CN115116614A - Health state evaluation method, device, equipment and storage medium - Google Patents
Health state evaluation method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN115116614A CN115116614A CN202210790883.5A CN202210790883A CN115116614A CN 115116614 A CN115116614 A CN 115116614A CN 202210790883 A CN202210790883 A CN 202210790883A CN 115116614 A CN115116614 A CN 115116614A
- Authority
- CN
- China
- Prior art keywords
- record
- diagnosis
- diagnostic
- symptom
- moment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/126—Character encoding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Primary Health Care (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Epidemiology (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The application discloses a health state assessment method, a device, equipment and a storage medium, the application obtains diagnosis records of the same object at various moments, semantically encodes the diagnosis records at various moments to obtain symptom representation characteristics, determines attention weights corresponding to the diagnosis records according to time intervals of the diagnosis records relative to the specified assessment moments and the symptom representation characteristics of the diagnosis records, thereby considering the difference of the influence of different diagnosis records on the final assessment results in time sequence evolution, further endows the corresponding attention weights to the diagnosis records according to the attention weights, performs weighted addition on the symptom representation characteristics corresponding to the diagnosis records according to the attention weights of the diagnosis records to obtain final symptom representation characteristics which can better reflect symptom representation of the object, based on the health state of the determined object at the specified evaluation moment, the obtained result is more accurate.
Description
Technical Field
The present application relates to the field of smart medical technology, and more particularly, to a method, an apparatus, a device, and a storage medium for health status assessment.
Background
Diagnostic analysis of a subject is important for the assessment of the health status of a subject. In real life and work, the situation that the health state of a subject needs to be evaluated is often involved, for example, the risk that a patient may suffer from a certain disease is evaluated in real time, or the risk that equipment may have a certain abnormal fault is evaluated in real time, and the like.
Generally, when health status assessment is performed, analysis processing can be performed by means of a diagnosis record of a subject, wherein symptom description information in the whole history process of the subject is recorded in the diagnosis record and belongs to time sequence data with irregular time distribution, multiple sources and isomerism. Taking the patient's diagnosis record as an example, it may include admission record, medical order, examination and examination report, ward round record, consultation record, operation record, discharge record, etc. of the patient during the hospital. Further, the diagnostic record of the device is taken as an example, and may include an installation record, a patrol record, a maintenance record, and the like of the device.
In the prior art, when the health state of a subject is evaluated, generally, diagnostic records are treated equally, and differences of influences of different diagnostic records on the evaluation result of the final health state are ignored, so that the accuracy of the finally obtained health state result is not high.
Disclosure of Invention
In view of the above problems, the present application is proposed to provide a health status assessment method, apparatus, device and storage medium for achieving the purpose of accurately assessing the health status of a subject. The specific scheme is as follows:
in a first aspect, a health status assessment method is provided, including:
acquiring diagnostic records of the same object at all times, wherein the diagnostic records at all times contain symptom description information of the object at the time;
semantic coding is carried out on the diagnosis record at each moment to obtain symptom representation characteristics;
for the diagnosis record at each moment, determining an attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to a specified evaluation moment and the symptom representation characteristics of the diagnosis record, wherein the attention weight represents the influence of the diagnosis record on the evaluation of the health state of the subject;
according to the attention weight of the diagnosis record at each moment, carrying out weighted addition on the symptom representation characteristics corresponding to the diagnosis record at each moment to obtain final symptom representation characteristics;
determining a health state of the subject at the specified assessment time based on the end symptom representative features.
In a second aspect, a health status assessment apparatus is provided, comprising:
a diagnostic record acquisition unit, configured to acquire diagnostic records of the same object at different times, where the diagnostic record at each time includes symptom description information of the object at the time;
the semantic coding unit is used for carrying out semantic coding on the diagnosis record at each moment to obtain symptom representation characteristics;
an attention weight determination unit, which is used for determining an attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to a specified evaluation time and the symptom representation characteristics of the diagnosis record for the diagnosis record at each time, wherein the attention weight represents the influence of the diagnosis record on the evaluation of the health state of the object;
the attention weighting unit is used for weighting and adding the symptom representation characteristics corresponding to the diagnosis records at each moment according to the attention weights of the diagnosis records at each moment to obtain final symptom representation characteristics;
and the health state evaluation unit is used for determining the health state of the object at the specified evaluation moment based on the final symptom representation characteristics.
In a third aspect, a health status assessment apparatus is provided, comprising: a memory and a processor;
the memory is used for storing programs;
the processor is configured to execute the program to implement the steps of the health status assessment method.
In a fourth aspect, a storage medium is provided, on which a computer program is stored, which, when being executed by a processor, carries out the individual steps of the state of health assessment method as described above.
