CN114998203A - System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence - Google Patents

System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence Download PDF

Info

Publication number
CN114998203A
CN114998203A CN202210450902.XA CN202210450902A CN114998203A CN 114998203 A CN114998203 A CN 114998203A CN 202210450902 A CN202210450902 A CN 202210450902A CN 114998203 A CN114998203 A CN 114998203A
Authority
CN
China
Prior art keywords
pneumoconiosis
occupational
diagnosis
stage
artificial intelligence
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
Application number
CN202210450902.XA
Other languages
Chinese (zh)
Inventor
沈江
骆春波
伍东升
孙文建
罗杨
吴霜
张艳
刘露
张洪静
王成龙
郑后军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NO4 WEST CHINA TEACHING HOSPITAL SICHUAN UNIVERSITY
Original Assignee
NO4 WEST CHINA TEACHING HOSPITAL SICHUAN UNIVERSITY
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by NO4 WEST CHINA TEACHING HOSPITAL SICHUAN UNIVERSITY filed Critical NO4 WEST CHINA TEACHING HOSPITAL SICHUAN UNIVERSITY
Priority to CN202210450902.XA priority Critical patent/CN114998203A/en
Publication of CN114998203A publication Critical patent/CN114998203A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • 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
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • 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
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30061—Lung

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Biomedical Technology (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Quality & Reliability (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses an accurate diagnosis system and method for occupational pneumoconiosis based on artificial intelligence, wherein X-ray high kilovolt and digital photography (DR) back front chest films with excellent technical quality are collected in a standardized manner, are subjected to zoning and stage marking, are used for determining stage results of occupational pneumoconiosis, and are analyzed to establish incidence relations between focus image characteristics and pathological bases and anatomy of the diseases; the deep learning model and the technical scheme are used for X-ray high kilovolt and DR posterior anterior chest radiograph image segmentation, label distribution construction and stage prediction, and the pneumoconiosis in the occupational pneumoconiosis image, the first stage pneumoconiosis, the second stage pneumoconiosis and the third stage pneumoconiosis are graded; taking the proposed deep learning diagnosis model as an auxiliary tool for clinical diagnosis, and carrying out comparison verification and optimization on the result; and establishing a diagnosis system to guide the prevention and diagnosis of the occupational pneumoconiosis and the assessment of disability and disability grades of occupational diseases. The invention realizes accurate diagnosis and early intervention of occupational pneumoconiosis.

