CN114822737B - Electronic medical record-based Litsea artificial liver preoperative diagnosis and treatment system and use method - Google Patents

Abstract

The invention discloses a Litsea artificial liver preoperative diagnosis and treatment system based on an electronic medical record, which comprises a medical record management subsystem and an intelligent analysis subsystem, wherein the medical record management subsystem comprises an electronic medical record uploading unit, a database storage unit and a search engine unit; the attention-based multi-feature fusion prediction unit comprises a text data feature extraction network, a numerical data extraction network and a feature fusion classification network, the invention also discloses a method for using the Litsea artificial liver preoperative diagnosis and treatment system based on the electronic medical record. According to the invention, the pure text data and the check value data in the electronic medical record information can be analyzed by adopting a deep learning network and a classification result is given.

Description

A Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records and its use method

Technical Field

The present invention relates to the field of computer-aided diagnosis and treatment and image recognition, and in particular to a Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records and a use method thereof.

Background Art

Liver diseases have always posed a serious threat to human life and health. The World Health Organization reported that liver cancer has become the second leading cause of death in humans. The Lee's artificial liver support system is an effective treatment for severe liver diseases. The Lee's artificial liver mainly uses the regenerative ability of liver cells and external equipment, including mechanical, physical, chemical or biological devices, to replenish the body's essential components while removing harmful substances produced by liver failure, stabilize the body's internal environment, and temporarily replace some liver functions.

With the rapid development of Internet technology and the popularization of informatization, the hospital's information management system is becoming more and more perfect. On this basis, using the existing data resources in the hospital information system to conduct preoperative analysis of artificial liver can not only improve the efficiency of diagnosis and treatment, but also provide auxiliary diagnosis and treatment opinions, which is of great significance to the effectiveness of artificial liver diagnosis and treatment. However, after admission to the hospital, patients with liver failure need to undergo a large number of routine examinations and comprehensive medical treatment. The amount of existing data on patients with liver failure in the hospital information system is very large, including admission records, medical records, index test results, CT images, and other medical data. Therefore, it takes a lot of time and energy for doctors to use traditional manpower methods for preoperative analysis, and the accuracy is limited by the doctor's experience and energy. Therefore, a method is needed to analyze the existing data resources in the hospital information system through deep learning algorithms to reduce the burden on doctors.

Summary of the invention

The technical problem to be solved by the present invention is to provide a Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records and a method of use, which is used to analyze the pure text category data and test numerical data in the electronic medical record information using a deep learning network and give classification results.

In order to solve the above technical problems, the present invention provides a preoperative diagnosis and treatment system of Lee's artificial liver based on electronic medical records and a method of use, including: a medical record management subsystem and an intelligent analysis subsystem, the medical record management subsystem includes an electronic medical record upload unit, a database storage unit and a search engine unit, the intelligent analysis subsystem includes an attention-based multi-feature fusion prediction unit; mutual signal connections between the electronic medical record upload unit and the database storage unit, between the database storage units and the search engine unit, and between the search engine unit and the intelligent analysis subsystem;

The attention-based multi-feature fusion prediction unit includes a text data feature extraction network, a numerical data extraction network and a feature fusion classification network, and the outputs of the text data feature extraction network and the numerical data extraction network are simultaneously used as inputs of the feature fusion classification network.

As an improvement of the Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records of the present invention:

The medical record management subsystem includes a medical record input unit for inputting electronic medical record information and a medical record upload unit for uploading electronic medical record information; the database storage unit includes an electronic medical record database storing electronic medical record information.

As a further improvement of the electronic medical record-based Lee's artificial liver preoperative diagnosis and treatment system of the present invention:

The structure of the text data feature extraction network is as follows: the first layer is a word embedding layer with an output size of 32*256*200; the second layer is a BiLSTM layer with an output size of 32*256*256; the third layer is an attention layer with an output size of 32*256*256; the fourth layer is a splicing layer with an output size of 32*256*456; the fifth layer is a fully connected layer with an output size of 32*256*64; the sixth layer is a pooling layer with an output size of 32*64; the first to sixth layers are connected in sequence, and the output of the word embedding layer of the first layer is connected to the input of the splicing layer of the fourth layer;

The structure of the numerical feature extraction network is as follows: the first layer is an input layer, and the output size is 32*200; the second layer is a fully connected layer, and the output size is 32*128; the third layer is a fully connected layer, and the output size is 32*64, and the first layer to the third layer are connected in sequence;

The structure of the feature fusion classification network is as follows: the first layer is a concatenation layer with an output size of 32*128; the second layer is a fully connected layer with an output size of 32*64; the third layer is a fully connected layer with an output size of 32*2. The evaluated classification result is a binary classification prediction result, and the first to third layers are connected in sequence.

