CN109887606B - Attention-based diagnosis and prediction method for bidirectional recurrent neural network - Google Patents

Abstract

The invention discloses a diagnosis and prediction method of a bidirectional recurrent neural network based on attention, and relates to the technical field of prediction and diagnosis. The method first embeds high-dimensional medical codes (i.e. clinical variables) into a low-code layer space, and then inputs the coded representation into an attention-based bidirectional recurrent neural network to generate a hidden state representation. The hidden representation is entered via the softmax layer to predict future access to the medical code. Experimental data show that by adopting the method provided by the embodiment, when future access information is predicted, different weights can be distributed to previous accesses by an attention mechanism, so that not only can a diagnosis and prediction task be effectively completed, but also a prediction result can be reasonably explained.

Description

Attention-based diagnosis and prediction method for bidirectional recurrent neural network

Technical Field

The invention relates to the technical field of prediction diagnosis, in particular to a diagnosis prediction method of a bidirectional recurrent neural network based on attention.

Background

Electronic Health Records (EHRs) consist of longitudinal patient Health data including a sequence of visits over time, with each visit containing multiple medical codes, including diagnosis, medication and program codes, that have been successfully applied to several predictive modeling tasks in healthcare. EHR data consists of a set of high-dimensional clinical variables (i.e., medical specifications). One of the key tasks is to predict future diagnosis, i.e., diagnostic prediction, from past EHR data of a patient.

The time of visit for each patient and the medical code for each visit may be of different importance in predicting a diagnosis. Therefore, the most important and challenging problem in diagnostic prediction is 1. How to correctly model these temporal and higher order two EHR data to significantly improve prediction performance; 2. how to reasonably explain the importance of the visit and medical discipline in predicting outcomes.

Diagnosis and prediction are challenging and significant works, and accurate prediction of prediction results is a difficult and key problem of medical prediction models. Many existing diagnostic predictive efforts employ deep learning techniques, such as Recurrent Neural Networks (RNNs), to model EHR data in time and high dimensions. However, RNN-based approaches may not be able to fully remember all previous access information, resulting in erroneous predictions.

Disclosure of Invention

The present invention is directed to a diagnostic prediction method for bidirectional recurrent neural networks based on attention, so as to solve the aforementioned problems in the prior art.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

a diagnostic prediction method of a bidirectional recurrent neural network based on attention comprises the following steps:

s1, constructing a medical code x in the electronic health medical record of the patient_t As models of diagnosis, intervention codes, admission type and time lapse sequences; according to the visit information fromtime 1 to t, the medical code x of the ith visit is coded_i ∈{0,1}^|C| Embedding into a vector representation v_i The method comprises the following steps:

v_i ＝ReLU(W_v x_i +b_c )

where | C | is the number of unique medical codes, m is the embedding dimension size, W_v ∈R^m×|C| Is the weight of the medical code, b_c ∈R^m Is a bias vector. ReLU is a rectifying linear unit defined as ReLU (v) = max (v, 0), max () is applied to a vector in element order;

s2, constructing different bidirectional recurrent neural networks according to different attention mechanism states:

will vector v_i Input Bidirectional Recurrent Neural Network (BRNN), and output hidden state h_i As a representation of the ith access, the set of hidden states is denoted as

Wherein,

in order to be in a forward-hidden state,

the state is a backward hidden state;

s3, according to relative importance alpha_t And

computing a context state vector c_t The following:

alpha is obtained by the following softmax function_t ：

α_t ＝Softmax([α_t1 ，α_t2 ，...，α_t(t-1) ]).

Wherein the relative importance α is calculated by an attention mechanism_t Alpha in (A)_ti ；

S4, according to the context state vector c_t And hidden state h of the tth access_t A simple connection layer is used to combine the information from the two vectors to generate an attention hiding state vector, as follows:

wherein, W_c ∈R^r×4p Is a weight matrix;

s5, hiding attention into the state vector

Generating t +1 th access information by softmax layer feeding, defined as:

wherein,

is the parameter to be learned and is,

is a unique number of categories.

