CN112951412A - Auxiliary diagnosis method and application thereof - Google Patents
Auxiliary diagnosis method and application thereof Download PDFInfo
- Publication number
- CN112951412A CN112951412A CN202110264436.1A CN202110264436A CN112951412A CN 112951412 A CN112951412 A CN 112951412A CN 202110264436 A CN202110264436 A CN 202110264436A CN 112951412 A CN112951412 A CN 112951412A
- Authority
- CN
- China
- Prior art keywords
- chest
- diagnosis
- model
- pneumonia
- ray
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 71
- 238000000034 method Methods 0.000 title claims abstract description 45
- 206010035664 Pneumonia Diseases 0.000 claims abstract description 49
- 238000011976 chest X-ray Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 15
- 201000010099 disease Diseases 0.000 claims abstract description 13
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 13
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- 238000013145 classification model Methods 0.000 claims description 22
- 206010011224 Cough Diseases 0.000 claims description 13
- 206010037660 Pyrexia Diseases 0.000 claims description 13
- 206010037833 rales Diseases 0.000 claims description 13
- 208000037656 Respiratory Sounds Diseases 0.000 claims description 11
- 206010008479 Chest Pain Diseases 0.000 claims description 10
- 208000000059 Dyspnea Diseases 0.000 claims description 10
- 206010013975 Dyspnoeas Diseases 0.000 claims description 10
- 208000000616 Hemoptysis Diseases 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 208000008745 Healthcare-Associated Pneumonia Diseases 0.000 claims description 5
- 206010061218 Inflammation Diseases 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 208000015181 infectious disease Diseases 0.000 claims description 4
- 230000002458 infectious effect Effects 0.000 claims description 4
- 230000004054 inflammatory process Effects 0.000 claims description 4
- 210000004072 lung Anatomy 0.000 claims description 4
- 206010006451 bronchitis Diseases 0.000 claims description 3
- 230000008602 contraction Effects 0.000 claims description 3
- 230000008451 emotion Effects 0.000 claims description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 4
- 230000002349 favourable effect Effects 0.000 abstract description 2
- 210000000038 chest Anatomy 0.000 description 10
- 208000024891 symptom Diseases 0.000 description 9
- 239000003814 drug Substances 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 206010011409 Cross infection Diseases 0.000 description 2
- 238000004195 computer-aided diagnosis Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000003864 humus Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000241 respiratory effect Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 206010029803 Nosocomial infection Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000020169 heat generation Effects 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000000399 orthopedic effect Effects 0.000 description 1
- 238000002601 radiography Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000001931 thermography Methods 0.000 description 1
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
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Biology (AREA)
- Pathology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The application belongs to the technical field of artificial intelligence, and particularly relates to an auxiliary diagnosis method and application thereof. The prior art can only diagnose pneumonia through images, and lacks a pneumonia diagnosis model trained under a large data set. The application provides an auxiliary diagnosis method, which collects a chest radiograph data set; dividing the chest X-ray image data set into first chest X-ray data and second chest X-ray data, and training a convolutional neural network model by adopting the first chest X-ray data; then inputting the second chest X-ray data into a trained convolutional neural network model to output a probability value of a diagnosis result, dividing the probability value into '0' or '1', and taking the '0' or '1' as a characteristic value of a chest X-ray map; extracting diagnostic information from a textual report, the textual report corresponding to the chest radiograph dataset; and inputting the diagnosis information and the characteristic value as new data into the interpretation model, and outputting the disease probability. The relation between each node can be better explained, and more favorable information is provided for doctors.
Description
Technical Field
The application belongs to the technical field of artificial intelligence, and particularly relates to an auxiliary diagnosis method and application thereof.
Background
With the rapid development of AI technology, computer-aided diagnosis has attracted much attention and has been successfully applied in many applications of healthcare and medicine. For certain specific tasks, learning-based systems may compete with, or even surpass, the performance of human experts. Although the intuition of the model is often not explicitly expressed, the impressive performance is attributed to the well-behaved and scalable nature of neural networks. However, for computer aided diagnosis, interpretability is very important, even as accurate as diagnosis. The invention mainly researches how to fuse various data and extracts the relationship among the data through the fusion of the various data, thereby achieving the interpretability of the pneumonia diagnosis model.
