CN114098779A - Intelligent pneumoconiosis grade judging method - Google Patents
Intelligent pneumoconiosis grade judging method Download PDFInfo
- Publication number
- CN114098779A CN114098779A CN202111315770.1A CN202111315770A CN114098779A CN 114098779 A CN114098779 A CN 114098779A CN 202111315770 A CN202111315770 A CN 202111315770A CN 114098779 A CN114098779 A CN 114098779A
- Authority
- CN
- China
- Prior art keywords
- pneumoconiosis
- shadow
- training
- sample
- image
- 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.)
- Withdrawn
Links
- 206010035653 pneumoconiosis Diseases 0.000 title claims abstract description 96
- 238000000034 method Methods 0.000 title claims abstract description 28
- 210000004072 lung Anatomy 0.000 claims abstract description 83
- 238000012549 training Methods 0.000 claims abstract description 78
- 239000000835 fiber Substances 0.000 claims abstract description 34
- 208000005069 pulmonary fibrosis Diseases 0.000 claims abstract description 29
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 16
- 230000006870 function Effects 0.000 claims description 33
- 239000013598 vector Substances 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000002601 radiography Methods 0.000 claims description 7
- 239000002131 composite material Substances 0.000 claims description 6
- 238000012850 discrimination method Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000010191 image analysis Methods 0.000 abstract description 2
- 230000006872 improvement Effects 0.000 description 9
- 238000003745 diagnosis Methods 0.000 description 6
- 230000003902 lesion Effects 0.000 description 5
- 230000011218 segmentation Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 206010016654 Fibrosis Diseases 0.000 description 2
- 208000019693 Lung disease Diseases 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 230000004761 fibrosis Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004083 survival effect Effects 0.000 description 2
- 210000003437 trachea Anatomy 0.000 description 2
- 201000003838 Idiopathic interstitial pneumonia Diseases 0.000 description 1
- 208000029523 Interstitial Lung disease Diseases 0.000 description 1
- 230000009798 acute exacerbation Effects 0.000 description 1
- 210000002376 aorta thoracic Anatomy 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001684 chronic effect Effects 0.000 description 1
- 238000004195 computer-aided diagnosis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003176 fibrotic effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000004199 lung function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000010837 poor prognosis Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000002685 pulmonary effect Effects 0.000 description 1
- 231100000241 scar Toxicity 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
Landscapes
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- Heart & Thoracic Surgery (AREA)
- High Energy & Nuclear Physics (AREA)
- Physics & Mathematics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Optics & Photonics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Radiology & Medical Imaging (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention relates to the technical field of medical image analysis, and discloses an intelligent determination method for pneumoconiosis grade, which comprises the following steps: dividing the image containing a plurality of DR chest film samples into a first training sample and a second training sample. The DR chest sample image is an image of each lung area with different sizes of pneumoconiosis fiber shadow pixels. And inputting a first training sample based on the convolutional neural network model, and training model parameters of the characteristics of the pulmonary fibrosis shadow to generate an intelligent judgment model of pulmonary fibrosis. In the intelligent pneumoconiosis grade judging method, the input image is replaced by the whole DR chest film from the image block for training, and the characteristic learning model is designed to replace the last three full-connection layers by the convolution layer, so that the aim of the design is to realize the end-to-end detection of the input image and the output image, and simultaneously, focus areas of different lung areas can be directly detected, the training time is reduced, and the recognition rate of pulmonary fibrosis is improved.
Description
Technical Field
The invention relates to the technical field of medical image analysis, in particular to an intelligent pneumoconiosis grade judging method.
Background
Pulmonary fibrosis is a common outcome of various lung diseases, mainly manifested as scar of lung tissue, and if the affected range is wide, the lung volume is reduced, the lung function is obviously reduced, and the life quality of patients is seriously affected. In particular idiopathic interstitial pneumonia is the most typical representative, with pathology and/or imaging manifested as a chronic progressive pulmonary disease of common interstitial pneumonia. IPF has unclear etiology and extremely poor prognosis, and the average survival period after diagnosis is only 3-5 years. IPF currently considers that individual survival time of patients varies greatly, some patients stably survive for many years for a long time, some patients progress slowly, and some patients die in a short time due to acute exacerbation. How to make more accurate disease severity assessment and prognosis judgment according to the condition of a patient is not a widely accepted assessment method at present.
