CN115810016B - Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image - Google Patents
Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image Download PDFInfo
- Publication number
- CN115810016B CN115810016B CN202310104254.7A CN202310104254A CN115810016B CN 115810016 B CN115810016 B CN 115810016B CN 202310104254 A CN202310104254 A CN 202310104254A CN 115810016 B CN115810016 B CN 115810016B
- Authority
- CN
- China
- Prior art keywords
- dense
- connection
- network
- cxr
- densenet
- 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.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an automatic recognition method, an automatic recognition system, a storage medium and a terminal for a lung infection CXR image, relates to automatic recognition of a pneumonia image, and aims to solve the technical problem that in the prior art, dense connection networks have poor automatic recognition effect of the pneumonia image due to data compression and network gradient reduction. The dense connection network is built, and each layer of dense blocks of the dense connection network are connected with the dense blocks of all the previous layers and serve as input of the dense blocks of the next layer; and the connection between the dense blocks comprises a left connection branch and a right connection branch which are arranged in parallel, wherein the left connection branch comprises two convolution layers, and the right connection branch comprises four convolution layers. The network structure of the invention is changed into a bidirectional compact layer and a transition layer, the compression-free operation can be carried out, the multi-scale characteristics of the pneumonia can be extracted, gradient disappearance can not be caused even if the dense connection network is deepened, and the network can be deepened to obtain a better pneumonia identification effect.
Description
Technical Field
The invention belongs to the technical field of image segmentation in artificial intelligence medicine, relates to automatic identification of a pneumonia image, and particularly relates to an automatic identification method, an automatic identification system, a storage medium and a terminal of a lung infection CXR image.
Background
The rapid spread of the new coronavirus infection (covd-19) disease has created unprecedented pressure on the medical system in many countries. For viral infections, which often manifest as upper respiratory tract infections or pneumonia, can also progress further to acute respiratory failure, multiple organ failure and death. Chest X-ray (CXR) photography remains the primary means of screening, classifying, and diagnosing various pneumonia in existing clinical diagnostics and therapy, including bacterial, viral, and other types of pneumonia. Recent developments in Artificial Intelligence (AI) provide new potential opportunities for rapid growth in radiodiagnostic applications, and AI models using weakly supervised classification or attention-based convolutional neural networks have also been investigated for diagnosis of common pulmonary and cardiac diseases based on CXRs. Therefore, through a machine learning method, relevant pathological image features in the images are automatically extracted, and then the judgment of the pneumonia is carried out based on the image features, so that the actual requirements of clinicians are met.
Automatic pneumonia recognition methods have been widely studied for several years, and these methods can be classified into conventional pneumonia recognition methods based on statistical learning and pneumonia recognition methods based on deep learning. However, the deep learning method is superior to the traditional machine learning method in many medical image analysis tasks, including tasks such as detection, classification, segmentation and the like.
Deep learning has made a great progress in various tasks of natural image recognition, the appearance of AlexNet (2012) expands the depth of LeNet, the precedent of deep learning as image processing is opened, and a deep neural network is widely applied. VGGNet (2014) further improves the performance of deep learning by increasing the network depth, reducing the convolution kernel size. ResNe (2015) uses residual modules to compose a more complex network, trained using standard random gradient descent methods. The structure can reduce gradient disappearance in the model training process. The recognition rate of the ImageNet dataset was reduced to 3.57% in the test. In the same year, the depth of the network is increased, the width of the network is also increased, the features are extracted by using convolution kernels with different sizes, convolution with different sizes is calculated, then the results of the convolution kernels are connected, the outputs of the filters are stacked along the channel dimension and transferred to the next layer, different information of an input image can be obtained, and the operations are processed in parallel and combined with all the results to obtain better image characterization. In a further development, researchers are interested in three-dimensional images, which have more spatial features. Researchers have proposed using three-dimensional convolutional neural networks for application in three-dimensional data, thereby making progress in three-dimensional object recognition tasks.
In the prior art, there are also deep learning methods applied to pneumonia recognition tasks: the CheXNet is a 121-layer dense convolution network trained on a data set of the Huant-ray 14, so that the effect of pneumonia identification is effectively improved; there are also signs of early diagnosis and better patient treatment based on pneumonia using Convolutional Neural Network (CNN) analysis.