At present, a plurality of pneumonia auxiliary diagnosis models exist, however, the diagnosis model methods have various defects. These models have the same characteristics, namely, the chest radiograph is input, but the chest radiograph cannot be directly used as the basis for diagnosing pneumonia. When a doctor diagnoses pneumonia, the doctor needs to know some other clinical manifestations of a patient, such as whether cough occurs or not, whether fever occurs or not, and some physical examinations are carried out, mainly to check whether crackle exists or not. The chest radiograph is an important diagnostic basis, but other important clinical information is also an important part of diagnosis.
In terms of interpretable models, Grad CAMs are the more widely interpretable methods currently in use. The interpretable method can explain the focus of the model on the image to a doctor during diagnosis, and provides the image interpretation in a thermal imaging mode. However, this method is also only used for image interpretation and cannot reasonably interpret other information required for diagnosis.
The existing technology can only diagnose pneumonia through images, and a pneumonia diagnosis model trained under a large data set is lacked, and the diagnosis result of pneumonia cannot be reasonably explained.
Disclosure of Invention
1. Technical problem to be solved
Based on the problems that the pneumonia can only be diagnosed through images in the prior art, a pneumonia diagnosis model trained under a large data set is lacked, and reasonable explanation on the diagnosis result of the pneumonia cannot be carried out, the application provides an auxiliary diagnosis method and application thereof.
2. Technical scheme
In order to achieve the above object, the present application provides a diagnosis assisting method, comprising the steps of: step 1: collecting a transthoracic data set; step 2: dividing the chest X-ray image data set into first chest X-ray data and second chest X-ray data, and training a convolutional neural network model by adopting the first chest X-ray data; then inputting the second chest X-ray data into a trained convolutional neural network model to output a probability value of a diagnosis result, dividing the probability value into '0' or '1', and taking the '0' or '1' as a characteristic value of a chest X-ray map; and step 3: extracting diagnostic information from a textual report, the textual report corresponding to the chest radiograph dataset; and 4, step 4: and inputting the diagnosis information and the characteristic value as new data into an interpretation model, and outputting the disease probability.
Another embodiment provided by the present application is: the pre-training image classification model is trained for 3 times to obtain 3 image classification models, the three image classification models are respectively tested to obtain the average value of the performance, and the generation of special cases of the trained models is prevented, so that the effect of the models can be better explained by obtaining the average performance of the models by training for three times.
Another embodiment provided by the present application is: and (4) obtaining a characteristic value of the output of the image classification model from the probability value chest chart by using the Johnson index as a threshold value.
Another embodiment provided by the present application is: the image classification model is a dense convolution network, and the interpretation model is a constraint-based Bayesian network model.
The application also provides an application of the auxiliary diagnosis method, and the auxiliary diagnosis method is applied to the diagnosis of pneumonia, the auxiliary diagnosis of some diseases such as bronchitis and the like, and can also be applied to scenes needing to be identified through multi-modal input such as emotion analysis.
Another embodiment provided by the present application is: the diagnostic information includes cough, hemoptysis, chest pain, fever, dyspnea, humectasia and rale 7-dimensional vector.
Another embodiment provided by the present application is: the constraint-based bayesian network model is improved according to a contraction algorithm.
Another embodiment provided by the present application is: the pneumonia includes infectious lung parenchymal inflammation that is suffered outside the hospital and nosocomial pneumonia.
Another embodiment provided by the present application is: the extraction of the diagnosis information comprises the steps of extracting information from the main complaints and the subjects, segmenting the main complaints by adopting Chinese punctuations, extracting and storing short sentences containing key words, marking the short sentences, and simultaneously carrying out manual examination on the short sentences.
Another embodiment provided by the present application is: the constraint-based Bayesian network model can explain the relationship among nodes and the parameter value of each node in the diagnosis process.
3. Advantageous effects
Compared with the prior art, the auxiliary diagnosis method provided by the application has the beneficial effects that:
the auxiliary diagnosis method provided by the application can be used for diagnosing by combining the image and the diagnosis report through training of the large data set, so that the diagnosis reliability is greatly improved, and the result can be explained to a certain extent.
The auxiliary diagnosis method provided by the application supports various data as input, and the model has certain interpretability and can provide more credible results for doctors.