The currently mainly used manual assessment method is to select 4 representative slices of the DR lung window: an aortic arch layer, a trachea bifurcation layer, a superior inferior leaflet basal segment trachea bifurcation layer and a right inferior septum top layer; each layer is provided with a left lung field and a right lung field, the whole lung is divided into 8 lung fields, and the 8 lung fields are divided into 100 small pieces; and (3) judging whether the area of the honeycomb change on each small piece is larger than one half of the area of the small piece as a positive or negative standard, and adding the positive pieces to obtain the proportion of the whole lung honeycomb change.
Currently, the clinical identification and evaluation of pulmonary fibrosis focus mainly depend on the visual judgment and rough evaluation of lung dr images by clinicians, and the evaluation accuracy completely depends on the personal experience of the clinicians. Moreover, the lung DR image data is increased explosively, the workload of doctors is greatly increased, and misdiagnosis and missed diagnosis are easily caused in the disease detection process.
With the development of big data computer vision, computer-aided diagnosis technology is also used to help complete the diagnosis of pulmonary fibrosis based on medical images, so as to reduce the workload of doctors. However, the existing method has low recognition rate on pulmonary fibrosis, and the judgment of the pneumoconiosis grade is not accurate.
Disclosure of Invention
In order to solve the technical problems mentioned in the background art, the invention provides an intelligent determination method for pneumoconiosis grade.
The invention is realized by adopting the following technical scheme: an intelligent pneumoconiosis grade judging method comprises the following steps:
s1, dividing the DR chest sample images into a first training sample and a second training sample; the DR chest radiography sample image is each lung area image with different sizes of pneumoconiosis fiber shadow pixels;
s2, inputting a training sample I based on the convolutional neural network model, and training model parameters of the characteristics of the pulmonary fibrosis shadow to generate an intelligent judgment model of pulmonary fibrosis;
s3, inputting a second training sample based on the intelligent lung fibrosis judgment model, and identifying the shadow shape and the corresponding coordinates of the pneumoconiosis fiber shadow pixels in each DR chest image in the second training sample;
s4, analyzing and predicting the pneumoconiosis fiber shadow shape in each DR chest image, determining the shadow density set to which each shadow shape belongs, calculating the overall density set of the shadow shapes, and performing pneumoconiosis grade judgment training by combining the lung area number of the coordinate where pneumoconiosis fiber shadow pixels are distributed so as to train a pneumoconiosis grade judgment model;
and S5, judging the pneumoconiosis grade according to each parameter of the pneumoconiosis judging model.
As a further improvement of the above solution, in step S1, for each DR chest sample image, it is necessary to calibrate the pneumoconiosis fiber shadow pixels of each type included in each lung region in each DR chest sample image in advance, and use the calibrated DR chest sample image as a sample label.
As a further improvement of the above scheme, in step S2, after each DR chest film sample image is sequentially convolved by several convolution layers of the convolutional neural network model, target feature vectors of different mapping scales are respectively generated;
and training the model parameters of the convolutional neural network model through a composite training loss function according to the obtained target feature vectors and the pre-calibrated sample labels in the training sample I.
As a further improvement of the above scheme, the convolutional neural network model comprises 8 convolutional layers connected in sequence, and each convolutional layer uses a 1 × 1 convolutional kernel to perform corresponding feature extraction;
after the DR chest sample image passes through convolutional layers conv4, conv5, conv6 and conv8, a plurality of target feature vectors for comparison with sample tags are generated respectively.
As a further improvement of the above solution, all of the image pixels of the convolutional layer conv1-3 are 512 × 515, and the image pixels of the convolutional layer conv4, the convolutional layer conv5, the convolutional layer conv6, the convolutional layer conv7 and the convolutional layer conv8 are respectively 256 × 256, 128 × 128, 64 × 64, 32 × 32 and 16 × 16 in sequence.