In addition, the invention patent application with the application number of 202011184333 discloses a novel coronavirus infection CT image processing method based on non-uniform pooling of lung, which comprises the following steps: s1, a full-automatic lung segmentation algorithm based on FPN; s2, identifying a double-lung central line; s3, pooling operation of the lung; s4, convolutional neural network based on pulmonary pooling. In step S1, a feature pyramid full convolution neural network based on the DenseNet121 is constructed, and a lung region is fully automatically segmented from a CT image. The feature pyramid full convolution neural network FPN uses a DenseNet network with pre-trained weights in an ImageNet as a basic network, extracts the output of the last layer of convolution layer from each DenseNet block in the DenseNet in the form of a feature pyramid as a multi-scale feature, then gradually upsamples and splices the features with different scales, and finally obtains a segmented lung region, namely a lung target region ROI in a full convolution network mode. The segmentation algorithm is characterized in that the feature pyramid full convolution neural network based on DenseNet121 is constructed, features of lung regions are amplified when the convolution neural network is pooled, the regions outside the lung are compressed, redundant features are eliminated, image information in the lung is enhanced, the precision of a novel coronavirus infection CT image processing method is improved, meanwhile, any manual labeling image is not relied on, and the practicability of the algorithm is improved.
The invention patent application of application number 202010666696 also discloses a novel recovery time prediction method and a system for a patient suffering from coronavirus infection based on deep learning, wherein the method comprises the steps of acquiring a plurality of CT sequence images of the patient suffering from coronavirus infection, and preprocessing the CT sequence images of the plurality of days; respectively inputting the lung lobe segmentation model and the pneumonia segmentation model, and respectively extracting the lung lobe area and the focus area for a plurality of days; calculating the ratio of the lesion area to the lung lobe area in a plurality of days to obtain a lesion area ratio value in a plurality of days; and (3) fitting a Gaussian process model by using a lesion area proportion value R for a plurality of days, and predicting the rehabilitation time of a novel coronavirus infected patient. The invention uses Densenet as the deep Lab V3+ frame of the backstone and the 3D UNet framework to divide lung lobes and pneumonia areas, the division is quick and effective, and the Gaussian process can accurately predict the rehabilitation time of patients and is used as a reference for medical resource allocation.
As in the two patent documents, there is a pneumonia image recognition network based on a DenseNet network in the prior art. However, in the prior art, denseNet networks typically have multiple layers of dense blocks, each of which is connected to the dense blocks of all the previous layers as input to the next layer of dense blocks. And when the dense blocks are connected, the connection is in a single branch structure, as shown in fig. 1, and comprises two convolution layers with a convolution kernel 1*1 and a convolution kernel 3*3, wherein the steps of the two convolution kernels are respectively 4k and k. Although the DenseNet network with the structure can realize automatic identification of the pneumonia image, the dense connection network can cause incomplete characteristics extracted by data compression operation, so that the extracted multi-scale characteristics of the pneumonia are limited, and the automatic identification effect of the pneumonia image is poor.
Disclosure of Invention
The invention aims at: in order to solve the technical problem that the automatic recognition effect of the pneumonia image is poor due to data compression of dense connection networks in the prior art, the invention provides a method, a system, a storage medium and a terminal for automatically recognizing the lung infection CXR image.
The invention adopts the following technical scheme for realizing the purposes:
an automatic identification method for CXR images of lung infection comprises the following steps:
step S1, obtaining sample data
Acquiring CXR image samples and corresponding chest pathology labels;
step S2, setting up a dense connection DenseNet network
Constructing a dense connection DenseNet network, wherein the dense connection DenseNet network comprises a plurality of layers of dense blocks, each layer of dense block is connected with the dense blocks of all the previous layers through dense connection, the next layer of dense block is connected with the dense blocks of all the previous layers through feedback connection, and each dense block extracts the output of the last layer of convolution layer, gradually up-samples along the direction close to the input direction, convolves the output and inputs the output to the first layer of dense block;
the dense connection and feedback connection between the dense blocks comprise a left connection branch and a right connection branch which are arranged in parallel, wherein the left connection branch comprises two convolution layers, and the right connection branch comprises four convolution layers;
step S3, training dense connection DenseNet network
Training the densely connected DenseNet network by adopting the CXR image sample and the chest pathology label obtained in the step S1 to obtain a mature densely connected DenseNet network;
step S4, identifying in real time
Inputting the CXR image acquired in real time into the mature dense connection DenseNet network obtained in the step S3, and outputting a chest pathology recognition result.
