regression layer; The linear regression layer is used to perform a linear transformation on the feature vector P and activate the transformation result to obtain the corresponding activation vector R; and derives and outputs the corresponding health status prediction vector Y based on the activation vector R. The reasoning method for the activation vector R is as follows: ； W4 and b4 are the weight matrix and bias vector of the linear regression layer; the weight matrix W4 has an M×D shape. s The bias vector b4 has a shape of M×1; sigmoid() is the sigmoid activation function; the activation vector R has a shape of M×1, consisting of M vector data r m Composition, each of the vector data r m The range of values ​​for all of them is the range of real numbers between 0 and 1; The health scores y of the health status prediction vector Y m The derivation method is as follows: ； round() is the rounding function.

3. The processing method for predicting health status based on multidimensional indicators according to claim 1, characterized in that, The construction of the first dataset through volunteer data collection and expert panel scoring specifically includes: Using the M-type chronic diseases corresponding to the health status prediction vector Y as the recruitment population feature, multiple volunteers are recruited to form a volunteer group; and multiple medical experts are recruited to form an expert group; the total number of volunteer types in the volunteer group is M+1, consisting of healthy volunteers and volunteers with M-type chronic diseases; the expert group includes multiple specialists or general practitioners in the field of Western medicine, multiple specialists or general practitioners in the field of traditional Chinese medicine, and multiple specialists or general practitioners in the field of integrated traditional Chinese and Western medicine. Each volunteer in the volunteer group is designated as the current volunteer. With authorization from the current volunteer or their guardian, data is collected on the current volunteer's organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3. The expert group then assigns scores to the M tags corresponding to the M types of chronic diseases based on the current volunteer's organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3, according to an expert consensus scoring method. A score assessment is performed; and the obtained M tag scores are used to evaluate the scores. The corresponding label vector R is formed. * ; and composed of the organ feature vector X1, the blood feature vector X2, the gas and blood feature vector X3, and the label vector R corresponding to the current volunteer. * Form a corresponding first data record; The first dataset is composed of all the first data records obtained.

4. The processing method for predicting health status based on multidimensional indicators according to claim 1, characterized in that, The process of training the health status prediction model on the first dataset under the condition that the model input vector is unmasked, and then training the health status prediction model on the first dataset again under the condition that the model input vector is masked, specifically includes: Step 41: Initialize the training round counter to 1; Step 42: Based on a preset first segmentation ratio, the first dataset is randomly divided into two subsets, denoted as the first training set and the first evaluation set. Both the first training set and the first evaluation set consist of multiple first data records; the total number of records in the first training set is denoted as N. tr The total number of records in the first evaluation set is denoted as N. av The ratio N of the total number of records in the first training set to the total number of records in the first evaluation set. tr :N av Satisfying the first segmentation ratio; the label vector R of each of the first data records in the first training set. * Record as the corresponding The label vector R * The scores of each of the aforementioned tags Record as the corresponding 1 ≤ index u ≤ N tr The label vector R of each of the first data records in the first evaluation set * Record as the corresponding The label vector R * The scores of each of the aforementioned tags Record as the corresponding 1 ≤ index v ≤ N tr ; Step 43: Take each of the first data records in the first training set as the current training record. When the training round counter is 1, directly input the three feature vectors of the current training record into the model for prediction processing. When the training round counter is 2, first randomly mask the three feature vectors of the current training record according to a random masking percentage, and then input the masked three feature vectors into the model for prediction processing. The activation vector R generated during this process is then used for prediction processing. u Its corresponding label vector Form a corresponding first prediction-label pair (R) u , ); Specifically: each of the first data records in the first training set is used as the current training record; and the training round counter is identified; if the training round counter is 1, the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 of the current training record are input into the health status prediction model for prediction processing; if the training round counter is 2, a percentage is randomly generated between 10% and a preset percentage α1. A percentage is randomly generated between 10% and a preset percentage α2. A percentage is randomly generated between 10% and a preset percentage α3. And calculate the corresponding three random occlusion numbers. , , And any of the organ feature vectors X1 recorded in the current training record The aforementioned inspection indicators Set as an invalid indicator, any one of the blood feature vectors X2 The aforementioned inspection indicators Set as an invalid indicator, any one of the Qi and Blood feature vectors X3 The aforementioned inspection indicators The organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3, which have been set as invalid indicators, are input into the health status prediction model for prediction processing; and the activation vector R generated during this processing is recorded as the corresponding activation vector R. u And the vector data r of the current activation vector R. m Let r be the corresponding r u,m And the activation vector R obtained this time u and the corresponding label vector Form a corresponding first prediction-label pair (R) u , ); in, , , int() is the integer function; Step 44, obtain N tr The first predicted-label pair (R) u , Substitute the value into the preset first model loss function L1 to calculate the corresponding first loss value; Wherein, the loss function L1 of the first model is: ； Step 45: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 46; if it does not, modulate the model parameters of the health state prediction model in one round based on the preset first model optimizer in the direction of minimizing the first model loss function L1, and return to step 43 when the first round of modulation ends. The first model optimizer includes the Adam optimizer and the SGD optimizer. Step 46: Take each of the first data records in the first evaluation set as the current evaluation record, and when the training round counter is 1, directly input the three feature vectors of the current evaluation record into the model for prediction processing. When the training round counter is 2, first randomly mask the three feature vectors of the current evaluation record according to the maximum masking percentage, and then input the masked three feature vectors into the model for prediction processing. The activation vector R generated in this process is then used for prediction processing. v Its corresponding label vector Form a corresponding second prediction-label pair (R) v , ); Specifically: each of the first data records in the first evaluation set is taken as the current evaluation record; and the training round counter is identified; if the training round counter is 1, the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 of the current evaluation record are input into the health status prediction model for prediction processing; if the training round counter is 2, the corresponding three random occlusion numbers are calculated. , , and any of the organ feature vectors X1 recorded in the current evaluation record The aforementioned inspection indicators Set as an invalid indicator, any one of the blood feature vectors X2 The aforementioned inspection indicators Set as an invalid indicator, any one of the Qi and Blood feature vectors X3 The aforementioned inspection indicators The organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3, after the invalidation of the indicator, are set as invalid indicators and input into the health status prediction model for prediction processing; the activation vector R generated during this processing is recorded as the corresponding activation vector R. v And the vector data r of the current activation vector R. m Let r be the corresponding r v,m And the activation vector R obtained this time v and the corresponding label vector Form a corresponding second prediction-label pair (R) v , ); Step 47, obtain N av The second prediction-label pair (R) v , Substitute the value into the preset first model evaluation function L2 to calculate the corresponding first evaluation value; Wherein, the first model evaluation function L2 is: ； Step 48: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 42; if it does, identify the training round counter; if the training round counter is 1, confirm the end of the first round of training, increment the training round counter by 1, and return to step 42; if the training round counter is 2, stop training and confirm the end of the second round of training.