As a further improvement of the above solution, a method for training model parameters of features of pulmonary fibrotic shadows by a composite training loss function comprises the steps of:
where W is the model parameter of the convolutional neural network model, and S ═ S1,S2,...,SNIs the set of training samples, N is the number of convolutional layers, αnIs an error function lnWeight of (1), XiIs a marked sample image block, Yi=(yi,bi) Is a labeled sample label, yiE, e {0, 1., K }, and the real target frame coordinates (x, y, w, h) respectively represent the center point coordinates and the width and length values of the target frame;
wherein m is the number of layers of the convolution layer where the prediction target frame is located during detection, i is the coordinate of the detected initial target frame, j is the coordinate of the real target frame, and the coordinate error function adopts smoothL1() Solving the coordinate errors of the predicted target frame and the real target frame, thenThe calculation method comprises the following steps:
wherein,represents the center, length and width coordinates of the predicted target frame,the coordinates of the center, length and width of the real target frame are shown,representing a coordinate difference value of the predicted target frame and the real target frame;
and (3) performing combined training and learning on the target classification error function and the coordinate error function, and fusing target feature vectors output by each convolutional layer, wherein the error function is expressed as:
wherein p (X) ═ p0(X),...,pK(X)) is the target classification probability value, λ is the balance parameter and is set to 1 by cross validation, the classification error function is calculated using the cross entropy function, Lobj(p(X),y)=-logpy(X)。
As a further improvement of the above solution, the width and length calculation formulas of the initial target box are respectively expressed as:
the central point of the ith row and jth column of each initial target box is set tom is the size of the m x m feature map.
As a further improvement of the above solution, the method trains and learns different targets of a special scale by using a tiled initial detection frame, and assuming that m convolutional layers are used for prediction, the initial value of each layer isThe size ratio of the starting target detection box is expressed as:
wherein s isminThat is, the lowest layer has a scale factor of smaxThe scale coefficient of the highest layer is k, which is the number of the initial target frames.
As a further improvement of the above solution, in step S4, a ratio k between pixels in the shadow-shaped total density set and pixels in the lung region where the pixels in the total density set are located is calculated, and a determination threshold value representing a small shadow density level of each lung region by the ratio k is expressed as:
wherein,pixel sum, N, representing that the ith lung region small shadow density set belongs to the j-level target detection boxiRepresenting the ith lung region pixel value,a decision value representing that the ith lung area cell shadow density set belongs to j;
and judging the pneumoconiosis grade based on the relationship between the pneumoconiosis fiber shadow pixel density of the lung area and the shadow density set j and the ratio of the pneumoconiosis fiber shadow pixel density of the lung area in the lung area.
The present invention also provides an intelligent pneumoconiosis grade discrimination apparatus using any one of the above-described intelligent pneumoconiosis grade discrimination methods, comprising:
the system comprises a sample acquisition module, a comparison module and a comparison module, wherein the sample acquisition module is used for dividing a plurality of DR chest film sample images into a first training sample and a second training sample; the DR chest radiography sample image is each lung area image with different sizes of pneumoconiosis fiber shadow pixels;
the first training module is used for inputting a first training sample based on a convolutional neural network model and is used for training model parameters of the characteristics of the pulmonary fibrosis shadow so as to generate an intelligent judgment model of pulmonary fibrosis;
the identification module is used for inputting a second training sample based on the intelligent lung fibrosis judgment model and identifying the shadow shape and the corresponding coordinates of the pneumoconiosis fiber shadow pixels in each DR chest image in the second training sample;
the training module II is used for analyzing and predicting the pneumoconiosis fiber shadow shape in each DR chest image, determining a shadow density set to which each shadow shape belongs, calculating an overall density set of the shadow shapes, and performing pneumoconiosis grade judgment training by combining the lung area number of coordinates where pneumoconiosis fiber shadow pixels are distributed so as to train a pneumoconiosis grade judgment model;
and the judging module is used for judging the pneumoconiosis grade through each parameter of the pneumoconiosis judging model.