Further, in step S1, the CXR image samples and corresponding chest pathology labels are from the ChestX-ray14 dataset, comprising 112120 frontal X-ray images of 30805 patients, each labeled with up to 14 different chest pathology labels.
Further, in step S2, the left connection branch includes two convolution layers of 1*1 and 2k and 3*3 and k/2, and the right connection branch includes four convolution layers of 1*1 and 2k, 3*3 and k/2, 3*3 and k/2, 3*3 and k/2.
Further, in step S3, a weighted binary cross entropy loss function is employed in training a dense connection DenseNet network
The method comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,
sample input for network, ++>
Binary label for network output +.>
；/>
Probability assigned to the network to tag 1, +.>
Probability of being assigned to tag 0 for the network, +.>
、/>
The number of positive cases and negative cases of pneumonia in the sample are respectively indicated.
Further, in step S3, when training the dense connection densnet network, an initial learning rate of 0.001 is used, each time the verification loss becomes stable after an epoch, it is attenuated by 10%, and a model with the lowest verification loss is selected; after each iteration learning, the network judges the total error of the pneumonia recognition result, if the current error is smaller than the error of the previous iteration, the current model is saved, and then training is continued until the maximum iteration times are reached.
An automated lung infection, CXR, image recognition system comprising:
the sample data acquisition module is used for acquiring CXR image samples and corresponding chest pathology labels;
the dense connection DenseNet network building module is used for building a dense connection DenseNet network, the dense connection DenseNet network comprises a plurality of layers of dense blocks, each layer of dense block is connected with the dense blocks of all the previous layers through dense connection, the next layer of dense block is connected with the dense blocks of all the previous layers through feedback connection, and each dense block extracts the output of the last layer of convolution layer, gradually upsamples along the direction close to the input direction, and inputs the output to the first layer of dense block after convolution operation;
the dense connection and feedback connection between the dense blocks comprise a left connection branch and a right connection branch which are arranged in parallel, wherein the left connection branch comprises two convolution layers, and the right connection branch comprises four convolution layers;
the dense connection DenseNet network training module is used for training the dense connection DenseNet network by adopting the CXR image sample and the chest pathology label acquired by the sample data acquisition module to obtain a mature dense connection DenseNet network;
and the real-time identification module is used for inputting the CXR image acquired in real time into the mature dense connection DenseNet network obtained by the dense connection DenseNet network training module and outputting a chest pathology identification result.
A storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method described above.
A terminal comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method described above.
The beneficial effects of the invention are as follows:
1. in the invention, not only the dense connection for connecting to the output direction, but also the feedback connection for connecting to the input direction are arranged between the dense blocks, and the dense connection and the feedback connection can be used for receiving the information to form an effective loop which is more in accordance with the biological neural network structure and the context to train the model to share parameter information; in addition, the output of the last layer of convolution layer is extracted from each dense block, the output is subjected to step-by-step up-sampling and convolution operation along the direction close to the input direction, and then the output is input to the first layer of dense block, so that the resolution of an image can be improved, meanwhile, connection is performed during up-sampling to fuse bottom layer position information with deep semantics, and finally, the output is input to the first dense block, so that more-dimensional characteristics can be extracted through training, and the improvement of model performance is facilitated.
2. In the invention, a structure based on dense connection network is adopted, and the link operation can enable the feature map learned by any layer to be used by other subsequent layers, so that the features can be reused, the model becomes more compact, the small model with fewer parameters can remarkably save bandwidth, reduce storage cost, and has faster calculation and strong fitting resistance.
3. In the invention, the network model is trained by adopting the optimized weighted binary cross entropy loss function, and the target evaluation index is directly optimized, so that the model can be more fully trained.
4. According to the invention, the trained network is adopted to be used for rapid detection and identification of pneumonia, so that manpower and material resources for primary screening are greatly saved, and more accurate detection conclusion is provided for subsequent medical staff.