By the technical scheme, the method obtains the diagnosis records of the same object at all times, semantically encodes the diagnosis records at all times to obtain the symptom representation characteristics, and determines the attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to the specified evaluation time and the symptom representation characteristics of the diagnosis record for all the times, wherein the attention weight represents the influence of the diagnosis record on the health state of the evaluation object, so that the method considers the difference of the influence of different diagnosis records on the final evaluation result in time sequence evolution, further assigns the corresponding attention weight to each diagnosis record according to the attention weight, performs weighted addition on the symptom representation characteristics corresponding to each diagnosis record according to the attention weight of each diagnosis record to obtain the final symptom representation characteristics, and the final symptom representation characteristics can better reflect the symptom representation of the object, based on the health state of the determined object at the specified evaluation moment, the obtained result is more accurate.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the application. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
fig. 1 is a schematic flow chart of a health status evaluation method according to an embodiment of the present disclosure;
FIG. 2 illustrates a schematic diagram of a pre-trained model structure;
FIG. 3 illustrates a process diagram of diagnostic record sliding window selection;
fig. 4 is a schematic structural diagram of a health status assessment apparatus according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a health status assessment apparatus according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making any creative effort belong to the protection scope of the present application.
The application provides a health state assessment scheme, which is applicable to a scene of assessing the health state of various types of objects by analyzing the diagnosis records of the objects, wherein the objects can be patients or machine equipment. The corresponding health status may be a risk level of the patient suffering from a certain disease, or a risk level of the device suffering from a certain malfunction, etc.
The scheme can be realized based on a terminal with data processing capacity, and the terminal can be a mobile phone, a computer, a server, a cloud terminal and the like.
Next, as described with reference to fig. 1, the health status assessment method of the present application may include the following steps:
step S100, obtaining diagnosis records of the same object at various times, wherein the diagnosis record at each time comprises symptom description information of the object at the time.
Specifically, in order to assess the health status of a subject, a diagnostic record of the subject at various times is first obtained in the present application. Wherein symptom description information of the subject is recorded in the diagnosis record. It should be noted that there may be multiple diagnostic records, and the time corresponding to each diagnostic record may be different. Taking the subject as a patient as an example, the corresponding diagnosis record may be a clinical record of the patient, and the clinical record may include admission record, medical order, examination and examination report, ward round record, consultation record, operation record, discharge record, and the like of the patient during the hospital. Taking the target as the machine equipment as an example, the corresponding diagnosis records may include installation records, inspection records, maintenance records, and the like of the equipment.
As can be seen from the above, the time corresponding to each diagnostic record of the same object in the time dimension may be different, and taking the currently specified evaluation time as an example, the time interval between the time of each diagnostic record and the currently specified evaluation time is short or long. In order to more accurately evaluate the health state of the object, the diagnostic records of the object at various moments are obtained in the step, namely richer diagnostic records are obtained, and a sufficient data basis is provided for the subsequent accurate evaluation of the health state of the object.
Step S110, semantic coding is carried out on the diagnosis records at each moment to obtain symptom representation characteristics.
Specifically, for the diagnosis records at different time instants, semantic coding may be performed respectively to obtain symptom representation features corresponding to the diagnosis records. The diagnostic record may be encoded semantically in various ways, for example, the encoding module is used for semantic encoding, and the encoding module may be a pre-trained encoding module using diagnostic record training data.
Step S120, aiming at the diagnosis record at each moment, determining the attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to the specified evaluation moment and the symptom representation characteristics of the diagnosis record.
Wherein the attention weight represents a magnitude of an impact of the diagnostic record on assessing the health status of the subject.
The specified evaluation moment is the moment when the health state of the object needs to be evaluated. Alternatively, the specified evaluation time may be the time of the most recent diagnostic record in the diagnostic records. Based on this, the manner of determining the time interval of each diagnostic record with respect to the specified evaluation time may include: the time interval of each case record relative to the specified evaluation moment is calculated.
The time intervals of different diagnostic records from the specified evaluation time are possibly different, some intervals are longer, some intervals are shorter, the difference of the influence of the diagnostic records at different times on the time sequence evolution on the final evaluation result is considered in the application, and then the corresponding attention weight is given to each diagnostic record according to the difference, so that the influence of different diagnostic records on the health state of the evaluation object is reflected.
Step S130 is to perform weighted addition on the symptom-representing features corresponding to the diagnostic record at each time point according to the attention weight of the diagnostic record at each time point, and obtain a final symptom-representing feature.
Specifically, when the sum of the attention weights of the diagnosis records at each time is obtained, the symptom representation features corresponding to the diagnosis records at each time can be added in a weighted manner, so that the final symptom representation feature can be obtained, the final symptom representation feature can better reflect the symptom representation of the object, the health state of the object at the designated evaluation time can be determined based on the final symptom representation feature, and the obtained result is more accurate.
And step S140, determining the health state of the object at the specified evaluation moment based on the final symptom representation characteristics.
The method provided by the embodiment of the application obtains the diagnosis records of the same object at each moment, semantically encodes the diagnosis records at each moment to obtain the symptom representation characteristics, and determines the attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to the specified evaluation moment and the symptom representation characteristics of the diagnosis record, wherein the attention weight represents the influence of the diagnosis record on the health state of the evaluation object, so that the method considers the difference of the influence of different diagnosis records on the final evaluation result in time sequence evolution, further gives the corresponding attention weight to each diagnosis record according to the attention weight, and performs weighted addition on the symptom representation characteristics corresponding to each diagnosis record according to the attention weight of each diagnosis record to obtain the final symptom representation characteristics which can better reflect the symptom representation of the object, based on the health state of the determined object at the specified evaluation moment, the obtained result is more accurate.