According to the auxiliary diagnosis method, the Bayesian network model is used as the classification model, certain interpretability can be provided, the relation between every two nodes can be better explained after a large number of data sets are learned, and more favorable information is provided for doctors.
The application of the auxiliary diagnosis method provided by the application assists a doctor in diagnosing pneumonia, so that the workload of the doctor is reduced, and the misdiagnosis rate of the doctor is reduced.
Drawings
Fig. 1 is a system schematic diagram of the auxiliary diagnostic method of the present application.
Detailed Description
Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, and it will be apparent to those skilled in the art from this detailed description that the present application can be practiced. Features from different embodiments may be combined to yield new embodiments, or certain features may be substituted for certain embodiments to yield yet further preferred embodiments, without departing from the principles of the present application.
With the rapid development of artificial intelligence, the auxiliary diagnosis of diseases is more and more widely applied. Pneumonia is one of the most common diseases, and the demand of assisting doctors in diagnosing pneumonia by using artificial intelligence is increasingly strong. The pneumonia diagnosis assisting device is mainly used for assisting doctors in diagnosing pneumonia, so that the workload of the doctors is reduced, and the misdiagnosis rate of the doctors is reduced.
The Bayesian network, also called belief network, is an extension of Bayes method, and is one of the most effective theoretical models in uncertain knowledge expression and reasoning field at present. Since its introduction by Pearl in 1988, it has become a hot spot of research in recent years. A bayesian network is a Directed Acyclic Graph (DAG) consisting of nodes representing variables and Directed edges connecting these nodes. The nodes represent random variables, the directed edges among the nodes represent the mutual correlation system (from father nodes to child nodes), the relation strength is expressed by conditional probability, and the prior probability is used for expressing information without father nodes. The node variables may be abstractions of any problem, such as test values, observations, opinion polls, etc. The method is applicable to expressing and analyzing uncertain and probabilistic events, and to making decisions that are conditionally dependent on a variety of control factors, and can make inferences from incomplete, inaccurate, or uncertain knowledge or information.
Youden index (Youden index): also called correct index, is a method for evaluating the authenticity of a screening test, and can be applied when the harmfulness of false negative (missed diagnosis rate) and false positive (misdiagnosis rate) is equal. The jotan index is the sum of sensitivity and specificity minus 1. Indicating that the screening method finds true patient and non-patient total ability. The larger the index, the better the screening experiment and the greater the authenticity.
Complaints, medical and psychological terms. It is the patient (visitor) who self-states his symptoms or (and) signs, nature, and duration.
The chief complaint is the first content in the medical record of the hospitalization, and the good chief complaint needs to be refined and accurate; the symptoms described by the patient are used as much as possible without diagnostic terms; to be consistent with the current medical history; the principle of objective and practical is followed.
Physical examination is performed by a doctor through a stethoscope or the like.
Referring to fig. 1, the present application provides a method of aiding diagnosis, the method comprising the steps of: step 1: collecting a transthoracic data set; step 2: dividing the chest X-ray image data set into first chest X-ray data and second chest X-ray data, and training a convolutional neural network model by adopting the first chest X-ray data; then inputting the second chest X-ray data into a trained convolutional neural network model to output a probability value of a diagnosis result, dividing the probability value into '0' or '1', and taking the '0' or '1' as a characteristic value of a chest X-ray map; and step 3: extracting diagnostic information from a textual report, the textual report corresponding to the chest radiograph dataset; and 4, step 4: and inputting the diagnosis information and the characteristic value as new data into an interpretation model, and outputting the disease probability.
The pneumonia data set of 35389 cases is constructed, and when the deep learning model is used, the performance of the model is greatly improved by the large pneumonia data set.
Further, the pre-trained image classification model is trained for 3 times to obtain 3 image classification models, then the three models are respectively tested, and then the average value of the performance is calculated, so that the generation of special cases of the trained models is prevented, and the effect of the models can be better explained by calculating the average performance of the models three times.
Further, the output of the image classification model is extracted from the characteristic value of the probability value thorax map by using the john index as a threshold value.
Further, the image classification model is a dense convolution network, and the interpretation model is a constraint-based Bayesian network model. A constraint-based Bayesian network model capable of combining medical common knowledge is constructed, and the model can be used for constructing the association among nodes more reasonably.