Drawings
FIG. 1 is a schematic diagram of a dense block to dense block connection in a prior art DenseNet network;
FIG. 2 is a schematic flow chart of the present invention;
FIG. 3 is a schematic diagram of a densely connected DenseNet network according to the present invention;
fig. 4 is a schematic diagram of the structure of dense block-to-dense block connection in the DenseNet network of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
The embodiment provides an automatic recognition method for lung infection CXR images, which is characterized in that a dense connection DenseNet network is built, the network model is trained, and the trained network model can realize automatic recognition for the lung infection CXR images. The identification method is shown in fig. 2, and specifically comprises the following steps:
step S1, obtaining sample data
And obtaining CXR image samples and corresponding chest pathology labels.
In step S1, the CXR image samples and corresponding chest pathology labels are from the ChestX-ray14 dataset, comprising 112120 frontal X-ray images of 30805 patients, each labeled with up to 14 different chest pathology labels.
An image with pneumonia was marked as a positive example for one annotated pathology, and all other images were marked as negative examples. In the pneumonia detection task, the dataset of CXR image samples was randomly divided into training (28744 patients, 98637 images), validation (1672 patients, 6351 images) and testing (389 patients, 420 images), with no overlap between adjacent groups. The CXR image samples were scaled down to 224X 224 and normalized according to the mean and standard deviation of the CXR image samples in the ImageNet training set prior to inputting the CXR image samples into the network.
Step S2, setting up a dense connection DenseNet network
A dense connection DenseNet network as shown in fig. 3 is constructed, the dense connection DenseNet network includes a plurality of layers of dense blocks, each layer of dense blocks is connected with the dense blocks of all the previous layers through dense connections, the next layer of dense blocks is connected with the dense blocks of all the previous layers through feedback connections, and each dense block extracts the output of the last layer of convolution layer and gradually up-samples along the direction close to the input, and the output is input to the first layer of dense blocks after convolution operation.
And the dense connection and feedback connection between the dense blocks are shown in fig. 4, which includes a left connection branch and a right connection branch arranged in parallel, the left connection branch includes two convolution layers, and the right connection branch includes four convolution layers.
The left connecting branch comprises two convolution layers with a convolution kernel of 1*1 and a step length of 2k, a convolution kernel of 3*3 and a step length of k/2, which are sequentially arranged, and the right connecting branch comprises four convolution layers with a convolution kernel of 1*1 and a step length of 2k, a convolution kernel of 3*3 and a step length of k/2, a convolution kernel of 3*3 and a step length of k/2, a convolution kernel of 3*3 and a step length of k/2, which are sequentially arranged.
The left path of the upper layer is connected to the filter connection through the 1*1 convolution layers and the 3*3 convolution layers in sequence, and the right path of the upper layer is connected to the filter connection through the 1*1 convolution layers and the 3 3*3 convolution layers in sequence. The densely connected DenseNet network, besides halving the filter of the original trunk branch (i.e. the left connection branch), also adds a new branch (i.e. the right connection branch) in which 3 convolutions of 3x3 size are used, the branch receptive field becomes the same as the 7x7 convolutions; the extracted features are not only single-scale, but also small targets and large targets; in densely connected DenseNet networks, the transition layer is used to reduce the spatial resolution of the feature map, and the number of channels in the transition layer is smaller than that of the previous layer: this operation is found to reduce the expressive power of the network characteristics, and thus the number of channels of the transition layer and the previous layer is set to be the same in the present network, so that the expressive power can be better obtained.
Step S3, training dense connection DenseNet network
Training the densely connected DenseNet network by adopting the CXR image sample and the chest pathology label obtained in the step S1 to obtain the mature densely connected DenseNet network.
In step S3, when training the dense connection DenseNet network, the weighted binary cross entropy loss function is adopted
The method comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,
sample input for network, ++>
Binary label for network output +.>
；/>
Probability assigned to the network to tag 1, +.>
Probability of being assigned to tag 0 for the network, +.>
、/>
The number of positive cases and negative cases of pneumonia in the sample are respectively indicated.
When training a dense connection DenseNet network, using an initial learning rate of 0.001, each time the validation loss tends to stabilize after an epoch, it decays by 10% and a model with the lowest validation loss is selected; after each iteration learning, the network judges the total error of the pneumonia recognition result, if the current error is smaller than the error of the previous iteration, the current model is saved, and then training is continued until the maximum iteration times are reached.