Preferably, in S2, the calculation of the relative importance α by the attention mechanism is described_t Alpha in (A)_ti The method specifically comprises the following steps:

location based attention function, according to the current hidden state h_i Weights are calculated separately, as follows:

wherein, W_α ∈R^2p ，b_α The epsilon R is a parameter to be learned;

or, attention is generally paid to functions, using a matrix W_α ∈R^2p×2p To capture h_t And h_i The weights are calculated as follows:

or, using a multi-layered perceptron MLP, the hidden state h of the tth access, based on the connected function_t And hidden state h of ith access_i Then by multiplying by a weight matrix W_α ∈R^q×4q Obtaining potential vectors, q being the potential dimensions, selecting tanh as the activation function, noting that the weight vectors are generated as follows:

wherein upsilon is_α ∈R^q Are the parameters to be learned.

Preferably, after S5, a step of interpreting the prediction is further included, specifically:

using non-negative matrices

Representing medical codes, then arranging each dimension of the attention hiding state vector in a reverse order according to values, and finally selecting the top k codes with the maximum values to obtain clinical explanation of each dimension as follows:

wherein,

to represent

Column i or dimension.

The beneficial effects of the invention are: the invention provides a diagnostic prediction method of an attention-based bidirectional recurrent neural network, which is characterized in that a high-dimensional medical code (namely a clinical variable) is embedded into a low-code layer space, and then a code representation is input into the attention-based bidirectional recurrent neural network to generate a hidden state representation. The hidden representation is entered via the softmax layer to predict future access to the medical code. Experimental data show that by adopting the method provided by the embodiment, when the future access information is predicted, different weights can be distributed to the previous access by the attention mechanism, so that the diagnosis and prediction task can be effectively completed, and the prediction result can be reasonably explained.

Drawings

FIG. 1 is a schematic flow chart of a diagnostic prediction method for an attention-based bidirectional recurrent neural network provided by the present invention;

FIG. 2 is a graph showing the results of attention mechanism analysis of five patients in the example embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.

As shown in fig. 1, the present invention provides a diagnostic prediction method for attention-based bidirectional recurrent neural network, comprising the following steps:

step 1) constructing a medical code x in an electronic health medical record of a patient_t As a diagnostic,A model of the sequence of intervention codes, admission types and time lapses;

step 2) constructing different bidirectional recurrent neural networks according to different attention mechanism states;

step 3) carrying out diagnosis prediction;

step 4) interpreting the prediction.

The detailed explanation about the above steps is as follows:

electronic Health Records (EHRs) consist of longitudinal patient Health data including a sequence of visits over time, with each visit containing a number of medical codes, including diagnosis, medication and program codes. All unique medical codes from the EHR data are noted as c₁ ,c₂ ,…,c_|C| E C, where | C | is the number of unique medical codes. Assuming there are N patients, the Nth patient has T in the EHR data⁽ⁿ⁾ The secondary access record. The patient can be accessed by a series of visits

To indicate. Each access V_i Comprising a subset of medical codes

Using binary vectors x_t ∈{0,1}^|C| Is shown, wherein if V_t Containing a code c_i Then the ith element is 1. Each access V_t With corresponding coarse-grained class representations

Wherein

Is the only number of categories. Each diagnostic code may map to a node of the international disease classification (ICD-9) and each process code may map to a node in the current process terminology. Due to the input of x_t Too sparse and of high dimension, it is natural to learn its low dimension and meaningful embedding.

According to the timeAccess information from 1 to t, the medical code x of the ith access can be encoded_i ∈{0,1}^|C| Embedding into vector representation v_i The method comprises the following steps:

v_i ＝ReLU(W_v x_i +b_c )

where m is the embedding dimension size, W_v ∈R^m×|C| Is the weight of the medical code, b_c ∈R^m Is a bias vector. ReLU is a rectifying linear unit defined as ReLU (v) = max (v, 0), where max () is applied to a vector in element order.