Description

System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence
Technical Field
The invention belongs to the technical field of artificial intelligence, and particularly relates to an accurate diagnosis system and method for occupational pneumoconiosis based on artificial intelligence.
Background
Occupational pneumoconiosis is a collective term for a group of occupational lung diseases, mainly diffuse fibrosis of lung tissues, caused by long-term inhalation of productive dust of different pathogenicity and retention in the lungs during professional activities. According to the classification and catalogue of occupational diseases of our country, the medicine mainly comprises twelve types of silicosis, coal dust lung, graphite dust lung, carbon black dust lung, asbestos lung, talcum dust lung, cement dust lung, mica dust lung, ceramic dust lung, aluminum dust lung, electric welding dust lung and casting dust lung. At present, the diagnosis of occupational pneumoconiosis in China is based on the standard of "diagnosis of occupational pneumoconiosis" (GBZ 70). The diagnosis principle of the occupational pneumoconiosis is based on reliable productive mineral dust contact history, and based on the chest film performance after high kilovolt X-ray or digital X-ray photography (DR), the diagnosis principle can be used for diagnosing by combining the occupational health study of workplaces, epidemiological investigation data of pneumoconiosis and occupational health monitoring data, referring to clinical performance and laboratory examination, and after other similar lung diseases are eliminated, contrasting the standard diagnosis film of the occupational pneumoconiosis. According to the diagnostic criteria, the diagnostician divides the diagnosis of pneumoconiosis into first, second and third stages according to the overall density of small X-ray chest radiographs, the lung region with small shadows, the presence or absence of small shadow aggregation, large shadows and pleural macula.
According to the national occupational disease report issued by the national health and health committee, the average proportion of occupational pneumoconiosis accounts for 85.4% in the 10-year-2007 and 2016, and the average annual newly reported case growth rate reaches 11.6%. The data show that the occupational pneumoconiosis is the most important occupational disease in the occupational activities of China, and the potential population suffering from the occupational pneumoconiosis is huge, wherein the number of the occupational pneumoconiosis patients in Sichuan province is higher than that in the third country. Pulmonary fibrosis caused by occupational pneumoconiosis is irreversible and has many complications, and the existing treatment means for occupational pneumoconiosis is limited, so that most patients suffering from pneumoconiosis gradually lose labor capacity, and the conditions of health and life quality of the patients are seriously reduced due to pathogenic poverty and disease return poverty. The diagnosis and recovery costs of pneumoconiosis impose a heavy economic burden on patients and society. According to measurement and calculation, the pneumoconiosis causes the loss of 1845 million yuan in China, including the direct economic loss of 250 million yuan and the indirect economic loss of 1595 million yuan, wherein the indirect economic loss is 6 times of the direct economic loss. The average life social productivity loss of pneumoconiosis patients was calculated to be 60.85 ten thousand yuan by Disability Adjusted Life Year (DALY) method. Therefore, early detection and timely intervention are critical to the prevention and treatment of occupational pneumoconiosis.
As legal occupational disease, the treatment and arrangement of patients with occupational pneumoconiosis are mainly based on the assessed disability grade, and the disability grade assessment depends on stages of occupational pneumoconiosis and the degree of lung function injury. Therefore, accurate diagnosis of the occupational pneumoconiosis involves important actions related to civilian welfare and welfare, such as national rescue, implementation of industrial injury insurance, and the like. Because the diagnosis levels of the occupational disease diagnosticians in different regions and different institutions are different and are easily interfered by various factors, the homogeneity of the diagnosis staging conclusion of the occupational pneumoconiosis is poor, and various social problems are caused.
However, in the diagnosis process of the occupational pneumoconiosis, since the cognition of different occupational disease diagnosticians on the occupational pneumoconiosis microstructure is greatly different and is easily interfered by external factors, the diagnosis conclusion of the occupational pneumoconiosis cannot be objectively reflected, and the real change of the lung damage of a patient can be accurately evaluated. Accurate diagnosis of occupational pneumoconiosis is realized, and the following key technical problems need to be solved:
(1) according to the current diagnosis standard of occupational pneumoconiosis, the doctor who obtains the quality of occupational disease diagnosis compares the chest film of the patient to be applied for the occupational pneumoconiosis diagnosis with the standard film, and after collective discussion, the diagnosis opinion is formed. In the process, the professional diagnostician is subjective in identifying and comparing the images, the images are easily interfered by external factors, and the identification result is poor in stability and repeatability.
(2) Because of different cognition and difference of working experience of occupational pneumoconiosis image changes, diagnosis levels of occupational disease diagnosis doctors in different regions and different diagnosis institutions are different, the homogeneity of diagnosis conclusions is poor, and even misdiagnosis and missed diagnosis occur.
(3) Due to the overlapping of the front and rear structures, the X-ray chest radiograph is not good enough in displaying fine lesions such as irregular small shadows and very small circular shadows, early pulmonary fibrosis lesions and complications such as pulmonary emphysema and pulmonary hypertension are not easy to find timely and accurately, and the requirement of early occupational pneumoconiosis diagnosis on an image auxiliary identification technology is urgent.
(4) The gray scale change of the image which can be distinguished by human eyes is limited, image information which is represented by numbers on the chest image cannot be captured by naked eyes, and the potential value of image data needs to be further mined.