5. An apparatus for performing the processing method for predicting health status based on multidimensional indicators as described in any one of claims 1-4, characterized in that, The device includes: a model building module, a data acquisition module, a model training module, and a model application module; The model building module is used to construct a deep learning model for predicting health status based on multi-dimensional indicators, denoted as the health status prediction model. The health status prediction model is used to perform health status prediction processing based on the organ feature vector X1, blood feature vector X2, and qi and blood feature vector X3 input to the model, and output the corresponding health status prediction vector Y. The organ feature vector X1 includes N1 examination indicators. , 1 ≤ index i ≤ N1, where N1 is the preset total number of first indicators, and each of the inspection indicators Corresponding to a class of Western medical examination indicators for human tissues and organs; the blood feature vector X2 includes N2 examination indicators. ,1≤indexj≤N2, where N2 is the preset total number of second indicators, and each of the inspection indicators Corresponding to a type of blood test index in Western medicine; the Qi and Blood Feature Vector X3 includes N3 test indicators. , 1 ≤ index k ≤ N3, where N3 is the preset total number of third indicators, and each of the inspection indicators Corresponding to a class of Qi and blood examination indicators in Traditional Chinese Medicine; the health status prediction vector Y consists of M health scores y m Composition, 1 ≤ index m ≤ M, where M is the preset total number of chronic diseases, and each of the health scores y m Corresponding to a type of chronic disease; the health score y m The score range is the integer score range from 0 to 10; The data acquisition module is used to construct a first dataset through volunteer data collection and expert panel scoring; the first dataset includes multiple first data records; the first data record includes the organ feature vector X1, the blood feature vector X2, the qi and blood feature vector X3, and the label vector R. * The label vector R * Score of M tags Composition; score of each of the aforementioned tags Corresponding to a type of chronic disease; the label score The score range is the real number range from 0 to 1; The model training module is used to perform a first round of training on the health status prediction model based on the first dataset under the condition that the model input vector is unmasked, and then perform a second round of training on the health status prediction model based on the first dataset under the condition that the model input vector is masked. The model application module is used to input the organ feature vector X1, the blood feature vector X2, and the qi and blood feature vector X3 input by the user into the health status prediction model for prediction processing after the model training is completed, and to feed back the health status prediction vector Y output by the model to the current user.

6. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-4; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1-4.

Images