The application also provides an application of the auxiliary diagnosis method, and the auxiliary diagnosis method is applied to the diagnosis of pneumonia, the auxiliary diagnosis of some diseases such as bronchitis and the like, and can also be applied to scenes needing to be identified through multi-modal input such as emotion analysis.
Further, the diagnostic information includes 7-dimensional vectors of cough, hemoptysis, chest pain, fever, dyspnea, humectasia and rale.
The used data not only contains images, but also contains information extracted from a diagnosis report, the image data is firstly learned through a convolutional neural network to obtain a trained network to provide characteristics in the images, then 7 pieces of clinical information including cough, hemoptysis, chest pain, fever, dyspnea, moist rales and dry rales are extracted from the diagnosis report, and finally all the characteristics are used as a mode of inputting the training Bayesian network, so that the multi-input and multi-modal capability can be achieved by combining the images and the characters.
Further, the constraint-based Bayesian network model is refined according to a contraction algorithm.
Further, the pneumonia includes infectious lung parenchymal inflammation and nosocomial pneumonia that are suffered outside the hospital.
Further, the extraction of the diagnosis information comprises extracting information from the main complaints and the subjects, segmenting the main complaints by adopting Chinese punctuations, extracting and storing short sentences containing key words, marking the short sentences, and simultaneously carrying out manual examination on the short sentences.
Further, the constraint-based Bayesian network model can explain the relationship between nodes and the parameter value of each node in the diagnosis process.
The model provides certain interpretability, and compared with the unexplainable performance of other existing deep learning models, the model has the advantage that the trust degree of a doctor on the model is increased. The Bayesian network itself serves as an interpretable model, and the construction of the Bayesian network enables interpretation of the relationships between the respective nodes and the parameter values of each node during the diagnosis process, thereby serving as an interpretation.
Examples
From the disease environment classification, pneumonia can be classified into CAP (infectious lung parenchymal inflammation occurring outside the hospital) and HAP (nosocomial pneumonia). In this work, CAP, i.e. pneumonia of extra-hospital infections, is of major concern, while avoiding the observation of complex and very disturbing cases of nosocomial pneumonia that is infected with reduced immunity due to other severe diseases. The doctor is inquired and a large amount of medical record data is checked to know that pneumonia patients in other departments, such as heart departments, orthopedics departments, oncology departments and the like, are more likely to have pneumonia caused by cross infection in hospitals after hospitalization, therefore, the data mainly come from three departments, namely the department of respiratory medicine, the department of respiratory and critical medical science and the department of pediatrics in the second people hospital of Guangdong province, and the pneumonia patients in the three departments are mostly suffered from the diseases in the community through manual medical record checking and doctor inquiry. Meanwhile, pneumonia cases are identified through ICD-10 coding values (J12-J18 are pneumonia) typed by a coding doctor in medical record diagnosis.
The data set was collected for a total of 35389 pneumonia cases including 13482 patients with pulmonary inflammation and 21907 common patients who were hospitalized and outpatient during 2016 to 2020 (to date). Some cases with pneumonia also have other diseases, while most cases without pneumonia have other diseases, not have them.
The required information needs to be extracted from the patient's medical record and made a valid marker of the bayesian network. Each patient (case) has its own label. According to the internal medicine of the teaching material of the human defense edition, whether cough, hemoptysis, chest pain, fever, dyspnea, humus and rale are important criteria for diagnosing pneumonia or not on clinical symptoms, therefore, relevant information of cough, hemoptysis, chest pain, fever and dyspnea is extracted from recorded chief complaints, and relevant information of humus and rale is extracted from recorded corpus. If the patient has symptoms, the corresponding bit in the label is set to "1", for example, the label "1001001" indicates that the patient has cough, fever, and humorous, and has no other symptoms.
The present application relates to a simple algorithm to extract information from complaints and physical examinations. Take "heat generation" information as an example: first, when recording whether a patient is hot, doctors generally use two words of "hot" and "no hot", so that the Chinese punctuations "are used,". "and"; "to segment the main complaint, and then extract and save the short sentence containing" hot ". When the keyword in the sentence is "hot", the bit of "hot" is marked as "1", otherwise "0". Secondly, considering that some doctors may have special expression modes, manually checking each short sentence containing 'hot', and if a special writing form is found, correcting the label. The extraction method of other symptoms is the same as the above-mentioned method.