In some embodiments of the present application, step S110 is described as a process of semantically encoding the diagnosis record at each moment to obtain symptom representation characteristics.
The embodiment provides an implementation mode for carrying out semantic coding on a diagnosis record through a pre-training coding module.
In view of the scarcity of training data carrying labels, in order to ensure the effect of the downstream health status evaluation model, in this embodiment, the coding module may be trained in an unsupervised pre-training manner to obtain an effective initial semantic representation vector, that is, a symptom representation feature.
The structure of the pre-training model can be seen with reference to fig. 2:
the pre-training module may include an encoding module and a prediction module, wherein the prediction module may adopt a neural network structure such as LSTM or RNN.
The prediction module is used for predicting the coding vector of the symptom description sentence at the t + k th moment based on the coding vectors of the symptom description sentences at the t moment and the previous moments.
By pre-training, the encoding module can be made to learn features related to health status. On the basis, considering that the health status of the subject is a quantity varying with time, it is desirable in this embodiment that the encoding module can further learn the correlation of the disease/fault condition on the time evolution based on learning the characteristics related to the health status, so that the symptom representation characteristics encoded by the encoding module can reflect the correlation between the symptom description contents in different diagnostic records.
As shown in fig. 2, x ═ x is defined 1 ，x 2 ，...，x N ]The symptom in the diagnostic record at each time representing one object describes a sentence, and N represents the total number of sentences.
The input x is represented as z by the coding vector obtained after passing through the coding module:
z＝enc(x)
enc () represents any coding function, such as BERT, CNN, etc.
The input of the prediction module is a coding vector of a symptom description sentence at any time t and each time before, and the prediction module predicts a characteristic vector c based on the input t 。
It will be appreciated that if the vector c is characterized t The coding vector z describing the sentence with the symptom at a time after t (for example, at t + k) t+k The more similar, the more the correlation of the previous and subsequent disease conditions/failure conditions on the time sequence evolution is learned by the pre-training model, that is, the information in the original diagnosis record can be fully expressed by the symptom representing characteristics coded by the coding module for the input sentence, and the preliminary prediction capability is also provided, that is, the capability of predicting the change of the disease conditions/failure conditions after the prediction is provided, and the symptom representing characteristics coded by the coding module at the time are considered to be excellent.
In order to achieve the above purpose, the method may be implemented as follows when training a pre-training model:
and S1, acquiring a training sample set, wherein each training sample comprises a symptom description sentence in the diagnosis record of each moment of a subject.
And S2, training the pre-training model by using the training sample set and adopting a contrast learning strategy, wherein the training process comprises the steps that positive example pairs are formed by symptom description sentences of the same object at different moments, and the negative example pairs are formed by the symptom description sentences of different objects.
In an alternative case, the trained loss function may be referred to as:
where X represents a set of training samples, X t+k Representing sentences at time t + k, x, of a sample j Is equal to x t+k Any sentence from different samples, z t+k X input for said pair of coding modules t+k The encoded vector obtained after encoding, T represents the vector transposition, z j X input for said pair of coding modules j A coded vector, W, obtained after coding k As a network parameter, c t Basing the prediction module on x t+k The coding vectors of sentences at the t-th time and the previous times in the sample to which the sentence belongs, and the predicted characterization vectors.
Training the pre-training model according to the loss function, when a set training end condition is reached, extracting a coding module in the pre-training model, and performing semantic coding on the diagnosis record of the object acquired in the step S100 at each moment by using the coding module to obtain a symptom representation feature.
Further, considering that in practical applications, the diagnostic record of the subject may contain many contents, and some of the contents may have no effect on evaluating the health status of the subject, that is, the diagnostic record may contain a large amount of redundant information, if the diagnostic record is directly used for semantic coding, it is difficult to screen out key features related to the evaluation of the health status of the subject from the large amount of features, and the computational complexity may be increased. Therefore, the embodiment of the application provides a scheme for simplifying the diagnosis record so as to extract the key information from the diagnosis record.
Specifically, before performing semantic coding on the diagnosis record at each time to obtain the symptom representation feature in the foregoing step S110, a process of performing compaction processing on the diagnosis record at each time may be further added in this embodiment to obtain a processed diagnosis record that only includes symptom description information related to the health issue to be evaluated.
In addition, in step S110, semantic coding is performed on the processed diagnosis record at each time to obtain symptom representation characteristics.
The health problem to be evaluated may be a measure of the health status of the subject, for example, when the subject is a patient, the problem to be evaluated may be a disease to be evaluated, that is, a measure of the risk of the patient suffering from the disease to be evaluated. For another example, when the object is a machine device, the problem to be evaluated may be a fault type to be evaluated, that is, the risk of the machine device having a fault of the fault type to be evaluated is measured.
By simplifying the diagnosis records at each moment, redundant information irrelevant to the health state of an evaluation object is removed from the processed diagnosis records, only symptom description information relevant to the health problem to be evaluated is reserved, the calculation complexity is simplified, meanwhile, the accuracy of the symptom representation characteristics after semantic coding is improved, and the health state of the object obtained by subsequent calculation is more accurate.