Step S4, identifying in real time
Inputting the CXR image acquired in real time into the mature dense connection DenseNet network obtained in the step S3, and outputting a chest pathology recognition result.
Example 2
The embodiment provides an automatic recognition system for CXR images of lung infection, which specifically comprises:
and the sample data acquisition module is used for acquiring CXR image samples and corresponding chest pathology labels.
In the sample data acquisition module, CXR image samples and corresponding chest pathology labels are from a ChestX-ray14 dataset, which includes 112120 frontal X-ray images of 30805 patients, and each X-ray image is labeled with up to 14 different chest pathology labels.
An image with pneumonia was marked as a positive example for one annotated pathology, and all other images were marked as negative examples. In the pneumonia detection task, the dataset of CXR image samples was randomly divided into training (28744 patients, 98637 images), validation (1672 patients, 6351 images) and testing (389 patients, 420 images), with no overlap between adjacent groups. The CXR image samples were scaled down to 224X 224 and normalized according to the mean and standard deviation of the CXR image samples in the ImageNet training set prior to inputting the CXR image samples into the network.
The dense connection DenseNet network building module is used for building a dense connection DenseNet network shown in fig. 3, wherein the dense connection DenseNet network comprises a plurality of layers of dense blocks, each layer of dense block is connected with the dense blocks of all the previous layers through dense connection, the next layer of dense block is connected with the dense blocks of all the previous layers through feedback connection, and each dense block extracts the output of the last layer of convolution layer, and the output is gradually up-sampled along the direction close to the input direction, and is input to the first layer of dense block after convolution operation.
And the dense connection and feedback connection between the dense blocks are shown in fig. 4, and the dense connection and feedback connection comprises a left connection branch and a right connection branch which are arranged in parallel, wherein the left connection branch comprises two convolution layers, and the right connection branch comprises four convolution layers;
the left connecting branch comprises two convolution layers with a convolution kernel of 1*1 and a step length of 2k, a convolution kernel of 3*3 and a step length of k/2, which are sequentially arranged, and the right connecting branch comprises four convolution layers with a convolution kernel of 1*1 and a step length of 2k, a convolution kernel of 3*3 and a step length of k/2, a convolution kernel of 3*3 and a step length of k/2, a convolution kernel of 3*3 and a step length of k/2, which are sequentially arranged.
The left path of the upper layer is connected to the filter connection through the 1*1 convolution layers and the 3*3 convolution layers in sequence, and the right path of the upper layer is connected to the filter connection through the 1*1 convolution layers and the 3 3*3 convolution layers in sequence. The densely connected DenseNet network, besides halving the filter of the original trunk branch (i.e. the left connection branch), also adds a new branch (i.e. the right connection branch) in which 3 convolutions of 3x3 size are used, the branch receptive field becomes the same as the 7x7 convolutions; the extracted features are not only single-scale, but also small targets and large targets; in densely connected DenseNet networks, the transition layer is used to reduce the spatial resolution of the feature map, and the number of channels in the transition layer is smaller than that of the previous layer: this operation is found to reduce the expressive power of the network characteristics, and thus the number of channels of the transition layer and the previous layer is set to be the same in the present network, so that the expressive power can be better obtained.
And the dense connection DenseNet network training module is used for training the dense connection DenseNet network by adopting the CXR image sample and the chest pathology label acquired by the sample data acquisition module to obtain a mature dense connection DenseNet network.
In the dense connection DenseNet network training module, a weighted binary cross entropy loss function is adopted when training the dense connection DenseNet network
The method comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,
sample input for network, ++>
Binary label for network output +.>
；/>
Probability assigned to the network to tag 1, +.>
Probability of being assigned to tag 0 for the network, +.>
、/>
Respectively represent the samplesThe number of positive cases and negative cases of pneumonia in this case.
When training a dense connection DenseNet network, using an initial learning rate of 0.001, each time the validation loss tends to stabilize after an epoch, it decays by 10% and a model with the lowest validation loss is selected; after each iteration learning, the network judges the total error of the pneumonia recognition result, if the current error is smaller than the error of the previous iteration, the current model is saved, and then training is continued until the maximum iteration times are reached.