After the image data of a case is put into a trained image convolution neural network model to run out the prediction probability and 0/1, other 7 clinical symptoms (0/1) of the same case and the prediction result of a simple image model are used as input, and the Bayesian network automatically calculates the probability of a default node (whether pneumonia occurs or not).
Specifically, the auxiliary pneumonia diagnosis method is based on a large-scale chest radiography data set, and comprises the following steps:
s1, training a pre-trained image classification model on CheXpert using the deep-learned model, using DenseNet121 as the model. DenseNet121 proposes a more fundamental dense connection mechanism compared to the traditional network. All layers are connected; specifically, each layer accepts all previous layers as its additional inputs. This connection enhances the reusability of functionality and allows the final classifier to make decisions based on all the characteristics of the entire network. The completed image classification model trained using this data set resulted in a test in the CheXpert validation set with an area under the subject's working characteristic curve of 0.74.
S2, next, the image classification model is continued to be trained using the transthoracic image dataset. In order to improve the reliability of the experiment and reduce accidental errors, the training is carried out three times to obtain 3 image classification models, and the average area under the working characteristic curve of the testee is 0.90. Its area value under the subject's working characteristic curve increased by 0.16 compared to the last test on the ChecXpert validation set.
S3, predict the chest radiograph dataset by the trained DenseNet121, and convert the output of the image classification model from a probability value to 0 or 1 using the johnson index as a threshold.
S4, extracting 7-dimensional vectors from the report dataset corresponding to the transthoracic map dataset, where the 7-dimensional vectors respectively indicate whether the patient has cough, hemoptysis, chest pain, fever, dyspnea, humorous, and rale, and if the patient has symptoms thereof, the corresponding bit in the label is set to "1", for example, the label is "1001001" indicating that the patient has cough, fever, and humorous, and has no other symptoms.
S5, connecting the output of the image classification model with each 7-dimensional vector extracted from the report, constructing a new input as in fig. 1, and using the new input as an input for training the bayesian network classification model to classify pneumonia. During the training process, 17040 cases were shared, including 21881 pictures (transthoracic images), which do not overlap with the data from the training convolutional neural network.
When the Bayesian network is constructed, how to construct the Bayesian network by combining data and medical common knowledge is the key point. Therefore, the feature nodes extracted by the constructed bayesian network comprise cough, hemoptysis, chest pain, fever, dyspnea, wet rale, dry rale and pictures, wherein the pictures represent the results output by the chest radiograph through the trained DenseNet121, and in medical science, the feature nodes are all necessary information for the diagnosis of a clinician, so that the feature nodes cannot be mutually independent from pneumonia nodes, the same feature nodes are required to be in the same Markov blanket with the pneumonia nodes, and all the feature nodes are required to point to pneumonia.
To represent these medical wisdom, a new Algorithm was proposed to construct a bayesian network, which was improved based on the shrinkage Algorithm (Grow-spring (gs) Algorithm), and the detailed Algorithm is described as follows:
1. computing a Markov blanket: let U be the set of all nodes, i.e. U { "cough", "hemoptysis", "chest pain", "fever", "dyspnea", "wet rale", "picture" }, first calculate all nodes except "picture" node, then self-define the markov blanket of "picture" node as { "cough", "hemoptysis", "chest pain", "fever", "dyspnea", "wet rale", "rale", thus can obtain markov blanket b (x).
2. Let T be the smaller of B (X) -Y and B (Y) -X for all nodes X E U, Y E B (X), X and Y are connected assuming that X and Y are interdependent with S for all S T.
3. For X epsilon to U- { X0Y ∈ N (X), T is the smaller of B (Y) - { X, Z } and B (Z) - { X, Y }, N (X) represents a node directly connected to X, and X0Representing a node "picture", if Z ∈ N (X) -N (Y) - { Y }, then Y → X. All of and X0The direct connected nodes point to X directions for the nodes0。
And S6, testing the pneumonia diagnosis model and checking the performance of the model.
Although the present application has been described above with reference to specific embodiments, those skilled in the art will recognize that many changes may be made in the configuration and details of the present application within the principles and scope of the present application. The scope of protection of the application is determined by the appended claims, and all changes that come within the meaning and range of equivalency of the technical features are intended to be embraced therein.