In some embodiments of the present application, a specific implementation manner of performing compaction processing on a diagnostic record is introduced, and the detailed steps are as follows:
s1, aiming at the diagnosis record at each moment, performing initial semantic coding on each word in the diagnosis record, and coding the risk factor label of the health problem to be evaluated.
Specifically, the risk factor label of the health problem to be evaluated may be preset, and includes labels of various types of risk factors of the subject presenting the health problem to be evaluated. Taking the health problem to be evaluated as an example of a tracheitis disease, the various types of risk factor labels may include: cough, phlegm, asthma, etc.
In this step, the initial semantic coding is performed on each word in the diagnostic record, and the process of coding the risk factor label of the health problem to be evaluated can be performed by using the pre-trained coding module.
S2, determining the correlation degree of each word and the health problem to be evaluated based on the coding characteristics of the risk factor labels of the health problems to be evaluated and the coding characteristics of each word in the diagnostic record.
Optionally, in this embodiment, an attention mechanism may be adopted to determine the correlation between each sentence and the health problem to be evaluated.
Specifically, the coding features of each sentence in the diagnostic record and the coding features of the risk factor labels are calculated through an attention mechanism, and the correlation degree of each sentence and the health problem to be evaluated is predicted:
wherein, Score (O) i |Q m ) Representing the relevance score, O, of the mth sentence in the diagnostic record relative to the ith risk factor label i Indicates the ith risk factor tag, Q m Representing the mth sentence in the diagnostic record,
respectively represent the coding characteristics of the ith risk factor label and the jth risk factor label,
respectively representing the coding characteristics of every m and nth sentences in the diagnosis record, and W represents a network parameter.
In this step, for each sentence in the diagnosis record, the highest score in the relevancy scores between the sentence and each risk factor label may be used as the relevancy between the sentence and the health problem to be evaluated.
And S3, based on the correlation degree between each sentence and the health problem to be evaluated, screening out sentences of which the correlation degrees meet set requirements from the diagnosis records to obtain processed diagnosis records.
Specifically, a relevance threshold may be preset, and then sentences in the diagnostic record, the relevance of which to the health problem to be assessed is greater than the set relevance threshold, are screened out to form a processed diagnostic record.
In the scheme of the embodiment, the sentences with the correlation degrees meeting the set requirements are screened out by combining the correlation degrees of each sentence in the diagnosis record and the risk factor label of the problem to be evaluated to form the processed diagnosis record, so that the information irrelevant to the health state of the evaluated object in the diagnosis record can be effectively removed, the content of the diagnosis record is simplified, the calculation complexity is reduced, and meanwhile, the key characteristics relevant to the evaluation of the health state of the object are reserved.
Further, considering that the diagnostic records of the subject at different times may be redundant and complicated, and the diagnostic records at different times may include many repetitive descriptions, taking the health status of the patient as an example, the diagnostic records may include admission records, first-time course records, examination reports, ward rounds, etc., as shown in table 1 below:
TABLE 1
As can be seen from table 1 above, there may be a large number of repetitive descriptions in different diagnostic records, such as admission records, first-time course records, and ward-round records.
If all diagnostic records are considered computationally, it consumes more computational resources and does not help the final evaluation result. For this reason, the present embodiment further provides a scheme for screening diagnostic records, specifically, in step S110, before semantic coding is performed on the diagnostic records at each time to obtain symptom representation features, the following steps are further added:
and screening the diagnostic records at each moment to obtain a plurality of screened diagnostic records. Step S110 may specifically include:
and semantic coding is carried out on the screened diagnosis records at each moment to obtain symptom representation characteristics.
In the embodiment, when the diagnostic records are screened, the closer to the specified evaluation time, the more diagnostic records are reserved, and the farther from the specified evaluation time, the fewer diagnostic records are reserved.
Therefore, the embodiment provides a sliding window screening method, specifically:
several diagnostic records can be screened from the diagnostic records at each moment by adopting a sliding window mode, and the window size of the sliding window is gradually increased and/or the sliding step length is gradually reduced according to the sequence that the time of each diagnostic record and the interval of the specified evaluation moment are from long to short.
Examples are as follows:
the obtained diagnosis records at each time are defined to be N parts, and finally the reserved diagnosis records to be screened are max _ num parts.
The previous (max _ num/2) diagnostic records are selected in order of short to long interval between the time of each diagnostic record and the designated evaluation time. And (max _ num/2) diagnostic records are selected from the rest (N-max _ num/2) diagnostic records.
The selection of (max _ num/2) diagnostic records from the remaining (N-max _ num/2) diagnostic records may be performed by uniform sampling at time intervals or by other methods, such as random selection.
Referring to fig. 3, an alternative screening approach for diagnostic records is illustrated. If all diagnostic records are 7 parts in total, and the diagnostic records retained in the final screening are defined as 4 parts, the 2 parts closest in time can be selected first, then one part is selected every 1 part in the remaining 5 parts, and then 2 parts are selected, so as to obtain 4 parts of diagnostic records selected by the final sliding window.
In some embodiments of the present application, the process of determining the attention weight corresponding to the diagnostic record in step S120 is described.