(5) The current diagnostic standard for the occupational pneumoconiosis classifies the concentration into four grades and twelve grades, the method becomes a main basis for reflecting the severity of the occupational pneumoconiosis, but the method mainly depends on the subjective judgment of a diagnostician, and the classification accuracy and the precision are not enough. With the development of medicine and the deep research on occupational pneumoconiosis, a more objective, accurate and comprehensive quantitative method is needed for evaluation and analysis.
Artificial Intelligence (AI) is a new technical science for the research and development of intelligent theories, methods, techniques and applications for simulating, extending and expanding people. Machine learning is one of the technical fields of artificial intelligence, and mainly researches how to improve the performance of a specific algorithm based on data driving in empirical learning. Deep Learning (Deep Learning) is a branch of machine Learning, an algorithm that attempts to perform high-level abstraction on data using multiple processing layers that contain complex structures or consist of multiple nonlinear transformations. In recent years, several deep learning frameworks such as multilayer neural networks, convolutional neural networks, deep signaling networks and recurrent neural networks have been applied to the fields of computer vision, speech recognition, audio recognition and bioinformatics, and have achieved good results.
In the field of medical imaging, machine learning is used for image segmentation, image registration, content-based image retrieval, and text analysis of diagnostic reports, and in particular, research on computer-aided detection and diagnosis is more. In the aspect of lung diseases, a machine learning method is applied to identify and classify lung nodules, the benign and malignant properties of the lung nodules are predicted, the accuracy of classification of lung cancer by different machine learning methods is compared, and the classification of pulmonary interstitial fibrosis disease types is carried out. An AI disease image diagnosis system based on deep learning is constructed at present, when distinguishing the pneumonia and the health state, the Area under the characteristic curve (AUC) of a subject with detection accuracy reaches 96.8%, the AUC for distinguishing the bacterial pneumonia and the viral pneumonia reaches 94%, and related results are published in Cell, top-level journal.
In China, the diagnosis and research based on artificial intelligence for the occupational pneumoconiosis is still in the beginning stage. The existing research technique team identifies diffuse focus of chest film of occupational pneumoconiosis based on lung field segmentation and texture feature extraction method of active deformation model; in the prior art, methods such as a support vector machine and a decision tree are adopted to automatically identify small shadows, assist in diagnosing pneumoconiosis and conduct stage research on Digital Radiography (DR) chest radiographs based on wavelet texture features and gray level enhancement technology, and good detection efficiency is displayed. However, all of the above methods are based on manual feature extraction, and have the problems of limited feature extraction, poor generalization performance, insufficient neural network layer number, and the like. At present, a research report for constructing an accurate diagnosis model of occupational pneumoconiosis based on a deep learning model is not searched.
The difficulty and significance for solving the technical problems are as follows: the invention establishes a scientific, accurate and efficient accurate diagnosis system for the occupational pneumoconiosis for the first time, and makes fundamental and original contributions.
The accurate, stable and reliable research on the diagnosis technology of the occupational pneumoconiosis has the following important meanings: (1) the diagnosis technology of the occupational pneumoconiosis can reduce the social contradiction and the sanitary cost expenditure due to the huge crowd contacting with the productive dust, and is accurate, stable and reliable. (2) The influence and the interference of the professional pneumoconiosis diagnosis caused by the business ability of the occupational disease diagnostician and social factors are eliminated to the maximum extent, the legal rights and interests of the occupational pneumoconiosis patient and the enterprise are protected, and the aims of reducing labor disputes, avoiding the initiation of group events and maintaining social fairness and justice are fulfilled.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an accurate diagnosis system and method for occupational pneumoconiosis based on artificial intelligence.
The invention is realized in such a way, and provides an artificial intelligence-based accurate diagnosis system and method for occupational pneumoconiosis, which specifically comprise the following steps:
the method comprises the following steps: standardizing X-ray high kilovolt and digital photography (DR) back front chest radiography with excellent quality, carrying out zoning and stage marking on the chest radiography, determining the stage result of occupational pneumoconiosis, analyzing and establishing the incidence relation between focus image characteristics and the pathological basis and anatomy of the disease;
step two: the deep learning model and the technical scheme are used for X-ray high kilovolt and DR posterior anterior chest radiograph image segmentation, label distribution construction and stage prediction, and the pneumoconiosis in the occupational pneumoconiosis image, the first stage pneumoconiosis, the second stage pneumoconiosis and the third stage pneumoconiosis are graded;
step three: taking the proposed deep learning diagnosis model as an auxiliary tool for clinical diagnosis, and carrying out comparison verification and optimization on the result; and establishing a diagnosis system to guide the prevention and diagnosis of the occupational pneumoconiosis and the assessment of disability and disability grades of occupational diseases.
Further, in the second step, diffuse distribution and lesions with small target lesions in the occupational pneumoconiosis image are identified and classified, and image data marking, diffuse lesion segmentation detection, label distribution construction and an occupational pneumoconiosis identification staging method are specifically adopted.
Further, in the first step, the image data marking specifically adopts the following steps:
dividing lung fields of a posterior and anterior chest image of a pneumoconiosis group according to a method in an occupational pneumoconiosis diagnosis standard, contrasting with a standard sheet, and respectively grading and marking the shadow form and the density of each region;