The invention has the beneficial effects that:
in the intelligent determination method for the pneumoconiosis grade, an input image is replaced by a whole DR chest film from an image block for training, and the characteristic learning model is designed to replace the last three full-connection layers by convolution layers.
The intelligent determination method for the pneumoconiosis grade provided by the invention firstly develops a pneumoconiosis grade determination mode, and determines the pneumoconiosis grade by respectively reading different lung fibrosis shadows and pixel ratios of different lung areas through establishing a model, so that the pneumoconiosis grade determination is faster and more effective.
Drawings
Fig. 1 is a schematic flow chart of an intelligent determination method for pneumoconiosis grade according to embodiment 1 of the present invention;
fig. 2 is a schematic diagram of a multi-scale feature mapping pneumoconiosis shadow detection model provided in embodiment 1 of the present invention;
fig. 3 is a schematic diagram of a multi-scale feature mapping module provided in embodiment 1 of the present invention;
FIG. 4 is an original DR chest radiograph, a lung field segmentation result and an overlapped object effect of the lung field region;
fig. 5 is a diagram of the lung segmentation effect.
Detailed Description
In order to make the technical solution and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and 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.
Example 1
Referring to fig. 1 to 3, the method for intelligently determining the pneumoconiosis grade includes the following steps:
s1, dividing the DR chest sample images into a first training sample and a second training sample; the DR chest radiography sample image is each lung area image with different sizes of pneumoconiosis fiber shadow pixels;
s2, inputting a training sample I based on the convolutional neural network model, and training model parameters of the characteristics of the pulmonary fibrosis shadow to generate an intelligent judgment model of pulmonary fibrosis;
s3, inputting a second training sample based on the intelligent lung fibrosis judgment model, and identifying the shadow shape and the corresponding coordinates of the pneumoconiosis fiber shadow pixels in each DR chest image in the second training sample;
s4, analyzing and predicting the pneumoconiosis fiber shadow shape in each DR chest image, determining the shadow density set to which each shadow shape belongs, calculating the overall density set of the shadow shapes, and performing pneumoconiosis grade judgment training by combining the lung area number of the coordinate where pneumoconiosis fiber shadow pixels are distributed so as to train a pneumoconiosis grade judgment model;
and S5, judging the pneumoconiosis grade according to each parameter of the pneumoconiosis judging model.
In step S1, for each DR chest sample image, it is necessary to calibrate the pneumoconiosis fiber shadow pixels of each type included in each lung region in each DR chest sample image in advance, and use the calibrated DR chest sample image as a sample label.
In the step S2, after each DR chest sample image is sequentially convolved by several convolution layers of the convolutional neural network model, target feature vectors with different mapping scales are respectively generated;
and training model parameters of the convolutional neural network model through a composite training loss function according to the obtained target characteristic vectors and sample labels calibrated in advance in a training set so as to train an intelligent judgment model for pulmonary fibrosis.
The convolutional neural network model comprises 8 convolutional layers which are sequentially connected, and each convolutional layer uses a 1 x 1 convolutional kernel to perform corresponding feature extraction;
after the DR chest sample image passes through convolutional layers conv4, conv5, conv6 and conv8, a plurality of target feature vectors for comparison with sample tags are generated respectively.
The image pixels of the convolutional layer conv1-3 are all 512 × 515, and the image pixels of the convolutional layer conv4, the convolutional layer conv5, the convolutional layer conv6, the convolutional layer conv7 and the convolutional layer conv8 are respectively 256 × 256, 128 × 128, 64 × 64, 32 × 32 and 16 × 16 in sequence.
The method for training the model parameters of the features of the pulmonary fibrosis shadow by the composite training loss function comprises the following steps:
where W is the model parameter of the convolutional neural network model, and S ═ S1,S2,...,SNIs the set of training samples, N is the number of convolutional layers, αnIs an error function lnWeight of (1), XiIs a marked sample image block, Yi=(yi,bi) Is a labeled sample label, yiE, e {0, 1., K }, and the real target frame coordinates (x, y, w, h) respectively represent the center point coordinates and the width and length values of the target frame;
wherein m is the number of layers of the convolution layer where the prediction target frame is located during detection, i is the coordinate of the detected initial target frame, j is the coordinate of the real target frame, and the coordinate error function adopts smoothL1() Solving the coordinate errors of the predicted target frame and the real target frame, thenThe calculation method comprises the following steps:
wherein,represents the center, length and width coordinates of the predicted target frame,the coordinates of the center, length and width of the real target frame are shown,and representing the coordinate difference value of the predicted target frame and the real target frame.