In this embodiment, the process of determining the attention weight and determining the final symptom representation feature is realized by setting an attention module.
The attention module captures the interdependence relation between the symptom representation characteristics of the diagnosis records at different moments based on the attention mechanism, on the basis, the time interval of each diagnosis record from the specified evaluation moment is considered, the influence of the diagnosis records at different time intervals on the final evaluation result is measured, and the determined final symptom representation characteristics are more accurate.
The process of specifically determining the attention weight may include:
and calculating the attention weight of the diagnostic record by using an attention module according to the symptom representation characteristics of the diagnostic record at each moment, wherein the time interval of the diagnostic record relative to the specified evaluation moment is taken as an attenuation term in the calculation process, and the larger the time interval is, the lower the calculated attention weight is.
It will be appreciated that, in general, diagnostic records closer to a given evaluation time have a greater effect on the evaluation of the health status of the subject, whereas diagnostic records further from the given evaluation time have a lesser effect on the evaluation of the health status of the subject. Based on this, when calculating the attention weight of the diagnostic record, the present application introduces an attenuation term of time interval, and as the time interval increases, the corresponding attenuation amount increases, that is, the finally calculated attention weight decreases.
In consideration of the time sequence of the diagnosis record, the attention module may perform a time-sequence preliminary encoding on the symptom-representing characteristics of the diagnosis record at each time (the diagnosis record may be encoded by using the aforementioned pre-trained encoding module to obtain the symptom-representing characteristics), for example, the time-sequence preliminary encoding is performed by using a network such as LSTM or RNN. Taking the primary coding by adopting an LSTM network as an example to obtain a coding characteristic H' R ：H′ R ＝h 1 ，h 2 ，...，h T ＝LSTM(H R )
Where T denotes the time of the diagnostic record closest to the specified evaluation instant, H R A symptom representative characteristic representing the diagnostic record.
The conventional attention mechanism is calculated according to the following formula:
Q＝W Q H′ R
K＝W K H′ R
V＝W V H′ R
Z R ＝Score*V
wherein, W Q 、W K 、W V Is three network parameters, Q, K, V denotes query, key and value in attention mechanism calculation, by encoding feature H' R Multiplying the three different network parameters respectively to obtain vector representations of three different spaces for calculating attention weight; score denotes attention weight, d k To set the parameters such that the variance stabilizes to 1, the gradient of softmax is not too small, Z R The final symptom representation characteristics after attention weighting are shown.
It can be seen that the attention weight of each diagnostic record calculated according to the conventional attention mechanism does not take into account the difference in the time interval between each diagnostic record and the specified evaluation time, and the influence on the final evaluation result is different. For this reason, an improved attention mechanism calculation method is proposed in this embodiment, further taking into account the time interval Δ t between the diagnostic record and the specified evaluation time, and at the same time, a trainable parameter β may be further added to control the influence of the time interval Δ t on the attention weight, and the improved attention weight calculation formula may be as follows:
wherein, Score t Representing the attention weight, q, of the diagnostic record at time t t The query feature of the diagnostic record at time t can be calculated by referring to the formula Q, k T And the key characteristic of the diagnostic record at each moment can be calculated by referring to the calculation formula of K in the preamble, wherein sigma () represents a sigmoid function, beta is a trainable parameter and is used for controlling the influence of a time interval on the attention weight, and delta t represents the time interval between the t-th moment and a specified evaluation moment.
As can be seen from the above formula, if the time interval Δ t between the time of the diagnostic record and the specified evaluation time is long, the attention weight of the calculated diagnostic record is significantly attenuated, whereas when the time interval Δ t is small, for example, Δ t is 0, the denominator part is equal to β, i.e., the attention weight is only slightly attenuated.
According to the above calculation method, attention weights of diagnostic records at various times can be calculated as follows:
Score＝(Score 1 ，Score 2 ，...，Score T )
in addition, in step S130, the process of obtaining the final symptom expression characteristic by performing weighted addition on the symptom expression characteristics corresponding to the diagnosis record at each time according to the attention weight of the diagnosis record at each time can be calculated according to the following formula:
Z R ＝Score*V
wherein Z is R Indicating the final symptom-indicating characteristic, calculatedReference is made to the preceding description.
The following describes the health status assessment device provided in the embodiments of the present application, and the health status assessment device described below and the health status assessment method described above may be referred to correspondingly.
Referring to fig. 4, fig. 4 is a schematic structural diagram of a health status assessment apparatus disclosed in the embodiment of the present application.
As shown in fig. 4, the apparatus may include:
a diagnostic record obtaining unit 11, configured to obtain diagnostic records of the same object at different times, where the diagnostic record at each time includes symptom description information of the object at the time;
a semantic coding unit 12, configured to perform semantic coding on the diagnosis record at each time to obtain symptom expression features;
an attention weight determination unit 13, configured to determine, for each diagnostic record at a time, an attention weight corresponding to the diagnostic record according to a time interval of the diagnostic record relative to a specified evaluation time and a symptom representation characteristic of the diagnostic record, where the attention weight represents a magnitude of an influence of the diagnostic record on evaluating a health state of the subject;
an attention weighting unit 14 for performing weighted addition on the symptom-representing features corresponding to the diagnosis records at each time point according to the attention weights of the diagnosis records at each time point to obtain final symptom-representing features;
a health status evaluation unit 15 for determining the health status of the subject at the specified evaluation moment based on the final symptom representation characteristics.