And the real-time identification module is used for inputting the CXR image acquired in real time into the mature dense connection DenseNet network obtained by the dense connection DenseNet network training module and outputting a chest pathology identification result.
Example 3
The present embodiment provides a storage medium that is a computer-readable storage medium, in which a computer program is stored, which when executed by a processor causes the processor to perform the steps of the above-described automatic identification method for lung infection CXR images.
Wherein the computer-readable storage medium stores an interface display program executable by at least one processor to cause the at least one processor to perform the steps of the lung infection CXR image automatic identification method as described above.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk), comprising several instructions for causing a terminal device (which may be a mobile phone, a computer, a server or a network device, etc.) to perform the automatic identification method for lung infection CXR images according to the embodiments of the present application.
Example 4
The present embodiment provides a terminal, which may be a computer device, including a memory and a processor, where the memory stores a computer program, and the computer program when executed by the processor causes the processor to execute the steps of the above-described automatic pulmonary infection CXR image identification method.
The computer equipment can be computing equipment such as a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The computer equipment can perform man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch pad or voice control equipment and the like.
The memory includes at least one type of readable storage medium including flash memory, hard disk, multimedia card, card memory (e.g., SD or D interface display memory, etc.), random Access Memory (RAM), static Random Access Memory (SRAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), programmable Read Only Memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card) or the like. Of course, the memory may also include both internal storage units of the computer device and external storage devices. In this embodiment, the memory is often used to store an operating system and various application software installed on the computer device, for example, program codes of the automatic identification method for a CXR image of pulmonary infection. In addition, the memory may be used to temporarily store various types of data that have been output or are to be output.
The processor may be a central processing unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments. The processor is typically used to control the overall operation of the computer device. In this embodiment, the processor is configured to execute the program code stored in the memory or process data, such as the program code for the automatic identification method of the lung infection CXR image.
Claims (7)
1. An automatic identification method for CXR images of lung infection is characterized by comprising the following steps:
step S1, obtaining sample data
Acquiring CXR image samples and corresponding chest pathology labels;
step S2, setting up a dense connection DenseNet network
Constructing a dense connection DenseNet network, wherein the dense connection DenseNet network comprises a plurality of layers of dense blocks, each layer of dense block is connected with the dense blocks of all the previous layers through dense connection, the next layer of dense block is connected with the dense blocks of all the previous layers through feedback connection, and each dense block extracts the output of the last layer of convolution layer, gradually up-samples along the direction close to the input direction, convolves the output and inputs the output to the first layer of dense block;
the dense connection and feedback connection between the dense blocks comprise a left connection branch and a right connection branch which are arranged in parallel, wherein the left connection branch comprises two convolution layers, and the right connection branch comprises four convolution layers;
step S3, training dense connection DenseNet network
Training the densely connected DenseNet network by adopting the CXR image sample and the chest pathology label obtained in the step S1 to obtain a mature densely connected DenseNet network;
step S4, identifying in real time
Inputting the CXR image obtained in real time into the mature dense connection DenseNet network obtained in the step S3, and outputting a chest pathology recognition result;
in step S3, when training the dense connection DenseNet network, the weighted binary cross entropy loss function L (x, y) used is:
wherein x is a sample input by the network, y is a binary label output by the network, and y is {0,1}; p (y=1/x) is the probability that the network is allocated to the label 1, P (y=0/x) is the probability that the network is allocated to the label 0, and |p| and |n| respectively represent the number of positive cases and negative cases of pneumonia in the sample.
2. An automatic identification method for a CXR image for pulmonary infection as defined in claim 1, wherein: in step S1, the CXR image samples and corresponding chest pathology labels are from the ChestX-ray14 dataset, comprising 112120 frontal X-ray images of 30805 patients, each labeled with up to 14 different chest pathology labels.
3. An automatic identification method for a CXR image for pulmonary infection as defined in claim 1, wherein: in step S2, the left connecting branch comprises two convolution layers with a convolution kernel of 1*1 and a step length of 2k, a convolution kernel of 3*3 and a step length of k/2, which are sequentially arranged, and the right connecting branch comprises four convolution layers with a convolution kernel of 1*1 and a step length of 2k, a convolution kernel of 3*3 and a step length of k/2, a convolution kernel of 3*3 and a step length of k/2, a convolution kernel of 3*3 and a step length of k/2, which are sequentially arranged.