And (3) performing combined training and learning on the target classification error function and the coordinate error function, and fusing target feature vectors output by each convolutional layer, wherein the error function is expressed as:
wherein p (X) ═ p0(X),...,pK(X)) is the target classification probability value, λ is the balance parameter and is set to 1 by cross validation, the classification error function is calculated using the cross entropy function, Lobj(p(X),y)=-logpy(X)。
The width and length calculation formulas of the initial target frame are respectively expressed as:
the central point of the ith row and jth column of each initial target box is set tom is the size of the m x m feature map.
The method for training and learning different targets with special scales by using the tiled initial detection frame mode is characterized in that the size proportion of the initial target detection frame of each layer is expressed as follows under the assumption that m convolutional layers are used for prediction:
wherein s isminThat is, the lowest layer has a scale factor of smaxThe scale coefficient of the highest layer is k, which is the number of the initial target frames.
In step S4, a ratio k between the pixels of the shadow-shaped total density set and the pixels of the total density set in the lung region where the pixels are located is calculated, and the determination threshold value representing the level of the shadow density of each lung region by the ratio k is represented as:
wherein,pixel sum, N, representing that the ith lung region small shadow density set belongs to the j-level target detection boxiRepresenting the ith lung region pixel value,a decision value representing that the ith lung area cell shadow density set belongs to j;
and judging the pneumoconiosis grade based on the relationship between the pneumoconiosis fiber shadow pixel density of the lung area and the shadow density set j and the ratio of the pneumoconiosis fiber shadow pixel density of the lung area in the lung area.
In the present embodiment, the lung field region is subdivided: the automatic segmentation of the lung field area is a very important link, and the quality of the segmentation effect directly influences the automatic discrimination accuracy of the stage of the pneumoconiosis.
The two lobes were divided into three regions according to diagnostic criteria guidelines: the high lung field, the middle lung field and the low lung field, so the left and right lung lobes have 6 lung regions. Referring to fig. 4, the left side of the graph shows the original DR chest radiography graph, the middle of the graph shows the result after lung field segmentation, and the right side of the graph shows the superimposed physical effect. The vertical distance between the lung tip and the diaphragm is calculated from the lung field regions divided from the original chest radiograph, then the left and right lung fields are divided into three regions by calculation, and the contour line is calculated according to the position of the horizontal line, that is, as shown in fig. 5.
The stage discrimination of pneumoconiosis is specifically as follows:
1. small shadow and large shadow
The shadow is divided into two types of circle and irregular shape according to the shape of the small shadow, p and s mean that the diameter (width diameter) is not more than 1.5mm, q and t mean that the diameter (width diameter) is between 1.5mm and 3mm, r and u mean that the diameter (width diameter) is between 3mm and 10mm, and the shadow with the diameter (width diameter) exceeding 10mm is called as a large shadow.
2. Term determination factor
Through the analysis of the national occupational pneumoconiosis diagnostic standard, a clinical expert judges whether a lung area generates fibrosis lesion and needs to have two elements, wherein the fibrosis small shadow distribution range refers to the number of the lung areas with small shadows of the existing grade 1 density (including grade 1), otherwise, a lesion distribution area is not calculated, and the lung area density is judged to require that the small shadow distribution at least accounts for two thirds of the area of the lung area, otherwise, the lesion distribution area is not calculated.
3. Small shadow density set
The diagnosis of the lesion region needs to determine the shadow density set to which the lesion region belongs according to the shadow form, then calculate the overall density set, namely the density of the lung region with the highest density in the whole lung, and finally comprehensively diagnose the pneumoconiosis stage by combining the number of the lung regions in distribution. Therefore, assuming that there are i e (1,2,. 6) lung regions, the small shadow density set is classified into 4 levels, i.e., j is (1,2,3,4) levels. The research adopts a threshold judgment method to diagnose the small shadow density set, and calculates the ratio k of the detected pixels of different small shadow density sets and the pixels of the lung area where the pixels are respectively located to express the small shadow density grade of each lung area, and the formula is as follows:
wherein,pixel sum, N, representing that the ith lung region small shadow density set belongs to the j-level target detection boxiRepresenting the ith lung region pixel value,and a decision value representing that the ith lung area cell shadow density set belongs to j.
According to research and analysis, the judgment rule of the lung region density set is that the small shadow distribution range occupies more than two thirds of the area of the lung region, so thatThe threshold is set to 0.67, and then there are two pneumoconiosis stage diagnosis decision parameters, one is the highest shadow density level j with k value exceeding 0.67, and the other is the number of lung areas, and the pneumoconiosis stage decision parameters are shown in table 1 below.
Note: j denotes the shadow density level
TABLE 1 auxiliary diagnosis of pneumoconiosis parameters
Example 2
The pneumoconiosis grade intelligent determination device using the pneumoconiosis grade intelligent determination method of embodiment 1 includes:
the system comprises a sample acquisition module, a comparison module and a comparison module, wherein the sample acquisition module is used for dividing a plurality of DR chest film sample images into a first training sample and a second training sample; the DR chest radiography sample image is each lung area image with different sizes of pneumoconiosis fiber shadow pixels;
the first training module is used for inputting a first training sample based on a convolutional neural network model and is used for training model parameters of the characteristics of the pulmonary fibrosis shadow so as to generate an intelligent judgment model of pulmonary fibrosis;
the identification module is used for inputting a second training sample based on the intelligent lung fibrosis judgment model and identifying the shadow shape and the corresponding coordinates of the pneumoconiosis fiber shadow pixels in each DR chest image in the second training sample;
the training module II is used for analyzing and predicting the pneumoconiosis fiber shadow shape in each DR chest image, determining a shadow density set to which each shadow shape belongs, calculating an overall density set of the shadow shapes, and performing pneumoconiosis grade judgment training by combining the lung area number of coordinates where pneumoconiosis fiber shadow pixels are distributed so as to train a pneumoconiosis grade judgment model;
and the judging module is used for judging the pneumoconiosis grade through each parameter of the pneumoconiosis judging model.
The present invention is not limited to the above preferred embodiments, and any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. An intelligent pneumoconiosis grade judging method is characterized by comprising the following steps:
s1, dividing the DR chest sample images into a first training sample and a second training sample; the DR chest radiography sample image is each lung area image with different sizes of pneumoconiosis fiber shadow pixels;
s2, inputting a training sample I based on the convolutional neural network model, and training model parameters of the characteristics of the pulmonary fibrosis shadow to generate an intelligent judgment model of pulmonary fibrosis;
s3, inputting a second training sample based on the intelligent lung fibrosis judgment model, and identifying the shadow shape and the corresponding coordinates of the pneumoconiosis fiber shadow pixels in each DR chest image in the second training sample;
s4, analyzing and predicting the pneumoconiosis fiber shadow shape in each DR chest image, determining the shadow density set to which each shadow shape belongs, calculating the overall density set of the shadow shapes, and performing pneumoconiosis grade judgment training by combining the lung area number of the coordinate where pneumoconiosis fiber shadow pixels are distributed so as to train a pneumoconiosis grade judgment model;
and S5, judging the pneumoconiosis grade according to each parameter of the pneumoconiosis judging model.
2. The method for intelligently discriminating a pneumoconiosis grade according to claim 1, wherein in step S1, for each DR chest radiograph sample image, it is necessary to calibrate in advance the pneumoconiosis fiber shadow pixels of each type contained in each lung region in each DR chest radiograph sample image, and use the calibrated DR chest radiograph sample image as a sample label.
3. The method according to claim 2, wherein in step S2, for each DR chest film sample image, after being convolved by several convolution layers of a convolutional neural network model in sequence, target feature vectors with different mapping scales are generated respectively;
and training the model parameters of the convolutional neural network model through a composite training loss function according to the obtained target feature vectors and the pre-calibrated sample labels in the training sample I.
4. The intelligent pneumoconiosis grade discrimination method according to claim 3, wherein the convolutional neural network model comprises 8 convolutional layers connected in sequence, each convolutional layer uses a 1 x 1 convolutional kernel to perform corresponding feature extraction;
after the DR chest sample image passes through convolutional layers conv4, conv5, conv6 and conv8, a plurality of target feature vectors for comparison with sample tags are generated respectively.
5. The method as claimed in claim 3, wherein the image pixels of the convolutional layer conv1-3 are 512 × 515, and the image pixels of the convolutional layer conv4, convolutional layer conv5, convolutional layer conv6, convolutional layer conv7 and convolutional layer conv8 are respectively 256 × 256, 128 × 128, 64 × 64, 32 × 32 and 16 × 16.
6. The intelligent pneumoconiosis grade discrimination method according to claim 3, wherein the method for training the model parameters of the features of the pulmonary fibrosis shadow by the composite training loss function comprises the following steps:
where W is the model parameter of the convolutional neural network model, and S ═ S1,S2,...,SNIs the set of training samples, N is the number of convolutional layers, αnIs an error function lnWeight of (1), XiIs a marked sample image block, Yi=(yi,bi) Is a labeled sample label, yiE, e {0, 1., K }, and the real target frame coordinates (x, y, w, h) respectively represent the center point coordinates and the width and length values of the target frame;
wherein m is the number of layers of the convolution layer where the prediction target frame is located during detection, i is the coordinate of the detected initial target frame, j is the coordinate of the real target frame, and the coordinate error function adopts smoothL1() Solving the coordinate errors of the predicted target frame and the real target frame, thenThe calculation method comprises the following steps:
wherein,represents the center, length and width coordinates of the predicted target frame,the coordinates of the center, length and width of the real target frame are shown,representing a coordinate difference value of the predicted target frame and the real target frame;
and (3) performing combined training and learning on the target classification error function and the coordinate error function, and fusing target feature vectors output by each convolutional layer, wherein the error function is expressed as:
wherein p (X) ═ p0(X),...,pK(X)) is the target classification probability value, λ is the balance parameter and is set to 1 by cross validation, the classification error function is calculated using the cross entropy function, Lobj(p(X),y)=-logpy(X)。
7. The intelligent pneumoconiosis grade discrimination method according to claim 6, wherein the width and length calculation formulas of the initial target box are respectively expressed as:
8. The method according to claim 6, wherein the training and learning of different targets with special scales by using the tiled initial detection boxes is performed, and assuming that m convolutional layers are used for prediction, the size ratio of the initial target detection box of each layer is expressed as:
wherein s isminThat is, the lowest layer has a scale factor of smaxThe scale coefficient of the highest layer is k, which is the number of the initial target frames.
9. The method for intelligently discriminating a pneumoconiosis level as claimed in claim 1, wherein in step S4, a ratio k between pixels in the shadow-shaped ensemble density set and pixels in the lung region where the pixels are located is calculated, and the decision threshold for expressing the shadow density level of each lung region by the ratio k is expressed as:
wherein,pixel sum, N, representing that the ith lung region small shadow density set belongs to the j-level target detection boxiRepresenting the ith lung region pixel value,a decision value representing that the ith lung area cell shadow density set belongs to j;
and judging the pneumoconiosis grade based on the relationship between the pneumoconiosis fiber shadow pixel density of the lung area and the shadow density set j and the ratio of the pneumoconiosis fiber shadow pixel density of the lung area in the lung area.