Optionally, the apparatus of the present application may further include:
the diagnostic record processing unit is used for processing the diagnostic record at each moment before the semantic coding unit processes the diagnostic record to obtain a processed diagnostic record only containing symptom description information related to the health problem to be evaluated;
the semantic coding unit performs semantic coding on the diagnosis record at each moment to obtain a process of symptom representation characteristics, which specifically includes:
and carrying out semantic coding on the processed diagnosis record at each moment to obtain symptom representation characteristics.
Optionally, the process of processing the diagnosis record at each moment by the diagnosis record processing unit to obtain the processed diagnosis record only containing the symptom description information related to the health issue to be evaluated may include:
aiming at the diagnosis record at each moment, performing initial semantic coding on each sentence in the diagnosis record, and coding the risk factor label of the health problem to be evaluated;
determining the correlation degree of each word and the health problem to be evaluated based on the coding features of the risk factor labels of the health problems to be evaluated and the coding features of each word in the diagnostic record;
and screening out sentences of which the correlation degrees meet set requirements from the diagnosis records based on the correlation degrees of each sentence and the health problems to be evaluated to obtain the processed diagnosis records.
Optionally, the process of determining the attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to the specified evaluation time and the symptom representation feature of the diagnosis record for the diagnosis record at each time by the attention weight determining unit may include:
and calculating the attention weight of the diagnostic record by using an attention module according to the symptom representation characteristics of the diagnostic record at each moment, wherein the time interval of the diagnostic record relative to the specified evaluation moment is taken as an attenuation term in the calculation process, and the larger the time interval is, the lower the calculated attention weight is.
Optionally, the apparatus of the present application may further include:
the diagnostic record screening unit is used for screening a plurality of diagnostic records from the diagnostic records at each moment in a sliding window mode before the processing of the semantic coding unit, and the window size of the sliding window is gradually increased and/or the sliding step length is gradually reduced according to the sequence of the interval between the time of each diagnostic record and the appointed evaluation moment from long to short;
the semantic coding unit performs semantic coding on the diagnosis record at each moment to obtain a process of symptom representation characteristics, which specifically includes:
and semantic coding is carried out on the screened diagnosis records at each moment to obtain symptom representation characteristics.
Optionally, the semantic coding unit performs semantic coding on the diagnosis record at each time to obtain the symptom representation feature, and the process may include:
and performing semantic coding on the diagnosis record at each moment by adopting a pre-trained coding module to obtain symptom representation characteristics.
Optionally, the pre-training process of the coding module may include:
acquiring a training sample set, wherein each training sample comprises a symptom description sentence in a diagnosis record of a subject at each moment;
training a pre-training model by using the training sample set and a comparison learning strategy, wherein the symptom description sentences of the same object at different moments form a positive example pair and the symptom description sentences of different objects form a negative example pair during training; the pre-training model comprises the encoding module and a prediction module, wherein:
the prediction module is used for predicting the coding vector of the symptom description sentence at the t + k th moment based on the coding vectors of the symptom description sentences at the t moment and the previous moments.
Optionally, the trained loss function is as follows:
where X represents a set of training samples, X t+k Representing sentences at time t + k, x, of a sample j Is equal to x t+k Any sentence from different samples, z t+k X input for said pair of coding modules t+k The encoded vector obtained after encoding, T represents the vector transposition, z j X input for said pair of coding modules j A coded vector, W, obtained after coding k As a network parameter, c t For the prediction module to be based on x t+k The coding vectors of sentences at the t-th time and the previous times in the sample to which the sentence belongs, and the predicted characterization vectors.
The health status evaluation device provided by the embodiment of the application can be applied to health status evaluation equipment, such as a terminal: mobile phones, computers, etc. Alternatively, fig. 5 shows a block diagram of a hardware structure of the health status evaluation device, and referring to fig. 5, the hardware structure of the health status evaluation device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;
in the embodiment of the application, the number of the processor 1, the communication interface 2, the memory 3 and the communication bus 4 is at least one, and the processor 1, the communication interface 2 and the memory 3 complete mutual communication through the communication bus 4;
the processor 1 may be a central processing unit CPU, or an application Specific Integrated circuit asic, or one or more Integrated circuits configured to implement embodiments of the present invention, etc.;
the memory 3 may include a high-speed RAM memory, and may further include a non-volatile memory (non-volatile memory) or the like, such as at least one disk memory;
wherein the memory stores a program and the processor can call the program stored in the memory, the program for:
acquiring diagnostic records of the same object at all times, wherein the diagnostic records at all times contain symptom description information of the object at the time;
semantic coding is carried out on the diagnosis record at each moment to obtain symptom representation characteristics;
for the diagnosis record at each moment, determining an attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to a specified evaluation moment and the symptom representation characteristics of the diagnosis record, wherein the attention weight represents the influence of the diagnosis record on the evaluation of the health state of the subject;
according to the attention weight of the diagnosis record at each moment, carrying out weighted addition on the symptom representation characteristics corresponding to the diagnosis record at each moment to obtain final symptom representation characteristics;
determining a health state of the subject at the specified assessment time based on the final symptom representative characteristics.