4. An automatic identification method for a CXR image for pulmonary infection as defined in claim 1, wherein: in step S3, when training a dense connection DenseNet network, using an initial learning rate of 0.001, each time the verification loss becomes stable after an epoch, it decays by 10%, and selecting the model with the lowest verification loss; after each iteration learning, the network judges the total error of the pneumonia recognition result, if the current error is smaller than the error of the previous iteration, the current model is saved, and then training is continued until the maximum iteration times are reached.
5. An automated lung infection, CXR, image recognition system comprising:
the sample data acquisition module is used for acquiring CXR image samples and corresponding chest pathology labels;
the dense connection DenseNet network building module is used for building a dense connection DenseNet network, the dense connection DenseNet network comprises a plurality of layers of dense blocks, each layer of dense block is connected with the dense blocks of all the previous layers through dense connection, the next layer of dense block is connected with the dense blocks of all the previous layers through feedback connection, and each dense block extracts the output of the last layer of convolution layer, gradually upsamples along the direction close to the input direction, and inputs the output to the first layer of dense block after convolution operation;
the dense connection and feedback connection between the dense blocks comprise a left connection branch and a right connection branch which are arranged in parallel, wherein the left connection branch comprises two convolution layers, and the right connection branch comprises four convolution layers;
the dense connection DenseNet network training module is used for training the dense connection DenseNet network by adopting the CXR image sample and the chest pathology label acquired by the sample data acquisition module to obtain a mature dense connection DenseNet network;
the real-time identification module is used for inputting CXR images acquired in real time into the mature dense connection DenseNet network obtained by the dense connection DenseNet network training module and outputting chest pathology identification results;
the DenseNet network comprises a plurality of layers of dense blocks, and when training dense connection of the DenseNet network, a weighted binary cross entropy loss function L (x, y) is adopted as follows:
wherein x is a sample input by the network, y is a binary label output by the network, and y is {0,1}; p (y=1/x) is the probability that the network is allocated to the label 1, P (y=0/x) is the probability that the network is allocated to the label 0, and |p| and |n| respectively represent the number of positive cases and negative cases of pneumonia in the sample.
6. A storage medium, characterized by: a computer program is stored which, when executed by a processor, causes the processor to perform the steps of the method according to any one of claims 1 to 4.
7. A terminal, characterized by: comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310104254.7A CN115810016B (en) | 2023-02-13 | 2023-02-13 | Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310104254.7A CN115810016B (en) | 2023-02-13 | 2023-02-13 | Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115810016A CN115810016A (en) | 2023-03-17 |
| CN115810016B true CN115810016B (en) | 2023-04-28 |
Family
ID=85487851
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310104254.7A Active CN115810016B (en) | 2023-02-13 | 2023-02-13 | Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115810016B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114119361A (en) * | 2021-08-05 | 2022-03-01 | 合肥工业大学智能制造技术研究院 | Method, system, equipment and medium for reconstructing underground image based on super-resolution of TESRGAN network |
Family Cites Families (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10579557B2 (en) * | 2018-01-16 | 2020-03-03 | Advanced Micro Devices, Inc. | Near-memory hardened compute blocks for configurable computing substrates |
| EP3928175A1 (en) * | 2019-03-22 | 2021-12-29 | Siemens Aktiengesellschaft | Inverse and forward modeling machine learning-based generative design |
| US11443514B2 (en) * | 2019-03-22 | 2022-09-13 | Qualcomm Technologies, Inc. | Recognizing minutes-long activities in videos |
| US11694081B2 (en) * | 2019-05-15 | 2023-07-04 | Advanced Micro Devices, Inc. | Accelerating neural networks with one shot skip layer pruning |
| CN110189308B (en) * | 2019-05-17 | 2020-10-23 | 山东财经大学 | Tumor detection method and device based on fusion of BM3D and dense convolution network |