according to the chest film quality and quality assessment standard C in the appendix of the national occupational health standards GBZ 70-2015 for diagnosis of the occupational pneumoconiosis, performing quality control on all collected images, eliminating the difference and the waste film, and ensuring high-quality image data;
according to the national diagnosis standard of occupational pneumoconiosis, the lung area of the posterior chest film is divided into 3 lung areas on the left side and the right side, and the total number of the lung areas is 6;
according to the national diagnosis standard sheet for occupational pneumoconiosis, the pneumoconiosis focus of the included cases is marked, and the focus form (regular small shadow p, q, r; irregular small shadow s, t, u; large shadow), the concentration (0 grade, 1 grade, 2 grade and 3 grade) and the overall concentration of each lung area are determined.
Image data marking is completed by 5 attending physicians who are engaged in breast image diagnosis for more than 5 years. All the quasi-marked image data are randomly selected, so that the selection bias is avoided; negotiating to resolve when the meanings are inconsistent in the marking process, or feeding back to a senior qualified physician for systematic analysis and determining a conclusion;
thirdly, 3 professional qualified physicians for diagnosing the national occupational pneumoconiosis disease adopt a few majority-compliant principles to evaluate the accuracy of each group of data markers and determine the stage of the occupational pneumoconiosis disease.
Further, in the second step, a lung field image in the chest piece is extracted through an image segmentation algorithm. The image segmentation algorithm can be selected from traditional image segmentation algorithms such as level set and graph cut at will, and can also be obtained by training U-Net by adopting a public chest radiograph lung field segmentation data set (Montgomery, JSRT).
Further, in the second step, label distribution corresponding to the period of the posterior-anterior chest radiograph is constructed, specifically: the label distribution construction method based on the lognormal distribution is characterized in that label distribution constructed by the method and one-hot labels are integrated, and a convolutional neural network is trained.
Further, in the second step, the stage of occupational pneumoconiosis is predicted, and the lung field image is input into the trunk network to predict the stage of pneumoconiosis to which the chest radiograph belongs. And (2) reducing the size of the lung field image obtained in the step (1) to 512 x 512 by a down-sampling algorithm, inputting the lung field image into a backbone network, wherein the number of output nodes of the last full-connection layer of the backbone network is 4, calculating loss, performing back propagation, optimizing the backbone network in a backbone network training stage, and directly outputting a backbone network prediction result in an inference stage.
Another object of the present invention is to provide an artificial intelligence-based precision diagnosis system for an occupational pneumoconiosis disease, which implements the artificial intelligence-based precision diagnosis method for an occupational pneumoconiosis disease, the artificial intelligence-based precision diagnosis system comprising: the system comprises a server, a client and a diagnosis system;
the diagnosis system comprises an information input module, a machine learning intelligent diagnosis module and a result output module.
The invention also aims to provide an information data processing terminal of the occupational pneumoconiosis accurate diagnosis system for realizing the artificial intelligence.
Another object of the present invention is to provide a mobile device for implementing the artificial intelligence accurate diagnosis system for occupational pneumoconiosis.
It is another object of the present invention to provide a computer-readable storage medium comprising instructions which, when executed on a computer, cause the computer to execute the artificial intelligent precision diagnosis system for occupational pneumoconiosis.
The diagnosis expert system comprises an AI server, a client and system software; the AI server is used for storing system data, automatically performing convolutional neural network training, automatically verifying the input case data at the background and feeding back the case data to the client; the client is a Web browser and is used for uploading historical cases and input cases to the server; the system software comprises an information input module and an automatic diagnosis module; the information input module is used for importing the information of the detector, diagnosis conclusions and the like; the automatic diagnosis module is used for selecting a deep learning model according to the uploaded training set data and test set data and the background according to modeling and carrying out automatic verification to generate a classified prediction result of the occupational pneumoconiosis and feed back diagnosis opinions.
The system adopts a B/S architecture and specifically comprises a presentation layer, a service logic layer and a data access layer; the presentation layer is used for inputting or feeding back an interface of diagnosis opinions; the business logic layer is used for carrying out deep learning on the uploaded data according to a set AI model, verifying the data requiring automatic diagnosis and feeding back a verification result to the presentation layer; the data access layer is used for directly operating the database on the done affairs.
The client is a Web browser and is used for inputting the historical case and the new case by adopting the Web browser, so that the cases with more resources can be obtained.
In summary, the advantages and positive effects of the invention are:
1. the front end marking data is reliable and accurate, and the marking is finished by a known expert who obtains the national diagnosis qualification of occupational diseases.
2. A label distribution construction method more conforming to the characteristics of stage ambiguity and asymmetric noise in a pneumoconiosis data set is provided.
3. The convolution neural network overfitting problem caused by asymmetric stage fuzzy and asymmetric noise in the pneumoconiosis data set can be well solved.
According to the method, the occupational pneumoconiosis image is segmented and accurately diagnosed based on the automatic feature extraction and the feature learning capability of the deep learning model, and finally, a clinically feasible occupational pneumoconiosis diagnosis model is constructed to guide the prevention and treatment of the occupational pneumoconiosis and the assessment of the occupational disease disability level. The method is used for deeply exploring the connotation of accurate medicine and transformation medicine, takes occupational pneumoconiosis with important clinical value and social significance as a starting point, and closely combines the basis and the clinic to realize accurate diagnosis and early intervention of the occupational pneumoconiosis.
The diagnosis method and the diagnosis technology provided by the invention are based on medical big data and an efficient artificial intelligence method, have the advantages of objectivity, high efficiency and homogeneity, can effectively assist various professional disease diagnosticians to accurately diagnose the occupational pneumoconiosis, are beneficial to popularization and have great significance. The achievement of the invention lays a solid foundation for the establishment of the intelligent diagnosis standard of the occupational pneumoconiosis, and the constructed large sample database provides high-quality open source data for the future work of the occupational pneumoconiosis. Actively responds to the national artificial intelligence development strategy, promotes the clinical application and the production process of the medical image auxiliary diagnosis system, and makes a contribution to the establishment of an intelligent medical system serving the aspect of preventing and treating occupational diseases.
The invention (1) analyzes and establishes the incidence relation between the focus image characteristics and the pathological basis and anatomy of the disease, and establishes the key image characteristic set of occupational pneumoconiosis imaging, in particular to the key image characteristic set of pneumoconiosis imaging caused by early pneumoconiosis and dust with low silicon dioxide content. (2) Aiming at a plurality of key problems existing in the diagnosis of the occupational pneumoconiosis, an accurate, stable and reliable diagnosis scheme of the occupational pneumoconiosis is constructed based on the strong capability of deep learning of feature extraction and feature learning. (3) And verifying the performance of the technical scheme and providing an algorithm optimization method by combining clinical practice. The label distribution construction method more suitable for the characteristic point that stage ambiguity and noise in the pneumoconiosis data set are asymmetric is provided, and the problem of convolution neural network overfitting caused by asymmetric stage ambiguity and asymmetric noise in the pneumoconiosis data set can be well solved. (4) Based on the formulation of prevention and treatment strategies for guiding occupational pneumoconiosis and the assessment of disability grades, a solid foundation is laid for the subsequent establishment of an intelligent medical system serving the prevention and treatment aspects of occupational diseases.
The invention establishes a scientific, accurate and efficient diagnosis system for the occupational pneumoconiosis for the first time, and makes fundamental and original contributions.
Drawings
FIG. 1 is a flowchart of a system and method for diagnosing occupational pneumoconiosis based on artificial intelligence according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a diagnostic system according to an embodiment of the present invention.
Fig. 3 is a flowchart of a pneumoconiosis chest radiograph staging method based on lognormal label distribution learning according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of an implementation process provided by an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
According to the method, the occupational pneumoconiosis image is segmented and accurately diagnosed based on the automatic feature extraction and the feature learning capability of the deep learning model, and finally, a clinically feasible occupational pneumoconiosis diagnosis model is constructed to guide the prevention and treatment of the occupational pneumoconiosis and the assessment of the occupational disease disability level.
The application principle of the present invention is further explained in detail with reference to the accompanying drawings;
as shown in fig. 1, the system and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence provided in the embodiments of the present invention specifically include the following steps:
s101: standardizing X-ray high kilovolt and digital photography (DR) back front chest radiographs with excellent quality, carrying out zoning and stage marking on the chest radiographs, determining stage results of occupational pneumoconiosis, and analyzing and establishing an association relation between the focus image characteristics and the pathological basis and anatomy of the disease;
s102: the deep learning model and the technical scheme are used for X-ray high kilovolt and DR posterior anterior chest radiograph image segmentation, label distribution construction and stage prediction, and the pneumoconiosis in the occupational pneumoconiosis image, the first stage pneumoconiosis, the second stage pneumoconiosis and the third stage pneumoconiosis are graded;
s103: taking the proposed deep learning diagnosis model as an auxiliary tool for clinical diagnosis, and carrying out comparison verification and optimization on the result; and (3) establishing a diagnosis system to guide prevention and treatment of the occupational pneumoconiosis and disability grade assessment of occupational diseases.
In step S101, the image data marking provided by the embodiment of the present invention specifically includes the following steps:
dividing lung fields of a posterior and anterior chest image of a pneumoconiosis group according to a method in an occupational pneumoconiosis diagnosis standard, contrasting with a standard sheet, and respectively grading and marking the shadow form and the density of each region;
image data marking is completed by 5 attending physicians who are engaged in breast image diagnosis for more than 5 years. All the quasi-marked image data are randomly selected, so that the selection bias is avoided; negotiating to resolve when the meanings are inconsistent in the marking process, or feeding back to a senior qualified physician for systematic analysis and determining a conclusion;
thirdly, 3 professional qualified physicians for diagnosing the national occupational pneumoconiosis disease adopt a few majority-compliant principles to evaluate the accuracy of each group of data markers and determine the stage of the occupational pneumoconiosis disease.
In step S102, in the segmentation detection of the diffuse lesion provided in the embodiment of the present invention, the lung field image in the chest radiograph is extracted through an image segmentation algorithm. The image segmentation algorithm can be selected from traditional image segmentation algorithms such as level set and graph cut, and can also be obtained by training U-Net through a public chest piece lung field segmentation data set (Montgomery, JSRT), which is not limited by the invention.
In step S102, a label distribution corresponding to the period to which the posterior chest radiograph belongs is constructed, specifically: the label distribution construction method based on the logarithmic normal distribution is characterized in that label distribution constructed by the method and one-hot labels are integrated, a convolutional neural network is trained, and the problem of overfitting of the convolutional neural network caused by asymmetric stage ambiguity and asymmetric noise in a pneumoconiosis data set is solved.
In step S102, in the occupational pneumoconiosis period prediction provided in the embodiment of the present invention, the lung field image is input to the trunk network, and the pneumoconiosis period to which the chest radiograph belongs is predicted. The lung field image size obtained in step 1 is reduced to 512 × 512 by a down-sampling algorithm, and then the lung field image is input into a backbone network (the selection of the backbone network is not limited by the present invention), and the number of output nodes of the last full connection layer of the backbone network is 4. And in the training stage of the backbone network, calculating loss, performing back propagation and optimizing the backbone network. And in the inference stage, directly outputting a prediction result of the backbone network.
The technical scheme of the invention is described in detail by combining experiments.
The basic idea of the invention is as follows: extracting a lung field image in the chest picture through an image segmentation algorithm; constructing label distribution of each stage of pneumoconiosis based on the characteristic that lognormal distribution is asymmetric; inputting the lung field image into a backbone network for prediction; in the stage of training a backbone network, training a convolutional neural network by combining label distribution and an original one-hot label; and in the inference stage, directly outputting a prediction result of the backbone network. The method can overcome the problem of convolution neural network overfitting caused by asymmetric stage ambiguity and asymmetric noise in the pneumoconiosis data set. The flowchart of the pneumoconiosis chest radiograph staging method based on lognormal label distribution learning provided by the invention is shown in fig. 3, and specifically comprises the following steps:
step 1: and extracting a lung field image from the chest picture through an image segmentation algorithm. The image segmentation algorithm can be selected from traditional image segmentation algorithms such as level set and graph cut, and can also be obtained by training U-Net by adopting a Montgomery JSRT (JSRT) data set, which is not limited by the invention.
Step 2: and constructing label distribution corresponding to the period of the chest radiograph based on the lognormal distribution. With S ═ S i |0≤i≤4，1≤s i ≦ 5} for the chest film tag set, s in the set 0 1 stands for healthy (without pulmonary nodules) chest film, s 1 ＝2、s 2 ＝3、s 3 ＝4、s 4 The term "5" stands for "without pneumoconiosis", "first stage", "second stage" and "third stage", respectively, four other pneumoconiosis chest tablets. Suppose the pneumoconiosis stage of the chest radiograph is s i Their corresponding label distribution
Figure BDA0003618582600000111
The value of (a) is calculated by the following formula:
Figure BDA0003618582600000112
wherein δ is 0.3. The construction of an alternate label distribution for pneumoconiosis requires three conditions to be met: 1)
Figure BDA0003618582600000113
is a probability distribution, so y j ∈[0，1](ii) a 2) Vector quantity
Figure BDA0003618582600000114
The sum of all probabilities in (1) is because the pneumoconiosis stage ambiguities and noise only occur between adjacent stages
Figure BDA0003618582600000115
3) In that
Figure BDA0003618582600000116
In all positions of (a), (b) y j＝i The probability value of (a) is maximum. To satisfy the above three conditions, it is necessary to set the average value μ ═ log(s) i ) Then, intercepting the values of the label distribution and carrying out L1 regularization treatment:
Figure BDA00036185826000001112
Figure BDA0003618582600000117
and step 3: and inputting the lung field image into a backbone network, and predicting the pneumoconiosis period of the chest film. The lung field image size obtained in step 1 is reduced to 512 × 512 by a down-sampling calculation method, and then the lung field image is input into a backbone network (the selection of the backbone network is not limited by the invention), and the number of output nodes of the last full-connection layer of the backbone network is 5.
And 4, step 4: and in the training stage of the backbone network, calculating loss, performing back propagation and optimizing the backbone network. Suppose the prediction result output by the backbone network is
Figure BDA0003618582600000118
Computing
Figure BDA0003618582600000119
KL divergence loss L from tag distribution y KL Calculating
Figure BDA00036185826000001110
And one-hot tag
Figure BDA00036185826000001111
Cross entropy loss between L CE . And weighting and summing the two loss values to obtain the total loss:
L＝αL KL +βL CE
the values of alpha and beta can be determined according to actual conditions, and alpha L can be ensured KL And beta L CE In the same order of magnitude. And (4) reversely transmitting the total loss L and optimizing the backbone network.
And 5: and in the inference stage, directly outputting a prediction result of the backbone network. Obtaining the prediction result output by the backbone network
Figure BDA0003618582600000121
In which the maximum value is locatedPosition indexing
Figure BDA0003618582600000122
The index value y is the pneumoconiosis stage to which the input chest radiograph belongs.
Compared with the prior art, the invention has the beneficial effects that: (1) a label distribution construction method more conforming to the characteristics of stage ambiguity and asymmetric noise in a pneumoconiosis data set is provided. (2) The problem of over-fitting of a convolutional neural network caused by asymmetric stage ambiguity and asymmetric noise in a pneumoconiosis data set can be well solved.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When used in whole or in part, is implemented in a computer program product that includes one or more computer instructions. The procedures or functions described in accordance with the embodiments of the invention may be carried out in whole or in part by loading or executing the computer program instructions on a computer. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (10)

1. An artificial intelligence-based precision diagnosis method for occupational pneumoconiosis, which is characterized by comprising the following steps:
the method comprises the following steps: standardizing X-ray high kilovolt and digital photography (DR) back front chest radiography with excellent quality, carrying out zoning and stage marking on the chest radiography, determining the stage result of occupational pneumoconiosis, analyzing and establishing the incidence relation between focus image characteristics and the pathological basis and anatomy of the disease;
step two: the deep learning model and the technical scheme are used for X-ray high kilovoltage and DR back and front chest radiography image segmentation, label distribution construction and stage prediction, and are used for grading pneumoconiosis in the occupational pneumoconiosis image, pneumoconiosis in the first stage, second stage and third stage;
step three: taking the proposed deep learning diagnosis model as an auxiliary tool for clinical diagnosis, and carrying out comparison verification and optimization on the result; and establishing a diagnosis system to guide the prevention and diagnosis of the occupational pneumoconiosis and the assessment of disability and disability grades of occupational diseases.
2. The method for accurately diagnosing occupational pneumoconiosis according to claim 1, wherein in the second step, diffuse distribution and small-focus lesions appearing in images of occupational pneumoconiosis are identified and classified, and image data marking, diffuse lesion segmentation detection, label distribution construction and an occupational pneumoconiosis identification staging method are specifically adopted.
3. The method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence of claim 1, wherein in the first step, the image data is labeled by the following steps:
dividing lung fields into regions according to a method in an occupational pneumoconiosis diagnosis standard for X-ray high kilovolt and DR posterior anterior chest radiograph images, and grading and marking the shadow form and the density of each region by contrasting standard sheets;
the marking of the image data is completed by 5 attending physicians engaged in the breast image diagnosis for more than 5 years; all the quasi-marked image data are randomly selected, so that the data selection bias is avoided; negotiating to resolve when the meanings are inconsistent in the marking process, or feeding back to a senior qualified physician for systematic analysis and determining a conclusion;
thirdly, 3 known diagnostic specialists for the occupational pneumoconiosis who acquire the qualification of the national diagnostic physicians for the occupational pneumoconiosis evaluate the accuracy of each group of data marks by adopting a few majority-obeying principles and determine the stage of the occupational pneumoconiosis.
4. The method for accurately diagnosing the occupational pneumoconiosis based on the artificial intelligence of claim 1, wherein in the second step, the lung field image in the chest piece is extracted through an image segmentation algorithm; the image segmentation algorithm can be selected from traditional image segmentation algorithms such as level set and graph cut at will, and can also be obtained by training U-Net by adopting a public chest piece lung field segmentation data set (Montgomery, JSRT).
5. The method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence of claim 1, wherein in the second step, label distribution corresponding to the period to which the posterior anterior chest radiograph belongs is constructed, specifically: the label distribution construction method based on the lognormal distribution is characterized in that label distribution constructed by the method and one-hot labels are integrated, and a convolutional neural network is trained.
6. The artificial intelligence-based accurate diagnosis method for the occupational pneumoconiosis according to claim 1, wherein in the second step, the occupational pneumoconiosis period is predicted, the lung field images are input into a backbone network, the pneumoconiosis period to which the chest slices belong is predicted, the lung field images obtained in the step 1 are reduced to 512 x 512 through a down-sampling algorithm, then the lung field images are input into the backbone network, the number of output nodes of the last full connection layer of the backbone network is 4, in the backbone network training stage, loss calculation, back propagation and backbone network optimization are performed, and in the reasoning stage, the prediction result of the backbone network is directly output.
7. An artificial intelligence-based precision diagnosis system for occupational pneumoconiosis that implements the artificial intelligence-based precision diagnosis method of claim 1, wherein the artificial intelligence-based precision diagnosis system for occupational pneumoconiosis comprises: the system comprises a server, a client and a diagnosis system;
the diagnosis system comprises an information input module, a machine learning intelligent diagnosis module and a result output module.
8. An information data processing terminal for implementing the artificial intelligence accurate diagnosis system for occupational pneumoconiosis according to claim 7.
9. A mobile device implementing the artificial intelligence accurate diagnosis system for occupational pneumoconiosis according to claim 7.
10. A computer-readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the artificial intelligence accurate diagnosis system for occupational pneumoconiosis as claimed in any one of claims 7.
CN202210450902.XA 2022-04-27 2022-04-27 System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence Pending CN114998203A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210450902.XA CN114998203A (en) 2022-04-27 2022-04-27 System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210450902.XA CN114998203A (en) 2022-04-27 2022-04-27 System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence

Publications (1)

Publication Number Publication Date
CN114998203A true CN114998203A (en) 2022-09-02

Family

ID=83025253

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210450902.XA Pending CN114998203A (en) 2022-04-27 2022-04-27 System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence

Country Status (1)

Country Link
CN (1) CN114998203A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117315378A (en) * 2023-11-29 2023-12-29 北京大学第三医院（北京大学第三临床医学院） Grading judgment method for pneumoconiosis and related equipment

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246888A (en) * 2012-02-03 2013-08-14 通用电气公司 System and method for diagnosing lung disease by computer
CN110680326A (en) * 2019-10-11 2020-01-14 北京大学第三医院（北京大学第三临床医学院） Pneumoconiosis identification and grading judgment method based on deep convolutional neural network
CN111816314A (en) * 2020-07-01 2020-10-23 深圳市职业病防治院 Method for selecting, marking and verifying pneumoconiosis chest radiograph through artificial intelligence screening
WO2021173826A1 (en) * 2020-02-25 2021-09-02 The Board Of Regents Of The University Of Texas System Systems and methods for screening and staging of pneumoconiosis
CN114098779A (en) * 2021-11-08 2022-03-01 安徽医学高等专科学校 Intelligent pneumoconiosis grade judging method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246888A (en) * 2012-02-03 2013-08-14 通用电气公司 System and method for diagnosing lung disease by computer
CN110680326A (en) * 2019-10-11 2020-01-14 北京大学第三医院（北京大学第三临床医学院） Pneumoconiosis identification and grading judgment method based on deep convolutional neural network
WO2021173826A1 (en) * 2020-02-25 2021-09-02 The Board Of Regents Of The University Of Texas System Systems and methods for screening and staging of pneumoconiosis
CN111816314A (en) * 2020-07-01 2020-10-23 深圳市职业病防治院 Method for selecting, marking and verifying pneumoconiosis chest radiograph through artificial intelligence screening
CN114098779A (en) * 2021-11-08 2022-03-01 安徽医学高等专科学校 Intelligent pneumoconiosis grade judging method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117315378A (en) * 2023-11-29 2023-12-29 北京大学第三医院（北京大学第三临床医学院） Grading judgment method for pneumoconiosis and related equipment
CN117315378B (en) * 2023-11-29 2024-03-12 北京大学第三医院（北京大学第三临床医学院） Grading judgment method for pneumoconiosis and related equipment

Similar Documents

Publication Publication Date Title
CN106372390B (en) A kind of self-service healthy cloud service system of prevention lung cancer based on depth convolutional neural networks
US8682693B2 (en) Patient data mining for lung cancer screening
US7711404B2 (en) Patient data mining for lung cancer screening
Yan et al. A fully automatic artificial intelligence–based CT image analysis system for accurate detection, diagnosis, and quantitative severity evaluation of pulmonary tuberculosis
CN112802000A (en) Intelligent auxiliary diagnosis and treatment system for multi-modal medical images
Lv et al. Development and validation of a clinically applicable deep learning strategy (HONORS) for pulmonary nodule classification at CT: a retrospective multicentre study
CN114582496A (en) Common gynecological disease prediction model construction method and prediction system
Shamrat et al. Analysing most efficient deep learning model to detect COVID-19 from computer tomography images
Han et al. Development and validation of a 3D-convolutional neural network model based on chest CT for differentiating active pulmonary tuberculosis from community-acquired pneumonia
Zhang et al. Artificial intelligence clinicians can use chest computed tomography technology to automatically diagnose coronavirus disease 2019 (COVID-19) pneumonia and enhance low-quality images
CN114998203A (en) System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence
Adams et al. Clinical impact and generalizability of a computer-assisted diagnostic tool to risk-stratify lung nodules with CT
Kakani et al. Post-covid chest disease monitoring using self adaptive convolutional neural network
Wang et al. Can peritumoral regions increase the efficiency of machine-learning prediction of pathological invasiveness in lung adenocarcinoma manifesting as ground-glass nodules?
CN103168305A (en) System and method for dynamic growing of a patient database with cases demonstrating special characteristics
CN115602327A (en) Construction method of prediction model for lung nodule lung cancer risk
Jeong et al. The EMory BrEast imaging Dataset (EMBED): a racially diverse, granular dataset of 3.5 M screening and diagnostic mammograms
Yang et al. Automated diagnosis and management of follicular thyroid nodules based on the devised small-dataset interpretable foreground optimization network deep learning: a multicenter diagnostic study
Wang et al. Fully Automated Interpretable Breast Ultrasound Assisted Diagnosis System
Fernandes et al. MIDAS–mammographic image database for automated analysis
CN115620899B (en) Multi-view field based 3D convolutional neural network integrated model pulmonary nodule diagnostic system
Chandrakantha et al. A Survey on Artificial Intelligence-based Lung Tumor Segmentation and Classification
Ahatsham et al. An efficient system for early diagnosis of breast cancer using support vector machine
Holt et al. A nationwide adaptive prediction tool for coronary heart disease prevention.
Gerbolés et al. A bias analysis on a breast cancer mammography dataset for deep learning applications

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