Will vector v_i Input Bidirectional Recurrent Neural Network (BRNN), and output hidden state h_i As a representation of the ith access, the set of hidden states is denoted as

According to relative importance α_t And

compute context state c_t The following are:

an attention weight vector alpha is obtained by the following softmax function_t ：

A BRNN contains a forward and backward Recurrent Neural Network (RNN). Forward direction of rotation

From x₁ To x_T Reading an input access sequence and calculating a sequence of forward hidden states

(

And p is the dimension of the hidden state). Reverse direction

Reading the access sequence in reverse order, i.e. from x_T To x₁ Generating a series of hidden states backwards

Forward hidden state over connection

And backward hidden state

A final layer vector representation can be obtained

There are three attention mechanisms that can be used to calculate the relative importance α_t Alpha in (1)_ti And capturing related information:

the location-based attention function is based on the current hidden state h_i Weights are calculated separately, as follows:

wherein W_α ∈R^2p And b_α e.R is the parameter to learn.

It is generally noted that the function is to use a matrix W_α ∈R^2p×2p To capture h_t And h_i The relation between the weight:

another calculation of alpha_ti The method of (4) is based on a connected function, using a multi-layer perceptron (MLP). First connect the hidden state h of the tth access_t And hidden state h of ith access_i Then by multiplying by a weight matrix W_α ∈R^q×4q Potential vectors can be obtained, q being a potential dimension. Tanh is chosen as the activation function. Note that the weight vector is generated as follows:

wherein upsilon is_α ∈R^q Are parameters that need to be learned.

According to the given context vector c_t And a current hidden state h_t A simple connected layer is used to combine the information from the two vectors to generate an attention hiding state, as follows:

wherein W_c ∈R^r×4p Is a weight matrix. Attention vector

Generating t +1 th access information by softmax layer feeding, defined as:

wherein

Is a parameter to be learned

Using real access information y_t And predictive access

The cross entropy of (c) to calculate the loss for all patients as follows

In medical careThe interpretability of the learned medical code and access representation is very important, it is necessary to understand the clinical meaning of each dimension of the medical code representation and to analyze which accesses are crucial for prediction. Since the proposed model is based on attention mechanisms, the importance of each visit to the prediction is easily discovered through analysis of the attention scores. Prediction of the t-th time, if the attention score is alpha_ti Very large, then the probability prediction for the (i + 1) th access to relevant information is currently high. First using a non-negative matrix

To represent a medical code. Each dimension of the attention hiding state vector is then sorted in reverse order of value. Finally, the first k codes with the largest value are selected as follows:

wherein

To represent

Column i or dimension. By analyzing the selected medical code, a clinical interpretation for each dimension can be obtained.

DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION

In order to illustrate the technical effect of the invention, the invention is verified by adopting a specific application example.

The experiment used two data sets: medical subsidy claims and diabetes claims. The medical assistance data set includes 2011 medical assistance applications. It contains data relating to 147,810 patients and 1,055,011 visits. The patients' visits were grouped by week, excluding patients with visits less than 2. The diabetes data set included medical assistance applications from 2012 and 2013, corresponding to patients diagnosed with diabetes (i.e., medical assistance members who had ICD-9 diagnostic code 250.Xx in the application). It contains the relevant data of 22,820 patients, and the number of times of visit of these patients is 466,732. The patients' visits were summarized in weeks, excluding patients with visits less than 5 times.

For each data set, the data set was randomly divided into training, validation and test sets in a ratio of 0.75. The validation data set was used to determine the optimal values for the parameters, perform 100 iterations, and report the best performance for each method.

Experiment one:

the statistical data set is shown in table 1:

TABLE 1

Experiment two:

the following baseline model was performed: (1) Med2Vec (model 1); (2) RETAIN (model 2); (3) RNN (model 3).

The following RNN-based prediction model is performed: (1) Calculating RNN of relative importance using location-based attention function_l (model 4); (2) Calculating RNN of relative importance using general attention function_g (model 5); (3) Calculating RNN of relative importance with connection-based attention function_c (model 6).

The following Dipole model was performed: (1) Dipole without any attention mechanism_- (model 7); (2) Computing Dipole of relative importance with location-based attention function_l (model 8); (2) Computing Dipole of relative importance with general attention function_g (model 9); (3) Computing Dipole of relative importance with a connection-based attention function_c (model 10)

Results and analysis of the experiments

Table 2 shows the accuracy of all methods on the diabetes data set

TABLE 2

It can be observed in table 2 that since most medical codes are for diabetes, med2Vec (model 2) can correctly learn the vector ratio representation on the diabetes data set. Thus, med2Vec achieves the best results in the three baselines, for the medical data set, RETAIN (model 3) accuracy is better than Med2Vec. The reason is that there are many diseases in the medical assistance data set and there are many more types of medical codes than in the diabetes data set. In this case, the attention mechanism may help RETAIN learn reasonable parameters to make the correct prediction.

The accuracy of RNN (model 1) is lowest in all methods of both datasets. This is because the prediction of RNNs depends mainly on recent access information. It cannot remember all past information. However, retain and proposed RNN variant RNN_l (model 4), RNN_g (model 5) and RNN_c (model 6) it is possible to take into full account all previous visit information, assign different attention scores to past visits, and achieve better performance when compared to RNNs.

Since most of the visits in the diabetes data set are related to diabetes, it is easy to predict the medical code of the next visit based on the past visit information. Retain uses a reverse temporal attention mechanism for prediction, which degrades prediction performance compared to methods using a temporal sequential attention mechanism. The performance of all three RNN variants was superior to that of RETAIN. However, RETAIN is more accurate than RNN variants because the data in the medical assistance data set is about different diseases. The use of a reverse time attention mechanism can help learn the correct access relationships.

RNN and proposed Dipole_- (model 7) none used any attention mechanism, but on the diabetes and medical assistance data sets, dipole_- The accuracy of (2) is higher than that of RNN. The results show that modeling access information from both directions can improve prediction performance. Therefore, it is reasonable to use a bi-directional recurrent neural network for diagnostic prediction.

Proposed isDipole_c (model 10) and Dipole_l (model 8) best performs on the diabetes and the medical assistance data sets, respectively, which shows that modeling the visits from two directions and assigning different weights to each visit can improve the accuracy of the medical diagnosis prediction task. On the diabetes data set, dipole_l And Dipole_c Superior to all baseline and proposed RNN variants. On the medical assistance data set, dipole_l 、Dipole_g And Dipolec were superior to baseline and RNN variants

FIG. 2 shows a case study predictive healthcare code at sixth visit (y)₅ ) Based on previous diabetes dataset access. According to the hidden state h₁ ,h₂ ,h₃ The connection-based attention weights for the second through fifth visits are calculated. In fig. 2, the x-axis is the patient and the y-axis is the attention weight for each visit. In this case study, 5 patients were selected for the trial. It can be observed that the attention scores learned by the attention mechanism are different for different patients. For the second patient in fig. 2, all diagnostic codes are listed in table 3:

TABLE 3

The weight α = [0.2386,0.0824,0.2386,0.0824] for four visits bypatient 2 was first obtained from fig. 2. From an analysis of this attention vector, it can be concluded that the medical coding at the second, fourth and fifth visit has a significant impact on the final prediction. As can be seen from table 3, the patient developed essential hypertension at the second and fourth visits, with diabetes being diagnosed at the fifth visit. Therefore, the probability of medical coding for diabetes and essential hypertension related diseases at visit 6 is high.

In summary, dipole can remedy the challenges of modeling EHR data and interpreting the predicted results. With a bi-directional recurrent neural network, dipole can remember the learned hidden knowledge in past and future accesses. Three attention mechanisms can reasonably explain the prediction result. Experimental results on two large real EHR data sets demonstrate the effectiveness of this Dipole in diagnostic predictive tasks. Analysis shows that in predicting future access information, the attention mechanism may assign different weights to previous accesses.

By adopting the technical scheme disclosed by the invention, the following beneficial effects are obtained: the invention provides a diagnostic prediction method of an attention-based bidirectional recurrent neural network, which is characterized in that a high-dimensional medical code (namely a clinical variable) is embedded into a low-code layer space, and then a code representation is input into the attention-based bidirectional recurrent neural network to generate a hidden state representation. The hidden representation is entered via the softmax layer to predict future access to the medical code. Experimental data show that by adopting the method provided by the embodiment, when future access information is predicted, different weights can be distributed to previous accesses by an attention mechanism, so that not only can a diagnosis and prediction task be effectively completed, but also a prediction result can be reasonably explained.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and improvements can be made without departing from the principle of the present invention, and such modifications and improvements should also be considered within the scope of the present invention.

Claims (3)

1. A diagnostic prediction method of a bidirectional recurrent neural network based on attention is characterized by comprising the following steps:

s1, constructing a medical code x in the electronic health medical record of the patient_t Including diagnosis, intervention codes, admission type and time lapse sequences as input parameters for the model; according to the visit information from time 1 to time t, the medical code x of the ith visit is coded_i ∈{0,1}^|C| Embedding into vector representation v_i The method comprises the following steps:

v_i ＝ReLU(W_v x_i +b_c )

where | C | is the number of unique medical codes, m is the embedding dimension size, W_v ∈R^m×|C| Is a medical substituteWeight of code, b_c ∈R^m Is an offset vector, reLU is a rectified linear unit defined as ReLU (v) = max (v, 0), max () is applied to vector v in element order;

s2, constructing different bidirectional recurrent neural networks according to different attention mechanism states:

will vector v_i Inputting bidirectional recurrent neural network, outputting hidden state h_i As a representation of the ith access, the set of hidden states is denoted as

Wherein,

in order to be in a forward-hidden state,

the state is a backward hidden state;

s3, according to relative importance α_t And

computing a context state vector c_t The following:

alpha is obtained by the following softmax function_t ：

α_t ＝Softmax([α_t1 ,α_t2 ,…,α_t(t-1) ]).

Wherein the relative importance α is calculated by an attention mechanism_t Alpha in (1)_ti ；

S4, according to the context state vector c_t And hidden state h of the tth access_t A simple connection layer is used to combine the information from the two vectors to generate an attention hiding state vector, as follows:

wherein, W_c ∈R^r×4p Is a weight matrix;

s5, hiding attention into a state vector

The t +1 th visit information is generated by softmax layer feed, defined as:

wherein,

is the parameter to be learned and is,

is the only number of categories.

2. The method of predicting diagnosis of attention-based bi-directional recurrent neural network of claim 1, wherein said calculating relative importance α by attention mechanism in S2_t Alpha in (A)_ti The method specifically comprises the following steps:

location based attention function, according to the current hidden state h_i Weights are calculated separately, as follows:

wherein, W_α ∈R^2p ，b_α E.g. R, is the parameter to be learned;

or, attention is generally paid to functions, using a matrix W_α ∈R^2p×2p To capture h_t And h_i The relationship between them, the weights are calculated as follows:

or, based on the function of the connection, using a multi-layered perceptron MLP, first connecting the hidden states h of the t-th access_t And hidden state h of ith access_i Then by multiplying by a weight matrix W_α ∈R^q×4q Obtaining potential vectors, q being the potential dimensions, selecting tanh as the activation function, noting that the weight vectors are generated as follows:

wherein, upsilon_α ∈R^q Are the parameters to be learned.

3. The method for predicting diagnosis of attention-based bi-directional recurrent neural network of claim 1, further comprising the step of interpreting the prediction after S5, specifically:

using non-negative matrices

Representing medical codes, then, carrying out reverse value arrangement on each dimension of the attention hiding state vector, and finally, selecting the first k codes with the maximum value to obtain clinical explanation of each dimension as follows:

wherein,

represent

Column i or dimension.

Images