Claims (10)
1. An aided diagnosis method, comprising: the method comprises the following steps:
step 1: collecting a transthoracic data set;
step 2: dividing the chest X-ray image data set into first chest X-ray data and second chest X-ray data, and training a convolutional neural network model by adopting the first chest X-ray data; then inputting the second chest X-ray data into a trained convolutional neural network model to output a probability value of a diagnosis result, dividing the probability value into '0' or '1', and taking the '0' or '1' as a characteristic value of a chest X-ray map;
and step 3: extracting diagnostic information from a textual report, the textual report corresponding to the chest radiograph dataset;
and 4, step 4: and inputting the diagnosis information and the characteristic value as new data into an interpretation model, and outputting the disease probability.
2. The aided diagnosis method according to claim 1, characterized in that: and training the pre-training image classification model for 3 times to obtain 3 image classification models, and testing the three image classification model models respectively to obtain an average value of performances.
3. The aided diagnosis method according to claim 2, characterized in that: and (4) obtaining a characteristic value of the output of the image classification model from the probability value chest chart by using the Johnson index as a threshold value.
4. The aided diagnosis method according to claim 1, characterized in that: the image classification model is a dense convolution network, and the interpretation model is a constraint-based Bayesian network model.
5. The application of the auxiliary diagnosis method is characterized in that: the use of the method for aiding diagnosis according to any one of claims 1 to 4 for diagnosis of pneumonia or bronchitis and emotion analysis.
6. Use of the method of aided diagnosis according to claim 5, characterized in that: the diagnostic information includes cough, hemoptysis, chest pain, fever, dyspnea, humectasia and rale 7-dimensional vector.
7. Use of the method of aided diagnosis according to claim 5, characterized in that: the constraint-based bayesian network model is improved according to a contraction algorithm.
8. Use of the method of aided diagnosis according to claim 5, characterized in that: the pneumonia includes infectious lung parenchymal inflammation that is suffered outside the hospital and nosocomial pneumonia.
9. Use of the method of aided diagnosis according to claim 5, characterized in that: the extraction of the diagnosis information comprises the steps of extracting information from the main complaints and the subjects, segmenting the main complaints by adopting Chinese punctuations, extracting and storing short sentences containing key words, marking the short sentences, and simultaneously carrying out manual examination on the short sentences.
10. Use of the method of aided diagnosis according to claim 5, characterized in that: the constraint-based Bayesian network model can explain the relationship among nodes and the parameter value of each node in the diagnosis process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110264436.1A CN112951412A (en) | 2021-03-11 | 2021-03-11 | Auxiliary diagnosis method and application thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110264436.1A CN112951412A (en) | 2021-03-11 | 2021-03-11 | Auxiliary diagnosis method and application thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112951412A true CN112951412A (en) | 2021-06-11 |
Family
ID=76228565
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110264436.1A Pending CN112951412A (en) | 2021-03-11 | 2021-03-11 | Auxiliary diagnosis method and application thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112951412A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117059283A (en) * | 2023-08-15 | 2023-11-14 | 宁波市鄞州区疾病预防控制中心 | Speech database classification and processing system based on pulmonary tuberculosis early warning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107038343A (en) * | 2017-04-10 | 2017-08-11 | 荣科科技股份有限公司 | Construction method and constructing system, the clinical diagnosing system of clinical diagnosis model |
| CN108846840A (en) * | 2018-06-26 | 2018-11-20 | 张茂 | Lung ultrasound image analysis method, device, electronic equipment and readable storage medium storing program for executing |
| CN109222914A (en) * | 2018-11-06 | 2019-01-18 | 北京三医智慧科技有限公司 | The information collection of Chinese medicine facial diagnosis and intelligent processing method based on deep learning and multi-optical spectrum imaging technology |
| CN111653356A (en) * | 2020-04-20 | 2020-09-11 | 浙江大学 | New coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning |
| CN112116552A (en) * | 2019-06-21 | 2020-12-22 | 斯特拉克斯私人有限公司 | Image analysis method and system |
| CN112382392A (en) * | 2020-11-25 | 2021-02-19 | 珠海圣美生物诊断技术有限公司 | System for be used for pulmonary nodule risk assessment |
-
2021
- 2021-03-11 CN CN202110264436.1A patent/CN112951412A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107038343A (en) * | 2017-04-10 | 2017-08-11 | 荣科科技股份有限公司 | Construction method and constructing system, the clinical diagnosing system of clinical diagnosis model |
| CN108846840A (en) * | 2018-06-26 | 2018-11-20 | 张茂 | Lung ultrasound image analysis method, device, electronic equipment and readable storage medium storing program for executing |
| CN109222914A (en) * | 2018-11-06 | 2019-01-18 | 北京三医智慧科技有限公司 | The information collection of Chinese medicine facial diagnosis and intelligent processing method based on deep learning and multi-optical spectrum imaging technology |
| CN112116552A (en) * | 2019-06-21 | 2020-12-22 | 斯特拉克斯私人有限公司 | Image analysis method and system |
| CN111653356A (en) * | 2020-04-20 | 2020-09-11 | 浙江大学 | New coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning |
| CN112382392A (en) * | 2020-11-25 | 2021-02-19 | 珠海圣美生物诊断技术有限公司 | System for be used for pulmonary nodule risk assessment |
Non-Patent Citations (1)
| Title |
|---|
| 王双成著: "《贝叶斯网络学习、推理与应用》", 立信会计出版社, pages: 159 - 166 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117059283A (en) * | 2023-08-15 | 2023-11-14 | 宁波市鄞州区疾病预防控制中心 | Speech database classification and processing system based on pulmonary tuberculosis early warning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Khobahi et al. | Coronet: A deep network architecture for semi-supervised task-based identification of covid-19 from chest x-ray images | |
| Shahin et al. | Machine learning approach for autonomous detection and classification of COVID-19 virus | |
| CN113241135A (en) | Disease risk prediction method and system based on multi-mode fusion | |
| Ren et al. | Interpretable pneumonia detection by combining deep learning and explainable models with multisource data | |
| Abdullayeva | Internet of Things‐based healthcare system on patient demographic data in Health 4.0 | |
| Khanna et al. | Radiologist-level two novel and robust automated computer-aided prediction models for early detection of COVID-19 infection from chest X-ray images | |
| Shoaib et al. | Hybrid classification structures for automatic COVID-19 detection | |
| Shirazi et al. | Deep learning in the healthcare industry: theory and applications | |
| Zhang et al. | Dynamic uncertain causality graph for computer-aided general clinical diagnoses with nasal obstruction as an illustration | |
| CN114191665A (en) | Method and device for classifying man-machine asynchronous phenomena in mechanical ventilation process | |
| CN112951412A (en) | Auxiliary diagnosis method and application thereof | |
| Kavuran et al. | COVID-19 and human development: An approach for classification of HDI with deep CNN | |
| Chiwariro et al. | Comparative analysis of deep learning convolutional neural networks based on transfer learning for pneumonia detection | |
| Ilyas et al. | Symptom analysis using fuzzy logic for detection and monitoring of covid-19 patients | |
| Saha et al. | LM-DNN: pre-trained DNN with LSTM and cross fold validation for detecting viral pneumonia from chest CT | |
| Han et al. | Mortality prediction for COVID-19 patients via Broad Learning System | |
| Radiuk et al. | Human-in-the-loop approach based on MRI and ECG for healthcare diagnosis | |
| He et al. | CovidNet: An automatic architecture for Covid-19 detection with deep learning from chest X-ray images | |
| Yazdani et al. | Supervised machine learning approach to COVID-19 detection based on clinical data | |
| CN114429822A (en) | Medical record quality inspection method and device and storage medium | |
| Nawaz et al. | Ensemble of Autoencoders for Anomaly Detection in Biomedical Data: A Narrative Review | |
| Devi et al. | Diagnosing the severity of covid-19 in lungs using CNN models | |
| Zillner | Reasoning-based patient classification for enhanced medical image annotation | |
| Gao et al. | Application of Machine Learning in Intelligent Medical Image Diagnosis and Construction of Intelligent Service Process | |
| Cheng et al. | DPN-SENet: A self-attention mechanism neural network for detection and diagnosis of COVID-19 from chest x-ray images |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210611 |