CN109935336B

CN109935336B - Intelligent auxiliary diagnosis system for respiratory diseases of children

Info

Publication number: CN109935336B
Application number: CN201910037320.7A
Authority: CN
Inventors: 倪鑫; 赵顺英; 王端民; 王海
Original assignee: Beijing Sipuke Software Co ltd
Current assignee: Beijing Sipuke Software Co ltd
Priority date: 2019-01-15
Filing date: 2019-01-15
Publication date: 2023-10-31
Anticipated expiration: 2039-01-15
Also published as: CN109935336A

Abstract

The application provides an intelligent auxiliary diagnosis system for respiratory diseases of children, which comprises: s1, acquiring auxiliary examination information and/or doctor inquiry record information of a child to be diagnosed; s2, screening key information according to the auxiliary examination information and/or doctor inquiry record information; s3, processing the key information by adopting a pre-trained typical symptom diagnosis model according to a pre-established knowledge base system of the children diseases to obtain a diagnosis result; the diagnostic result includes: at least one possible disease, and each possible disease corresponds to a characteristic of the child to be diagnosed. The intelligent auxiliary diagnosis system can help non-expert doctors to improve diagnosis level of difficult and complicated diseases and reduce misdiagnosis.

Description

Intelligent auxiliary diagnosis system for respiratory diseases of children

Technical Field

The application relates to an artificial intelligence technology, in particular to an intelligent auxiliary diagnosis system for respiratory diseases of children.

Background

The current situations of personnel shortage and uneven capacity distribution in the existing children medical health service in China are still very serious, basic staff lack abundant clinical experience, and the traditional medical training means are difficult to play a desired role. The direct manifestation is that it is often difficult for the primary physician to take on the proper responsibility and for the patient to get high quality medical service.

For this reason, the industry is concerned with technology based on artificial intelligence as a core, the capability of copying the most excellent pediatric medical resources in China, and the orderly and effective sinking of the high-quality medical resources is realized through the technology.

Currently, the latest achievement of artificial intelligence in the medical field, namely the Baidu medical brain, is deduced by hundred degrees, and the artificial intelligence technology is formally applied to the medical health industry, wherein the Baidu medical brain is a specific application of the Baidu brain in medical scenes. The specific application scenario comprises providing intelligent assistance for online consultation of hundreds of doctors, providing assistance for hospitals and establishing user portraits for patients so as to carry out chronic disease management and the like.

The hundred-degree medical brain cannot realize diagnosis and treatment of childhood diseases. Since children do not express, their individual features are not clearly known. At present, artificial intelligence is still blank in aspects of diagnosis and treatment of respiratory diseases of children, especially difficult and complicated diseases, the existing artificial intelligence image analysis of the lung radiographic image of adults is only limited, and is mostly focused on interpretation of lung cancers, so that researches on difficult and complicated diseases of respiratory departments of children are few, and the application of the radiographic image is strictly limited clinically by children due to the specificity of children.

How to realize intelligent diagnosis of difficult and complicated diseases of children is a problem which needs to be solved currently.

Disclosure of Invention

Aiming at the problems in the prior art, the application provides an intelligent auxiliary diagnosis system for children respiratory diseases.

In a first aspect, the present application provides an intelligent auxiliary diagnostic system for respiratory diseases in children, comprising:

the first information collection module is used for collecting doctor inquiry record information of the child to be diagnosed;

the second information collection module is used for acquiring auxiliary examination information of the child to be diagnosed when auxiliary examination items exist in the doctor inquiry record information;

the processing module is used for screening the key information according to the auxiliary examination information and/or doctor inquiry record information; processing the key information according to a pre-established knowledge base system for the child diseases and by adopting a pre-trained typical symptom diagnosis model to obtain a diagnosis result;

the display module is used for displaying the diagnosis result, and the diagnosis result comprises: at least one possible disease, and each possible disease corresponds to a characteristic of the child to be diagnosed;

the auxiliary inspection information includes: one or more of pathological examination information, ultrasonic auxiliary examination information, X-ray auxiliary examination information, CT and nuclear magnetic auxiliary examination information; and/or the number of the groups of groups,

the doctor inquiry record information includes: auxiliary examination information of non-expert doctors, recorded family history information and allergy history information;

the child disease knowledge base system is a term dictionary expressed by vectorization, one of the term dictionaries is classified into a multidimensional vector, each dimension in the multidimensional vector is a disease feature in the corresponding classification, and weights of the disease features in the classifications are expressed by one vector;

the weights of all words in the key information are represented by a vector, the key information is a multidimensional vector, and each disease feature is a dimensional information.