10. An intelligent pneumoconiosis grade discrimination apparatus using the intelligent pneumoconiosis grade discrimination method according to any one of claims 1 to 9, comprising:
the system comprises a sample acquisition module, a comparison module and a comparison module, wherein the sample acquisition module is used for dividing a plurality of DR chest film sample images into a first training sample and a second training sample; the DR chest radiography sample image is each lung area image with different sizes of pneumoconiosis fiber shadow pixels;
the first training module is used for inputting a first training sample based on a convolutional neural network model and is used for training model parameters of the characteristics of the pulmonary fibrosis shadow so as to generate an intelligent judgment model of pulmonary fibrosis;
the identification module is used for inputting a second training sample based on the intelligent lung fibrosis judgment model and identifying the shadow shape and the corresponding coordinates of the pneumoconiosis fiber shadow pixels in each DR chest image in the second training sample;
the training module II is used for analyzing and predicting the pneumoconiosis fiber shadow shape in each DR chest image, determining a shadow density set to which each shadow shape belongs, calculating an overall density set of the shadow shapes, and performing pneumoconiosis grade judgment training by combining the lung area number of coordinates where pneumoconiosis fiber shadow pixels are distributed so as to train a pneumoconiosis grade judgment model;
and the judging module is used for judging the pneumoconiosis grade through each parameter of the pneumoconiosis judging model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111315770.1A CN114098779A (en) | 2021-11-08 | 2021-11-08 | Intelligent pneumoconiosis grade judging method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111315770.1A CN114098779A (en) | 2021-11-08 | 2021-11-08 | Intelligent pneumoconiosis grade judging method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114098779A true CN114098779A (en) | 2022-03-01 |
Family
ID=80377556
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111315770.1A Withdrawn CN114098779A (en) | 2021-11-08 | 2021-11-08 | Intelligent pneumoconiosis grade judging method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114098779A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998203A (en) * | 2022-04-27 | 2022-09-02 | 四川大学华西第四医院 | System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence |
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 |
-
2021
- 2021-11-08 CN CN202111315770.1A patent/CN114098779A/en not_active Withdrawn
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998203A (en) * | 2022-04-27 | 2022-09-02 | 四川大学华西第四医院 | System and method for accurately diagnosing occupational pneumoconiosis based on artificial intelligence |
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 |
---|---|---|
CN110992382B (en) | Fundus image optic cup optic disc segmentation method and system for assisting glaucoma screening | |
CN109523535B (en) | Pretreatment method of lesion image | |
CN118674677A (en) | Construction method and application of gastric cancer image recognition model | |
CN109859168A (en) | A kind of X-ray rabat picture quality determines method and device | |
CN114098779A (en) | Intelligent pneumoconiosis grade judging method | |
CN110781953B (en) | Lung cancer pathological section classification method based on multi-scale pyramid convolution neural network | |
CN112819821B (en) | Cell nucleus image detection method | |
CN112365973B (en) | Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN | |
CN112132800B (en) | Deep learning-based pulmonary fibrosis detection and severity assessment method and system | |
CN111161287A (en) | Retinal vessel segmentation method based on symmetric bidirectional cascade network deep learning | |
CN109448854A (en) | A kind of construction method of pulmonary tuberculosis detection model and application | |
CN113782184A (en) | Cerebral apoplexy auxiliary evaluation system based on facial key point and feature pre-learning | |
CN117036288A (en) | Tumor subtype diagnosis method for full-slice pathological image | |
CN113610118A (en) | Fundus image classification method, device, equipment and medium based on multitask course learning | |
CN115579128A (en) | Multi-model feature-enhanced disease screening system | |
CN114842960A (en) | Method for evaluating disease progression and prognosis of new coronary patient based on CT (computed tomography) image and clinical data | |
Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans | |
CN113902738A (en) | Heart MRI segmentation method and system | |
Mridha et al. | Deep learning in lung and colon cancer classifications | |
KR101162599B1 (en) | An automatic detection method of Cardiac Cardiomegaly through chest radiograph analyses and the recording medium thereof | |
Rahman et al. | Deep learning-based left ventricular ejection fraction estimation from echocardiographic videos | |
CN112690815A (en) | System and method for assisting in diagnosing lesion grade based on lung image report | |
JP7445672B2 (en) | Gate area estimation program, gate area estimation device, learning model generation method | |
CN108345769A (en) | A kind of integral image spatial information based on tissue points whole-genome association method | |
CN111652840B (en) | Turbid screening and classifying device for X-ray chest X-ray image lung |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220301 |
|
WW01 | Invention patent application withdrawn after publication |