Alternatively, the detailed function and the extended function of the program may be as described above.
Embodiments of the present application further provide a storage medium, where a program suitable for execution by a processor may be stored, where the program is configured to:
acquiring diagnostic records of the same object at all times, wherein the diagnostic records at all times contain symptom description information of the object at the time;
semantic coding is carried out on the diagnosis record at each moment to obtain symptom representation characteristics;
for the diagnosis record at each moment, determining an attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to a specified evaluation moment and the symptom representation characteristics of the diagnosis record, wherein the attention weight represents the influence of the diagnosis record on the evaluation of the health state of the subject;
according to the attention weight of the diagnosis record at each moment, carrying out weighted addition on the symptom representation characteristics corresponding to the diagnosis record at each moment to obtain final symptom representation characteristics;
determining a health state of the subject at the specified assessment time based on the final symptom representative characteristics.
Alternatively, the detailed function and the extended function of the program may be as described above.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, the embodiments may be combined as needed, and the same and similar parts may be referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (12)
1. A health state assessment method, comprising:
acquiring diagnostic records of the same object at all times, wherein the diagnostic records at all times contain symptom description information of the object at the time;
semantic coding is carried out on the diagnosis record at each moment to obtain symptom representation characteristics;
for the diagnosis record at each moment, determining an attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to a specified evaluation moment and the symptom representation characteristics of the diagnosis record, wherein the attention weight represents the influence of the diagnosis record on the evaluation of the health state of the subject;
according to the attention weight of the diagnosis record at each moment, carrying out weighted addition on the symptom representation characteristics corresponding to the diagnosis record at each moment to obtain final symptom representation characteristics;
determining a health state of the subject at the specified assessment time based on the final symptom representative characteristics.
2. The method of claim 1, wherein prior to semantically encoding the diagnostic record at each time instant to obtain symptom-indicative features, further comprising:
processing the diagnosis record at each moment to obtain a processed diagnosis record only containing symptom description information related to the health problem to be evaluated;
a process of semantically encoding the diagnostic record at each time point to obtain symptom representation characteristics, comprising:
and carrying out semantic coding on the processed diagnosis record at each moment to obtain symptom representation characteristics.
3. The method of claim 2, wherein processing the diagnostic record at each time to obtain a processed diagnostic record containing only symptom descriptive information related to the health issue being assessed comprises:
aiming at the diagnosis record at each moment, performing initial semantic coding on each sentence in the diagnosis record, and coding the risk factor label of the health problem to be evaluated;
determining the correlation degree of each word and the health problem to be evaluated based on the coding features of the risk factor labels of the health problems to be evaluated and the coding features of each word in the diagnostic record;
and screening out sentences of which the correlation degrees meet set requirements from the diagnosis records based on the correlation degrees of each sentence and the health problems to be evaluated to obtain the processed diagnosis records.
4. The method of claim 1, wherein determining, for each diagnostic record at a time, an attention weight corresponding to the diagnostic record based on a time interval of the diagnostic record relative to a specified evaluation time and a symptom-representative characteristic of the diagnostic record comprises:
and calculating the attention weight of the diagnostic record by using an attention module according to the symptom representation characteristics of the diagnostic record at each moment, wherein the time interval of the diagnostic record relative to the specified evaluation moment is taken as an attenuation term in the calculation process, and the larger the time interval is, the lower the calculated attention weight is.
5. The method of claim 1, wherein prior to said semantically encoding the diagnostic record at each time instant to obtain symptom representation features, the method further comprises:
screening a plurality of diagnostic records from the diagnostic records at each moment in a sliding window mode, and gradually increasing the window size of a sliding window and/or gradually reducing the sliding step length according to the sequence of the interval between the time of each diagnostic record and the specified evaluation moment from long to short;
then said semantically encoding the diagnostic record at each time to obtain symptom representation features, including:
and semantic coding is carried out on the screened diagnosis records at each moment to obtain symptom representation characteristics.
6. The method of claim 1, wherein semantically encoding the diagnostic record at each time point to obtain symptom-indicative features comprises:
and performing semantic coding on the diagnosis record at each moment by adopting a pre-trained coding module to obtain symptom representation characteristics.
7. The method of claim 6, wherein the pre-training process of the coding module comprises:
acquiring a training sample set, wherein each training sample comprises a symptom description sentence in a diagnosis record of a subject at each moment;
training a pre-training model by using the training sample set and a comparison learning strategy, wherein the symptom description sentences of the same object at different moments form a positive example pair and the symptom description sentences of different objects form a negative example pair during training;
the pre-training model comprises the encoding module and a prediction module, wherein:
the prediction module is used for predicting the coding vector of the symptom description sentence at the t + k th moment based on the coding vectors of the symptom description sentences at the t moment and the previous moments.
8. The method of any one of claims 1-7, wherein the specified evaluation time is the time of the most recent diagnostic record in each diagnostic record;
the manner in which the time interval of each diagnostic record relative to a given evaluation time is determined includes:
the time interval of each case record relative to the specified evaluation time is calculated.