Optionally, the first information collecting module includes:

the onset recording unit is used for recording onset classification information of the child to be diagnosed;

a symptom-performance recording unit for recording physical state information of the child to be diagnosed;

the symptom examination recording unit is used for recording the preliminary examination information of the current doctor on the child to be diagnosed;

the past medical history inquiry recording unit is used for recording medical record information of the response of the child to be diagnosed;

and the auxiliary examination item confirming unit is used for recording auxiliary examination items needing auxiliary examination.

Optionally, the first information collection module and the display module are simultaneously located on an information display interface of the auxiliary diagnostic system.

Alternatively, the typical symptom diagnostic model is a BP neural network model,

the number of nodes at the input layer of the BP neural network model is consistent with the dimension of the input vector,

the learning step length is 0.01-0.8; the hidden layer node number is the node number determined by a node deleting method and an expanding method according to the network structure complexity and the error requirement;

the output layer of the BP neural network model is one layer, and the number of nodes is consistent with the number of output vectors.

The application has the beneficial effects that:

according to the application, based on the acquisition of the diagnosis standard and priori knowledge of the difficult and complicated diseases of the respiratory department of the children, the artificial intelligence technology is utilized to develop the artificial intelligence auxiliary diagnosis system for the difficult and complicated diseases of the respiratory department of the children, so that doctors are helped to improve the cognition and diagnosis level of the difficult and complicated diseases, and powerful support is provided for reducing misdiagnosis, optimizing treatment scheme and early prevention.

In addition, the system based on the application can realize the beneficial support of the basic medical treatment, so that the child patient can perform preliminary examination and screening on the basic layer, and the problem is found and then transferred to the expert, thereby reasonably distributing the medical resources and ensuring the reasonable use of the medical resources.

The system provided by the application has been tested, the resolution ratio of the system for treating the difficult and complicated respiratory diseases can reach 99%, and the system can effectively assist doctors to check.

Drawings

FIG. 1 is a schematic flow chart of an intelligent auxiliary diagnosis method for respiratory diseases of children according to an embodiment of the application;

fig. 2 is a schematic diagram of an information display interface of an intelligent auxiliary diagnosis system for respiratory diseases of children according to an embodiment of the present application;

FIG. 3 is a flowchart of training BP neural network according to an embodiment of the present application;

FIG. 4 is a schematic diagram illustrating a process of diagnosing a BP neural network after training according to an embodiment of the present application;

fig. 5 is a schematic diagram of an intelligent auxiliary diagnosis method for respiratory diseases of children according to an embodiment of the present application.

Detailed Description

The application will be better explained by the following detailed description of the embodiments with reference to the drawings.

Example 1

As shown in fig. 1, the intelligent auxiliary diagnosis method for the respiratory disease of the child of the present embodiment includes the following steps:

s1, auxiliary examination information and/or doctor inquiry record information of the child to be diagnosed are obtained.

For example, the auxiliary inspection information includes: one or more of ultrasound-assisted inspection information, X-ray-assisted inspection information, and nuclear magnetic-assisted inspection information;

the doctor inquiry record information includes: auxiliary examination information of non-expert doctors, recorded family history information and allergy history information.

S2, screening key information according to the auxiliary examination information and/or doctor inquiry record information.

For example, personal disease history data of a child to be diagnosed stored by an automatic diagnostic system is obtained;

and screening the key information according to the personal disease history data, auxiliary examination information and/or doctor inquiry record information of the child to be diagnosed.

The key information of the present embodiment may be represented by a vector, and each disease feature in the key information may be a vector of one dimension.

The auxiliary inspection information in the present embodiment includes: one or more of pathological examination information, ultrasonic auxiliary examination information, X-ray auxiliary examination information, CT and nuclear magnetic auxiliary examination information;

S3, processing the key information by adopting a pre-trained typical symptom diagnosis model according to a pre-established knowledge base system of the children diseases to obtain a diagnosis result; as shown in fig. 5.

The diagnostic result includes: at least one possible disease, and each possible disease corresponds to a characteristic of the child to be diagnosed.

The typical symptom diagnostic model in this embodiment may be a BP neural network model with pre-trained input, convolution, full connection, etc., according to adaptive changes in threshold, weight, etc. in the BP neural network model after each different disease feature is trained.

For example, the following steps may be performed to build a knowledge base system for a disease of a child and train a BP neural network model based on the built knowledge base system for a disease of a child before step S3.

A01, collecting various symptom information and diagnosis information of the difficult and complicated diseases of the respiratory department of the children.

For example, the diagnostic information includes: diagnostic thinking, diagnostic methods, diagnostic procedures, manifestations of disease symptoms, differential diagnosis of disease. The diagnostic information includes text information and/or image data.

A02, establishing a term dictionary of the dyspnea and the difficult and complicated diseases and information of each case in the term dictionary according to the collected symptom information, the diagnosis information and the priori knowledge of an expert.

In this embodiment, the term dictionary includes a plurality of classification information, and each classification information may include a plurality of case information.

A03, constructing a child disease knowledge base system comprising a training set, a verification set and a test set through the case information in the term dictionary;

a04, training the established typical symptom diagnosis model by adopting the constructed knowledge base system for the children diseases, and obtaining the trained typical symptom diagnosis model.

In the training process, aiming at the defect that the distribution of each disease in the training set is uneven, in this embodiment, each disease can be balanced by adopting a data up-sampling mode, for example, an oversampled/undersampled method is adopted for processing.

The above description is directed to the currently pediatric corresponding respiratory dyscrasia.

Of course, in practical application, the collection of each symptom of the respiratory department is also possible, and the collection is not necessarily limited to the difficult and complicated respiratory department. At this time, the step a01 can be adaptively adjusted as follows: collecting various symptom information and diagnosis information of the children respiratory department.

The partial classification information of the term dictionary is shown as follows.

Sequence number	Department of science	ICD coding	Disease name
				1	Department of respiratory	B37.101+J99.8*	Pulmonary fungal infection-pulmonary candidiasis
2	Department of respiratory	B44.101+J99.8*	Allergic bronchopulmonary aspergillosis
				3	Department of respiratory	B44.102+J99.8*	Pulmonary fungal infection-aspergillosis
4	Department of respiratory	B45.001+J99.8*	Lung fungus infection-Cryptococcosis pneumonecosis
				5	Department of respiratory	C85.908	Lung lymphoma
6	Department of respiratory	C95.0018	Lymphocytic interstitial pneumonia
				7	Department of respiratory	D71xx03	Chronic granulomatosis
8	Department of respiratory	D76.015	Histiocytosis of langerhans cells
				9	Department of respiratory	D80.501	Immunodeficiency disease with high lgM
10	Department of respiratory	D81.901	Combined immunodeficiency disease
				11	Department of respiratory	D83.901	Common variant immunodeficiency disease
12	Department of respiratory	E84.902	Cystic fibrosis
				13	Department of respiratory	J47xx01	Bronchiectasis
14	Department of respiratory	J67.902	Allergic alveolitis
				15	Department of respiratory	J69.001	Aspiration pneumonia
16	Department of respiratory	J84.001	Alveolar proteinosis
				17	Department of respiratory	J84.003	Diffuse alveolar hemorrhage
18	Department of respiratory	J84.902	Interstitial pneumonia
				19	Department of respiratory	M30.101	Eosinophilic granulomatous vasculitis
20	Department of respiratory	G24.902	Primary ciliated dyskinesia

The child disease knowledge base system in this embodiment may include a training set, a validation set, and a test set.

In some embodiments of the present application, a plurality of terms may be collected from various websites or web pages, such as Baidu encyclopedia/medical searching, electronic medical record data, etc., to construct a multi-type term dictionary; and can adopt CRF (conditional random field), hidden Markov, deep learning method, etc. to process word segmentation, in this embodiment, the pediatric transfer can correct the word segmentation information to obtain the corpus in the term dictionary.

Each word in the corpus is represented by a vector, so that keywords in the corpus can be characterized in a vectorization manner.

Because of the large dimensions of the vector, there are fewer non-0 elements, and therefore, storage is typically done in a sparse manner. For example, each word is assigned a numerical ID, e.g., "fever" is noted 3 and "fever" is noted 5 (assuming a tag starting from 0). When programming is realized, the number assignment work of each word can be completed through the Hash table. Word vectors may be trained using a Skip-gram model.

Further, the electronic medical record can be processed into a format which can be recognized by a format computer required by model input from the text of the electronic medical record, for example, the electronic medical record is extracted from a database and then processed into the format required by model input; and acquiring image data such as CT, ultrasound, MR, pathology and the like from a system of a hospital, and performing preprocessing operation (such as standardization or binarization processing) on the image data.

Since the diagnostic information includes text information and image data, the standardized processing of the image data may include:

the image data of the training set, the verification set and the test set are subjected to standardized processing, for example, in this embodiment, the mean value and the standard deviation of the training set are calculated first, and then the data of the verification set and the test set are subjected to standardized processing according to the mean value and the standard deviation of the training set, so that the data set including the classification information and the diagnosis information is processed by using a unified standard. In addition, data enhancement operation can be performed on the data of the training set, so that the data volume in the data set processing process is increased, and the training precision in the data set processing process is improved.

The child disease knowledge base system in this embodiment is a term dictionary expressed by vectorization, and each term dictionary has a classification of a multidimensional vector, each dimension can be understood as a disease feature in the classification, and weights of multiple classified disease features are also expressed by a vector.

In addition, the model used in the present embodiment is a BP neural network model. The process of BP neural network training may be described as follows:

as shown in fig. 3 and 4, in the diagnostic model, a BP network may be applied to build a disease prediction model, and the basic process of modeling is as follows: firstly, collecting main influencing factors and disease occurrence results of disease occurrence, such as a training set, a testing set and a verification set of a children disease knowledge base system; and inputting influencing factors and disease results into the designed neural network model for repeated training until the network converges (i.e. the expected training error is reached), adopting a certain skill in the training process to lead the training speed of the network to be the fastest, the error to be the smallest and the model to be the optimal, and finally, using the established model for disease prediction.

Design of input layer

The BP neural network is constructed, and the number of nodes of an input layer depends on the dimension of an input vector, namely the dimension of a feature vector. Therefore, when the respiratory disease diagnosis simulation system is designed by the respiratory department, variables related to the disease, namely age z, symptoms (cough.z, chest pain.22.), fever z, expectoration zs, hemoptysis e, dyspnea z, cold sweat zs, pestle fingers z.) and examination (radiographic examination z., bronchoscopy z) are selected as input data in respiratory disease diagnosis, and the feature vector dimension is obtained after preprocessing of the data so as to determine the network input layer node number.

Design of hidden layer

Hidden layer (i.e., hidden layer) node number design

In the BP network, the selection of the hidden layer node number is very important, the hidden layer node number has great influence on the performance of the established neural network model, and if the hidden layer node number is too small, the network can not be trained at all or the network performance is poor; if the number of hidden nodes is too large, the systematic error of the network can be reduced, but on one hand, the training time of the network is prolonged, on the other hand, the training is easy to sink into local minimum points, so that the optimal point cannot be obtained, and the hidden nodes are also the internal cause of 'overfitting' in the training. Therefore, the reasonable hidden layer node number should be determined by a node deletion method and an expansion method under the condition of comprehensively considering the complexity degree and the error magnitude of the network structure.

In this embodiment, the number of hidden nodes is determined, so that the phenomenon of "overfitting" during training is avoided as much as possible, and the network performance and generalization capability which are high enough are ensured, and the most basic principle for determining the number of hidden nodes is as follows: under the premise of meeting the precision requirement, the structure is as compact as possible, namely the number of hidden layer nodes is as small as possible. According to the empirical formulah is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a constant. Alternatively, the empirical formula is h=n (input layer node number) +m (output layer node number) +a (constant).

Learning step size

The learning step length influences the stability of the BP neural network learning process. The large learning rate may cause the correction amount of the network weight to be too large every time, and even cause the weight to exceed the minimum value of a certain error in the correction process to take irregular jump and not converge; however, too small a learning rate results in too long a learning time, but convergence to a certain minimum value can be ensured. The general trend is to choose a smaller learning rate to ensure convergence (stability) of the learning process, typically between 0.01 and 0.8.

Initial connection weights for networks

The BP algorithm determines that the error function generally has a plurality of local minimum points, and different network initial weights directly determine which local minimum point or global minimum point the BP algorithm converges to.

Determination of transfer function

The BP neural network commonly comprises a linear transfer function purelin, a tan-sigmoid transfer function tan sig and a log-sigmoid transfer function logsig.

1) The linear transfer function is a linear transfer function that computes the output of the output layer from the input of the network in the form of: f (x) =purelin (x).

2) The tangent sign-type transfer function is used to map the input range of neurons from plus infinity to minus infinity to (-1, 1). The tangent Sigmoid type transfer function tan () is a micro-able function and is therefore well suited for training neural networks using the BP algorithm, which is chosen as the transfer function for the hidden and output layers of the network in this embodiment. The format is as follows: f (x) =tan sig (x).

3) The logarithmic Sigmoid type transfer function is used to map the input range of neurons from positive infinity to negative infinity to (0, 1). The logarithmic Sigmoid-type transfer function logsig () is also a microtransaction suitable for training neural networks using the BP algorithm, in the format: f (x) =logsig (x).

Design of output layer

The BP neural network has only one output layer, and the number of nodes is determined by the number of output vectors. If the respiratory study is 10 diseases (pneumonia, asthma, nervous system binding, acute bronchitis … …), the output vector is 10.

Training of neural network systems

The training of the BP network is to continuously adjust the network weight by applying the error back-transmission principle so that the square sum of the error between the network model output value and the known training sample output value is minimum or smaller than a certain expected value. And (3) a reasonable BP neural network model is built through the study (training) of training samples in a training set, the trained BP neural network model is utilized, the simulation diagnosis is carried out by using data in a test set and a verification set, the experimental result is compared with the diagnosis of a doctor with abundant experience, and the diagnosis coincidence rate is more than 90 percent and is qualified.

Example 2

As shown in fig. 2, this embodiment provides an intelligent auxiliary diagnosis system for respiratory diseases of children, including:

the processing module is used for screening the key information according to the auxiliary examination information and/or doctor inquiry record information; processing the key information by adopting a pre-trained typical symptom diagnosis model according to a pre-established knowledge base system of the children diseases to obtain a diagnosis result;

the display module is used for displaying the diagnosis result, and the diagnosis result comprises: at least one possible disease, and each possible disease corresponds to a characteristic of the child to be diagnosed.

As shown in fig. 2, the first information collecting module includes:

In practical application, the first information collection module and the display module are simultaneously located on an information display interface of the auxiliary diagnosis system.

Through machine learning, a typical symptom diagnosis model is formed in an early stage of occurrence of a difficult and complicated respiratory disease/respiratory department. Firstly, in the clinical standard diagnosis and treatment quality control process of respiratory department cooperatives in the children hospital, high-sensitivity screening reminding is formed for key clinical work elements; secondly, in the critical difficult and complicated disease parting work, high-specificity auxiliary accurate diagnosis parting is realized, and auxiliary judgment is made on the state risk and subsequent transition of the disease.

The product is simple to operate, has a particularly wide application range, is suitable for various medical and health systems, and is used for assisting in checking according to disease symptoms which are arranged on the left side of the product by respiratory specialists in Beijing children hospitals, including the age of patients, the onset of the disease, main symptoms, special examination, past history family history and auxiliary examination; the right side shows the disease diagnosis results. The doctor can display the diagnosis result in real time by clicking the symptom expression of the patient.

The diagnostic result format provided on the right side of the intelligent diagnosis is disease name + diagnostic weight, and the diagnostic results are arranged from large to small according to the weight. Typical symptoms of the disease are listed below for each diagnosis, with blue highlighting indicating that the current patient is present and gray indicating that the current patient is not present. If the diagnosis result of the disease is difficult and complicated symptoms, the upper right corner gives a red prompt to remind the patient of the possibility of the difficult and complicated symptoms, and the doctor can be helped to diagnose the difficult and complicated symptoms in a differential mode.

It should also be noted that the exemplary embodiments mentioned in this disclosure describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the above-described steps, that is, the steps may be performed in the order mentioned in the embodiments, or may be performed in a different order from the order in the embodiments, or several steps may be performed simultaneously.

The above embodiments may be referred to each other, and the present embodiment is not limited to the embodiments.

Finally, it should be noted that: the embodiments described above are only for illustrating the technical solution of the present application, and are not limiting; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced with equivalents; such modifications and substitutions do not depart from the spirit of the application.

Claims

1. An intelligent auxiliary diagnostic system for respiratory diseases of children, which is characterized by comprising:

2. The intelligent auxiliary diagnostic system according to claim 1, wherein,

the first information collection module includes:

3. The intelligent auxiliary diagnostic system of claim 1, wherein the first information collection module and the presentation module are simultaneously located at an information presentation interface of the auxiliary diagnostic system.

4. The intelligent aided diagnosis system of claim 1, wherein the typical symptom diagnosis model is BP neural network model,