As a further improvement of the electronic medical record-based Lee's artificial liver preoperative diagnosis and treatment system of the present invention:

The electronic medical record information includes patient basic information, attending physician information, pure text data and test numerical data; patient basic information includes patient name, gender and medical record number; attending physician information includes attending physician name, department and attending physician number; pure text data includes personal history, past history and chief complaint; test numerical data includes various test index data.

As a further improvement of the electronic medical record-based Lee's artificial liver preoperative diagnosis and treatment system of the present invention:

The training and testing process of the attention-based multi-feature fusion prediction unit is as follows: construct a training set, a validation set and a test set, the batch_size of the multi-feature fusion prediction unit is set to 32, the learning rate is 0.001, the number of iterations is 30, the Dropout value is set to 0.1, the optimization strategy is the Adam gradient descent method, and the activation function is Relu: the training set and the validation set are input into the attention-based multi-feature fusion prediction unit for training to obtain a trained attention-based multi-feature fusion prediction unit; then the test set is input into the trained attention-based multi-feature fusion prediction unit for testing, and the output binary classification prediction results are compared with the corresponding annotations in the test set to meet the requirements for online use.

As a further improvement of the electronic medical record-based Lee's artificial liver preoperative diagnosis and treatment system of the present invention:

The process of constructing the training set, validation set and test set is as follows: electronic medical record information of patients with liver failure is collected, each electronic medical record is annotated by a professional doctor, and the pure text category data and test value data in each electronic medical record information are preprocessed respectively, and then the preprocessed pure text category data and test value data with annotations are divided into the training set, the test set and the validation set in a ratio of 7:2:1.

The preprocessing of the test numerical data is to normalize the test numerical data according to the following formula (1):

Where x _norm , x _min , and x _max are the normalized test value, the minimum value of the test value data, and the maximum value of the test value data respectively;

The preprocessing of the pure text category data is to first use a word segmentation tool to perform word segmentation processing on the pure text category data to obtain a group of keywords, and then convert the keywords into a 200-dimensional vector.

The present invention also provides a method for using the Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records, comprising the following steps:

Step S01: Electronic medical record information management

Step S101: Upload electronic medical record information

Enter and upload electronic medical record information to the database storage unit through the electronic medical record upload unit, including pure text data and test numerical data; the data of the electronic medical record information is uniformly stored in the database storage unit;

Step S102: Electronic medical record search

The user inputs keywords, including "patient name", "medical record number", "doctor name" or "doctor number" through the search engine unit; the search engine unit searches and matches the electronic medical record information in the database storage unit according to the input keywords and presents the search results on the host computer;

Step S103: Update electronic medical record information

First, the electronic medical record information to be updated is searched and obtained in step S102, and then the updated electronic medical record information is input and uploaded to the database storage unit in step S101;

Step 02: Prediction Evaluation

Step 0201, searching and acquiring electronic medical record information through step S102 and transmitting the electronic medical record information to the intelligent analysis subsystem;

Step 0202, the intelligent analysis subsystem pre-processes the plain text data and the test value data in the electronic medical record information respectively;

Step 0203: Input the 200-dimensional vector obtained by preprocessing the pure text data and the normalized test numerical data obtained by preprocessing the test numerical data into the multi-feature fusion prediction unit, obtain the binary classification prediction result after prediction evaluation, and display the prediction result on the host computer.

The beneficial effects of the present invention are mainly reflected in:

1. The present invention extracts features from pure text category data and test value data in electronic medical records respectively. The feature fusion classification network can fuse the features of the extracted pure text category data and test value data and give classification results, thereby improving the doctor's work efficiency and reducing time and labor costs;

2. The present invention uses a deep learning network to analyze and evaluate electronic medical records, which can improve the accuracy of evaluation results and reduce the number of misjudgments.

BRIEF DESCRIPTION OF THE DRAWINGS

The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

FIG1 is an overall architecture diagram of a Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records of the present invention;

FIG2 is a schematic diagram of the network structure of the attention-based multi-feature fusion prediction unit of the present invention.

DETAILED DESCRIPTION

The present invention is further described below in conjunction with specific embodiments, but the protection scope of the present invention is not limited thereto:

Embodiment 1, a preoperative diagnosis and treatment system of Lee's artificial liver based on electronic medical records and a method of use, as shown in Figure 1-2, through the Lee's artificial liver preoperative diagnosis and treatment system based on electronic medical records (including a medical record management subsystem and an intelligent analysis subsystem), doctors can operate the system on a host computer to manage, update, and upload electronic medical records. At the same time, it can indirectly call the deep learning algorithm model to output classification results, providing a reference for doctors to judge whether patients are suitable for Lee's artificial liver instruments.

The medical record management subsystem includes an electronic medical record upload unit, a database storage unit and a search engine unit; the electronic medical record upload unit includes a medical record input unit and a medical record upload unit, which are used to record electronic medical record information and upload and store it in the database storage unit. The electronic medical record information includes basic patient information (including patient name, gender, medical record number, etc.), attending physician information (including attending physician name, department, attending physician number, etc.), pure text data (including personal history, past history, chief complaint, etc.), and test numerical data (including 200 test index data such as blood pressure and blood sugar); the database storage unit includes an electronic medical record database for storing all electronic medical record information; the search engine unit can search and match the patient's relevant information in each electronic medical record information of the database storage unit through keywords such as "patient name", "medical record number", "doctor name" or "doctor number";

The intelligent analysis subsystem includes an attention-based multi-feature fusion prediction unit. After the electronic medical record is processed by the intelligent analysis subsystem, the classification results of the attention-based multi-feature fusion prediction unit can be obtained, which can be used by doctors to determine whether the patient is suitable for the Lee's artificial liver instrument. The attention-based multi-feature fusion prediction unit innovatively integrates the pure text category data and test numerical data in the electronic medical record data, making the evaluation effect more accurate.

The attention-based multi-feature fusion prediction unit includes a text data feature extraction network, a numerical data extraction network, and a feature fusion classification network, as shown in Figure 2. The text data feature extraction network is based on the TextRCNN network with an added attention layer and a splicing layer. The specific structure is as follows: the first layer is a word embedding layer with an output size of 32*256*200; the second layer is a BiLSTM layer with an output size of 32*256*256; the third layer is an attention layer with an output size of 32*256*256. The attention layer added to the text data feature extraction network is mainly used to optimize the problem of large text data information capacity and unremarkable feature representation in the medical record text due to the large number of non-related words, so as to enhance the prediction of the disease. The influence of key vocabulary features is measured to improve the prediction accuracy and generalization ability of the model; the fourth layer is a concatenation layer with an output size of 32*256*456; the fifth layer is a fully connected layer with an output size of 32*256*64; the sixth layer is a pooling layer with an output size of 32*64; the first to sixth layers are connected in sequence, and the output of the word embedding layer of the first layer is connected to the input of the concatenation layer of the fourth layer, which is used to concatenate the output of the word embedding layer of the first layer and the output of the attention layer of the third layer, and the features of the attention layer and the word embedding layer can be retained at the same time. The text data feature extraction network is used to extract the features of pure text category data in electronic medical record information.

The numerical feature extraction network is used to extract the features of the test numerical data in the electronic medical record information; the specific structure of the numerical feature extraction network is: the first layer is the input layer, and the output size is 32*200; the second layer is the fully connected layer, and the output size is 32*128; the third layer is the fully connected layer, and the output size is 32*64, and the first to third layers are connected in sequence. The test numerical data in the electronic medical record information includes 200 types of test data such as blood pressure and blood sugar. Because the features of the test numerical data are of different dimensions and are not comparable, if they are directly sent to the network, the features of the data with higher values will be more significant, while the features of the data with lower values will be weaker. Therefore, the test numerical data needs to be normalized first:

Among them, x _norm , x _min , and x _max are the normalized test value, the minimum value of the test value data, and the maximum value of the test value data, respectively.

The feature fusion classification network is mainly used to fuse the features of pure text data with the features of test numerical data and perform classification prediction; the specific structure of the feature fusion classification network is that the first layer is a concatenation layer with an output size of 32*128; the second layer is a fully connected layer with an output size of 32*64; the third layer is a fully connected layer with an output size of 32*2, and the first to third layers are connected in sequence. The classification results evaluated are binary classification prediction results.

The attention-based multi-feature fusion prediction unit needs to be trained. The host computer used for training is configured with the Centos6.7 operating system, and uses an E5-2667 v4@3.20GHz, 8-core CPU, and 2 Tesla P4 GPUs. The data set used for training is the electronic medical record information set of liver failure provided by a tertiary hospital. There are a total of 1,024 electronic medical records, and each electronic medical record is classified and labeled by a professional doctor according to the patient's actual condition. Among them, 498 records are classified as 1, indicating that they are suitable for the Lee's artificial liver instrument, and 526 records are classified as 2, indicating that they are not suitable for the Lee's artificial liver instrument.

Then, the pure text data and test value data in the 1024 electronic medical records were preprocessed respectively. The preprocessing of the pure text data included:

1) Use Jieba word segmentation tool to segment the pure text data to obtain a set of keywords;

2) Use the Glove tool to convert the obtained keywords into a 200-dimensional vector.

The preprocessing of the test numerical data is to perform normalization operation on the test numerical data according to formula (1).

The 200-dimensional vector obtained by preprocessing, the normalized test numerical data and the corresponding classification labels are divided into training set, test set and validation set in a ratio of 7:2:1. The batch_size of the network is set to 32, the learning rate is 0.001, the number of iterations is 30, the Dropout value is set to 0.1, the optimization strategy is the Adam gradient descent method, the activation function is Relu, the formula is as follows, the advantage is that it can make the network converge quickly and improve the problem of gradient disappearance of the neural network.

The training set and validation set are input into the attention-based multi-feature fusion prediction unit for network training. During the training process, the training set data is used to optimize the network parameters, and the validation set is used for the preliminary evaluation of the network. Finally, a trained attention-based multi-feature fusion prediction unit is obtained; then the test set is input into the trained attention-based multi-feature fusion prediction unit for testing, and the binary classification prediction results output by the network are compared with the corresponding annotations in the test set. The ratio of the number of correctly classified samples to the total number of samples is 91.9%, which meets the requirements for online use.

The mutual signal connections between the electronic medical record upload unit and the database storage unit, between the database storage units and the search engine unit, and between the search engine unit and the intelligent analysis subsystem of the electronic medical record-based Lee's artificial liver preoperative diagnosis and treatment system of the present invention have a variety of flexible setting methods. For example, the medical record management subsystem and the intelligent analysis subsystem can be simultaneously set in the user's computer or the cloud server, or the database storage unit is set in the cloud server, and the electronic medical record upload unit, the search engine unit and the intelligent analysis subsystem are set in the user's computer.

The method of use of the present invention is as follows:

Step S01: Electronic medical record information management

Step S101: Upload electronic medical record information

Enter and upload electronic medical record information to the database storage unit through the electronic medical record upload unit, including records of pure text data such as patient information, personal history, past history, and records of test numerical data such as blood pressure and blood sugar; the data of the electronic medical record information is uniformly stored in the database storage unit;

Step S102: Electronic medical record search

The user inputs keywords including "patient name", "medical record number", "doctor name" or "doctor number" through the search engine unit;

The search engine unit searches and matches the electronic medical record information in the database storage unit according to the input keywords and presents the search results on the host computer;

Step S103: Update electronic medical record information

First, the electronic medical record information record to be updated is searched and obtained according to step S102, and then the updated electronic medical record information is entered and uploaded to the database storage unit according to step S101;

Step 02: Prediction Evaluation

Step 0201, searching and acquiring electronic medical record information through step S102 and transmitting the electronic medical record information to the intelligent analysis subsystem;

Step 0202, the intelligent analysis subsystem pre-processes the plain text data in the electronic medical record information: the Jieba word segmentation tool is used to segment the plain text data to obtain a number of keywords; then the Glove tool is used to convert the obtained keywords into 200-dimensional vectors, and at the same time, the intelligent analysis subsystem pre-processes the test numerical data in the electronic medical record information by normalizing the data according to formula (1);

Step 0203: Input the 200-dimensional vector after preprocessing of the pure text data and the normalized test numerical data into the attention-based multi-feature fusion prediction unit, obtain the binary classification prediction result after prediction evaluation, and display the prediction result on the host computer.

experiment:

The experimental data set is consistent with the training set, validation set and test set used for training in Example 1. Four prior art networks and the attention-based multi-feature fusion prediction unit of the present invention are trained on the same training set and validation set, and the trained networks are compared and tested using the same test set to obtain the experimental results of each network, and the accuracy, precision, recall and F1 value indicators are used to statistically analyze the experimental results of each comparison network.

During the training process, the training set data is used to optimize the network parameters, and the validation set is used for the preliminary evaluation of the network. The four comparison networks are: bag-of-words model + support vector machine (BOW+SVM), word frequency-inverse text frequency + support vector machine (TF-IDF+SVM), Glove+TextRNN and word2vec+TextRCNN. The bag-of-words model + support vector machine (BOW+SVM) and word frequency-inverse text frequency + support vector machine (TF-IDF+SVM) are built according to reference 1 (Hu Jing, Liu Wei, Ma Kai. Hypertension medical record text classification based on machine learning [J]. Science Technology and Engineering, 2019, 19(33): 296-301.); Glove+TextRNN is built according to reference 2 (Graves A, Schmidhuber J. Framewise Phoneme Classification with Bidirectional LSTM and Other Neural Network Architectures [J]. Neural Networks, 2005, 18(5–6): 602-610.); word2vec+TextRCNN is built according to reference 3 (Lai S, Xu L, Liu K, et al. Recurrent Convolutional Neural Networks for Text Classification[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2015,29(1).).

The evaluation indicators of this experimental result are accuracy, precision, recall and F1 value. The specific expressions are as follows:

Among them, TP (True Positive) means that the positive sample is classified as positive, TN (True Negative) means that the negative sample is classified as positive, FP (False Positive) means that the negative sample is classified as positive, and FN (False Negative) means that the negative sample is classified as negative.

The parameters for training are: batch_size is set to 32, learning rate is 0.001, number of iterations is 30, Dropout value is set to 0.1, optimization strategy is Adam gradient descent method, activation function is Relu. The final experimental results are shown in Table 1:

Table 1. Comparison of experimental results

From the results in Table 1, it can be seen that the attention-based multi-feature fusion prediction unit in the present invention has the highest evaluation indicators, which proves that after the attention-based multi-feature fusion prediction unit of the present invention fuses plain text data and test numerical data, it can better mine the information in the electronic medical record, thereby improving indicators such as accuracy.

Finally, it should be noted that the above examples are only some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or associated with the content disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims (6)

1. The Litsea artificial liver preoperative diagnosis and treatment system based on the electronic medical record is characterized by comprising a medical record management subsystem and an intelligent analysis subsystem, wherein the medical record management subsystem comprises an electronic medical record uploading unit, a database storage unit and a search engine unit, and the intelligent analysis subsystem comprises a multi-feature fusion prediction unit based on attention; mutual signal connection between the electronic medical record uploading unit and the database storage unit, between the database storage units and between the search engine units, and between the search engine units and the intelligent analysis subsystem;

The attention-based multi-feature fusion prediction unit comprises a text data feature extraction network, a numerical data extraction network and a feature fusion classification network, wherein the output of the text data feature extraction network and the numerical data extraction network is simultaneously used as the input of the feature fusion classification network;

The text data feature extraction network has the structure that: the first layer is a word embedding layer, and the output size is 32×256×200; the second layer is BiLSTM layers, and the output size is 32×256×256; the third layer is an attention layer, the output size is 32×256×256, the fourth layer is a splicing layer, and the output size is 32×256×456; layer 5 is a fully connected layer with an output size of 32 x 256 x 64; layer 6 is a pooling layer with an output size of 32 x 64; the first layer is connected with the sixth layer in sequence, and the output of the word embedding layer of the first layer is connected with the input of the splicing layer of the fourth layer;

the numerical feature extraction network has the structure that: the first layer is an input layer, and the output size is 32 x 200; the second layer is a full-connection layer, and the output size is 32 x 128; the third layer is a full-connection layer, the output size is 32 x 64, and the first layer to the third layer are sequentially connected;

The feature fusion classification network has the structure that: the first layer is a splicing layer, and the output size is 32 x 128; the second layer is a full-connection layer, and the output size is 32 x 64; the third layer is a full-connection layer, the output size is 32 x 2, the estimated classification result is a classification prediction result, and the first layer to the third layer are sequentially connected.

2. The electronic medical record-based pre-operative diagnosis and treatment system for Litsea artificial liver of claim 1, wherein:

The medical record management subsystem comprises a medical record input unit for transmitting the sub medical record information and a medical record uploading unit for uploading the electronic medical record information; the database storage unit comprises an electronic medical record database storing electronic medical record information.

3. The electronic medical record-based pre-operative diagnosis and treatment system for Litsea artificial liver of claim 2, wherein:

the electronic medical record information comprises basic information of a patient, information of a main doctor, pure text data and check value data; the patient base information includes patient name, sex and patient number; the information of the main doctor comprises the name, department and number of the main doctor; pure text data includes personal history, past history, and complaints; the test value data includes various kinds of test index data.

4. A system for pre-operative diagnosis and treatment of liver in a li-shi artificial system based on electronic medical records as claimed in claim 3, wherein:

The training and testing process of the attention-based multi-feature fusion prediction unit comprises the following steps: constructing a training set, a verification set and a test set, setting the batch_size of the multi-feature fusion prediction unit to be 32, the learning rate to be 0.001, the iteration number to be 30, setting the Dropout value to be 0.1, and the optimization strategy to be an Adam gradient descent method, wherein the activation function is Relu: inputting the training set and the verification set into the attention-based multi-feature fusion prediction unit for training to obtain a trained attention-based multi-feature fusion prediction unit; and then inputting the test set into a trained multi-feature fusion prediction unit based on attention for testing, and comparing the output two classification prediction results with corresponding labels in the test set to meet the requirement of online use.

5. The electronic medical record-based pre-operative diagnosis and treatment system for Litsea artificial liver of claim 4, wherein:

The process for constructing the training set, the verification set and the test set comprises the following steps: collecting electronic medical record information of a liver failure patient, marking each electronic medical record by a professional doctor, respectively preprocessing pure text data and check value data in each electronic medical record information, and then using 7:2:1 into training set, test set and validation set

The preprocessing of the test value data is to normalize the test value data according to the following formula (1):

Wherein x _norm、x_min、x_max is the normalized test value, the minimum value of the test value data, and the maximum value of the test value data, respectively;

the preprocessing of the pure text data comprises the steps of firstly using a word segmentation tool to segment the pure text data to obtain a group of keywords, and then converting the keywords into 200-dimensional vectors.

6. A method for using an electronic medical record-based system for pre-operative diagnosis and treatment of liver in a li-shi artificial system according to any one of claims 1-5, wherein:

the method comprises the following steps:

Step S01, electronic medical record information management

Step S101, uploading electronic medical record information

The electronic medical record uploading unit is used for inputting and uploading the electronic medical record information to the database storage unit, wherein the electronic medical record information comprises pure text data and check value data; the data of the electronic medical record information are uniformly stored in a database storage unit;

step S102, searching electronic medical records

The user inputs keywords including a 'patient name', 'patient number', 'doctor name' or 'doctor number' through the search engine unit; the search engine unit searches and matches the electronic medical record information in the database storage unit according to the input keywords and presents search results on the upper computer;

step S103, updating the electronic medical record information

Firstly, searching and obtaining electronic medical record information to be updated according to the step S102, and then inputting and uploading the updated electronic medical record information to a database storage unit through the step S101;

step 02, predictive assessment

Step 0201, searching and acquiring electronic medical record information through step S102 and transmitting the electronic medical record information to the intelligent analysis subsystem;

step 0202, the intelligent analysis subsystem respectively preprocesses the plain text data and the check value data in the electronic medical record information;

Step 0203, inputting the 200-dimensional vector obtained by preprocessing the plain text data and the normalized check value data obtained by preprocessing the check value data into a multi-feature fusion prediction unit based on attention, obtaining a two-class prediction result after prediction evaluation, and displaying the prediction result on an upper computer.