9. The method of any one of claims 1-7, wherein the subject is a user and the diagnostic record is a clinical record of the user;
or the like, or, alternatively,
the object is a device and the diagnostic record is a service record of the device.
10. A state of health assessment apparatus, comprising:
a diagnostic record acquisition unit, configured to acquire diagnostic records of the same object at different times, where the diagnostic record at each time includes symptom description information of the object at the time;
the semantic coding unit is used for carrying out semantic coding on the diagnosis record at each moment to obtain symptom representation characteristics;
an attention weight determination unit, which is used for determining an attention weight corresponding to the diagnosis record according to the time interval of the diagnosis record relative to a specified evaluation time and the symptom representation characteristics of the diagnosis record for the diagnosis record at each time, wherein the attention weight represents the influence of the diagnosis record on the evaluation of the health state of the object;
the attention weighting unit is used for weighting and adding the symptom representation characteristics corresponding to the diagnosis records at each moment according to the attention weights of the diagnosis records at each moment to obtain final symptom representation characteristics;
and the health state evaluation unit is used for determining the health state of the object at the specified evaluation moment based on the final symptom representation characteristics.
11. A health state evaluation apparatus characterized by comprising: a memory and a processor;
the memory is used for storing programs;
the processor is configured to execute the program to implement the steps of the health status assessment method according to any one of claims 1 to 9.
12. A storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of the state of health assessment method according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210790883.5A CN115116614A (en) | 2022-07-05 | 2022-07-05 | Health state evaluation method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210790883.5A CN115116614A (en) | 2022-07-05 | 2022-07-05 | Health state evaluation method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115116614A true CN115116614A (en) | 2022-09-27 |
Family
ID=83332890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210790883.5A Pending CN115116614A (en) | 2022-07-05 | 2022-07-05 | Health state evaluation method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115116614A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117637153A (en) * | 2024-01-23 | 2024-03-01 | 吉林大学 | Informationized management system and method for patient safety nursing |
-
2022
- 2022-07-05 CN CN202210790883.5A patent/CN115116614A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117637153A (en) * | 2024-01-23 | 2024-03-01 | 吉林大学 | Informationized management system and method for patient safety nursing |
| CN117637153B (en) * | 2024-01-23 | 2024-03-29 | 吉林大学 | Informationized management system and method for patient safety nursing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111191791B (en) | Picture classification method, device and equipment based on machine learning model | |
| CN111460101B (en) | Knowledge point type identification method, knowledge point type identification device and knowledge point type identification processor | |
| CN111160959B (en) | User click conversion prediction method and device | |
| CN111784595B (en) | Dynamic tag smooth weighting loss method and device based on historical record | |
| CN110379521B (en) | Medical data set feature selection method based on information theory | |
| CN113723007A (en) | Mechanical equipment residual life prediction method based on DRSN and sparrow search optimization BilSTM | |
| CN112489740A (en) | Medical record detection method, training method of related model, related equipment and device | |
| Ma et al. | Rare Disease Prediction by Generating Quality-Assured Electronic Health Records∗ | |
| CN115116614A (en) | Health state evaluation method, device, equipment and storage medium | |
| CN115579141A (en) | Interpretable disease risk prediction model construction method and disease risk prediction device | |
| CN111144473B (en) | Training set construction method, training set construction device, electronic equipment and computer readable storage medium | |
| CN116384223A (en) | Nuclear equipment reliability assessment method and system based on intelligent degradation state identification | |
| CN115660060A (en) | Model training method, detection method, device, equipment and storage medium | |
| US20210241147A1 (en) | Method and device for predicting pair of similar questions and electronic equipment | |
| CN110334353B (en) | Analysis method, device, equipment and storage medium for word sequence recognition performance | |
| CN115188475A (en) | Risk prediction method for lupus nephritis patient | |
| CN112529637B (en) | Service demand dynamic prediction method and system based on context awareness | |
| CN114372618A (en) | Student score prediction method and system, computer equipment and storage medium | |
| CN114444654A (en) | NAS-oriented training-free neural network performance evaluation method, device and equipment | |
| CN114388128A (en) | Method, apparatus, device and medium for detecting personal motion intelligent PAI index | |
| CN116415137B (en) | Emotion quantification method, device, equipment and storage medium based on multi-modal characteristics | |
| CN116610783B (en) | Service optimization method based on artificial intelligent decision and digital online page system | |
| CN110703924A (en) | Cold start method and equipment for new user based on input method application | |
| CN117197613B (en) | Image quality prediction model training method and device and image quality prediction method and device | |
| CN110728615B (en) | Steganalysis method based on sequential hypothesis testing, terminal device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 230000 floor 23-24, building A5, No. 666, Wangjiang West Road, high tech Zone, Hefei City, Anhui Province Applicant after: IFLYTEK Medical Technology Co.,Ltd. Address before: 230088 floor 23-24, building A5, No. 666, Wangjiang West Road, high tech Zone, Hefei, Anhui Province Applicant before: Anhui Xunfei Medical Co.,Ltd. |
|
| CB02 | Change of applicant information |