| CN112052642A (en) * | 2019-05-20 | 2020-12-08 | 台湾积体电路制造股份有限公司 | System and method for ESL modeling for machine learning |
| CN110570353B (en) * | 2019-08-27 | 2023-05-12 | 天津大学 | Super-resolution reconstruction method for generating single image of countermeasure network by dense connection |
| CN112560544A (en) * | 2019-09-10 | 2021-03-26 | 中科星图股份有限公司 | Method and system for identifying ground object of remote sensing image and computer readable storage medium |
| US11751832B2 (en) * | 2020-01-30 | 2023-09-12 | GE Precision Healthcare LLC | CTA large vessel occlusion model |
| US11392833B2 (en) * | 2020-02-13 | 2022-07-19 | Soundhound, Inc. | Neural acoustic model |
| CN111860386B (en) * | 2020-07-27 | 2022-04-08 | 山东大学 | Video semantic segmentation method based on ConvLSTM convolutional neural network |
| CN112258394B (en) * | 2020-11-11 | 2022-04-29 | 珠海大横琴科技发展有限公司 | Data processing method, ship tracking method, device, equipment and storage medium |
| CN112419303B (en) * | 2020-12-09 | 2023-08-15 | 上海联影医疗科技股份有限公司 | Neural network training method, system, readable storage medium and device |
| CN113096028A (en) * | 2021-03-03 | 2021-07-09 | 台州学院 | Magnetic resonance image enhancement method based on multitask cooperative network |
| CN112991177A (en) * | 2021-03-23 | 2021-06-18 | 数量级(上海)信息技术有限公司 | Infrared image super-resolution method based on antagonistic neural network |
| CN113129293B (en) * | 2021-04-26 | 2022-08-23 | 长沙理工大学 | Medical image classification method, medical image classification device, computer equipment and storage medium |
| CN115132275B (en) * | 2022-05-25 | 2024-02-27 | 西北工业大学 | Method for predicting EGFR gene mutation state based on end-to-end three-dimensional convolutional neural network |
-
2023
- 2023-02-13 CN CN202310104254.7A patent/CN115810016B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114119361A (en) * | 2021-08-05 | 2022-03-01 | 合肥工业大学智能制造技术研究院 | Method, system, equipment and medium for reconstructing underground image based on super-resolution of TESRGAN network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115810016A (en) | 2023-03-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10417788B2 (en) | Anomaly detection in volumetric medical images using sequential convolutional and recurrent neural networks | |
| CN113421652B (en) | Method for analyzing medical data, method for training model and analyzer | |
| CN112418329B (en) | Cervical OCT image classification method and system based on multi-scale textural feature fusion | |
| US10460150B2 (en) | Deep learning automated dermatopathology | |
| CN111882560B (en) | Lung parenchyma CT image segmentation method based on weighted full convolution neural network | |
| CN109544518B (en) | Method and system applied to bone maturity assessment | |
| CN111291825B (en) | Focus classification model training method, apparatus, computer device and storage medium | |
| CN110197492A (en) | A kind of cardiac MRI left ventricle dividing method and system | |
| CN110838108A (en) | Medical image-based prediction model construction method, prediction method and device | |
| Li et al. | Automated pneumonia detection in chest x-ray images using deep learning model | |
| Pawar et al. | Two‐Stage Hybrid Approach of Deep Learning Networks for Interstitial Lung Disease Classification | |
| JP2023553111A (en) | System and method for evaluating radiology images of pets | |
| CN110570425A (en) | Lung nodule analysis method and device based on deep reinforcement learning algorithm | |
| Xu et al. | Identification of benign and malignant lung nodules in CT images based on ensemble learning method | |
| Sangeetha et al. | Diagnosis of pneumonia using image recognition techniques | |
| CN111210398A (en) | White blood cell recognition system based on multi-scale pooling | |
| CN115810016B (en) | Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image | |
| CN115762721A (en) | Medical image quality control method and system based on computer vision technology | |
| CN113393445B (en) | Breast cancer image determination method and system | |
| Thakral et al. | An innovative intelligent solution incorporating artificial neural networks for medical diagnostic application | |
| Thammarach et al. | AI chest 4 all | |
| CN110458186B (en) | Breast ultrasound image classification method and system based on local reference similarity coding | |
| Tyagi et al. | A novel approach of lung tumor segmentation using a 3D deep convolutional neural network | |
| Marathe et al. | Detection of Lung Cancer using CT Scans with Deep Learning Approach | |
| CN116630628B (en) | Aortic valve calcification segmentation method, system, equipment and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |