CN114818844B - Pulmonary embolism identification device, terminal device and storage medium based on parameter sharing - Google Patents
Pulmonary embolism identification device, terminal device and storage medium based on parameter sharing Download PDFInfo
- Publication number
- CN114818844B CN114818844B CN202110114241.9A CN202110114241A CN114818844B CN 114818844 B CN114818844 B CN 114818844B CN 202110114241 A CN202110114241 A CN 202110114241A CN 114818844 B CN114818844 B CN 114818844B
- Authority
- CN
- China
- Prior art keywords
- feature
- pulmonary embolism
- classification network
- trained
- samples
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a pulmonary embolism recognition device based on parameter sharing, terminal equipment and a storage medium. The device inputs an image to be detected into a first feature extractor which is trained in advance to obtain first features, inputs the first features into a first classification network which is trained in advance to obtain the probability that the image to be detected contains pulmonary embolism, inputs the image to be detected, of which the probability reaches a first threshold value, into a region detection network which is trained in advance to detect the position of the pulmonary embolism and intercept a corresponding region image, inputs the region image into a second feature extractor which is trained in advance to obtain second features, fuses the first features and the second features to obtain mixed features, and inputs the mixed features into a second classification network and a third classification network which are trained in advance to confirm the property type and the position type of the pulmonary embolism. The accuracy of the identification detection result is improved through the commonality and characteristic combination of the feature identification result corresponding to different focus information dimensions.
Description
Technical Field
The embodiment of the invention relates to the technical field of image processing, in particular to a pulmonary embolism recognition device based on parameter sharing, terminal equipment and a storage medium.
Background
Pulmonary embolism is a disease caused by obstruction of the pulmonary artery, and may be a serious or even life threatening condition that occurs when every breath is stressed and painful. Therefore, diagnosis and correct treatment can be performed in time, and the death rate can be greatly reduced.
The clinical symptoms and signs of pulmonary embolism lack of specificity, and are easy to misdiagnose and miss diagnosis clinically. And the doctor spends much time and effort diagnosing the pulmonary embolism, so that the doctor is liable to overdiagnosis. If machine learning can be utilized to assist in more accurate diagnosis and identification, not only can the physician's effort be greatly reduced, but also the patient's management and treatment can be more effective.
Currently, CT (Computed Tomography, electronic computer tomography) pulmonary angiography is the most common medical image type for assessing pulmonary embolism patients. CT scans consist of hundreds of images that require detailed examination to identify blood clots within the pulmonary artery. As the use of imaging continues to grow, radiologist time constraints may lead to delayed diagnosis. How to more accurately identify pulmonary embolism by using chest CT pulmonary angiography image (CT image) and machine learning technology is a urgent problem to be solved. In particular, to improve the diagnostic efficiency, a more specific judgment result on pulmonary embolism is required.
The inventor finds that most of the schemes based on artificial intelligence for predicting the pulmonary embolism are only used for simply predicting whether the suspected pulmonary embolism exists in the chest CT pulmonary angiography image, but cannot provide depth accurate information for identifying the pulmonary embolism when verifying the existing schemes for identifying the pulmonary embolism in the chest CT pulmonary angiography image through machine learning.
Disclosure of Invention
The invention provides a pulmonary embolism recognition device, terminal equipment and storage medium based on parameter sharing, which are used for solving the technical problem that the depth accurate information for pulmonary embolism recognition in the prior art is insufficient.
In a first aspect, an embodiment of the present invention provides a pulmonary embolism identifying device based on parameter sharing, including:
The first extraction unit is used for inputting the image to be detected into a pre-trained first feature extractor so as to obtain first features;
The probability calculation unit is used for inputting the first characteristic into a pre-trained first classification network to obtain the probability that the image to be detected contains pulmonary embolism;
the regional screenshot unit is used for inputting an image to be detected, the probability of which reaches a first threshold value, into a pre-trained regional detection network so as to detect the position of a pulmonary embolism and intercept a corresponding regional image;
A second extraction unit for inputting the region image to a pre-trained second feature extractor to obtain a second feature;
the feature fusion unit is used for fusing the first feature and the second feature to obtain a mixed feature;
and the comprehensive judging unit is used for respectively inputting the mixed characteristics into the pre-trained second classification network and the pre-trained third classification network so as to confirm the property type and the position type of the pulmonary embolism.
In a second aspect, an embodiment of the present invention further provides a terminal device, including:
One or more processors;
A memory for storing one or more programs;
The one or more programs, when executed by the one or more processors, cause the one or more processors to implement parameter sharing based pulmonary embolism identification as described in the first aspect.
In a third aspect, embodiments of the present invention also provide a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the parameter sharing based pulmonary embolism identification as described in the first aspect.
The pulmonary embolism identifying device, the terminal equipment and the storage medium based on parameter sharing are used for inputting an image to be detected into a first feature extractor which is trained in advance to obtain first features, inputting the first features into a first classification network which is trained in advance to obtain the probability that the image to be detected contains pulmonary embolism, inputting the image to be detected, of which the probability reaches a first threshold value, into a region detection network which is trained in advance to detect the position of pulmonary embolism and intercept a corresponding region image, inputting the region image into a second feature extractor which is trained in advance to obtain second features, fusing the first features and the second features to obtain mixed features, and respectively inputting the mixed features into a second classification network which is trained in advance and a third classification network to confirm the property type and the position type of pulmonary embolism. The type and the position of a focus area and a focus in an image to be detected are confirmed in a grading detection and feature fusion mode, hierarchical relations and shared feature information among different focus information dimensions are mined by a multi-layer grading scheme, the commonality and the feature of a feature recognition result are combined, and the accuracy of the recognition and detection result is improved.
Drawings
FIG. 1 is a flowchart of a pulmonary embolism recognition method based on parameter sharing according to an embodiment of the present invention;
Fig. 2 is a schematic diagram of an image processing procedure of a pulmonary embolism recognition method based on parameter sharing according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a first feature extractor according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a first classification network according to an embodiment of the present invention;
Fig. 5 is a schematic structural diagram of a pulmonary embolism recognition device based on parameter sharing according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a terminal device according to an embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are for purposes of illustration and not of limitation. It should be further noted that, for convenience of description, only some, but not all of the structures related to the present invention are shown in the drawings.
It should be noted that the present disclosure is not limited to all the alternative embodiments, and those skilled in the art who review this disclosure will recognize that any combination of the features may be used to construct the alternative embodiments as long as the features are not mutually inconsistent.
For example, in one implementation of the embodiment, a technical feature is described that the first feature extractor comprises 5 convolution layers, and in another implementation of the embodiment, another technical feature is described that the first classification network comprises 2 convolution layers and 1 fully-connected layer, and after reading the description of the present application, a person skilled in the art should recognize that an implementation having both features is also an alternative implementation, that is, in a specific implementation, the first feature extractor comprises 5 convolution layers, and the first classification network comprises 2 convolution layers and 1 fully-connected layer.
The following describes each embodiment in detail.
Fig. 1 is a flowchart of a pulmonary embolism recognition method based on parameter sharing according to an embodiment of the present invention. The pulmonary embolism identifying method based on parameter sharing provided in the embodiment may be executed by an operation device corresponding to the pulmonary embolism identifying method based on parameter sharing, where the operation device may be implemented by software and/or hardware, and the operation device may be configured by two or more physical entities or may be configured by one physical entity.
In the scheme, firstly, the problem related data is defined and described by mathematical symbols. For basic training data, useCT picture representing data set and class label corresponding to the picture, whereinRepresenting the number of CT pictures contained in the dataset,Represent the firstA CT picture is displayed on a screen,Tags for the presence or absence of pulmonary embolism for the CT picture, i.e.Indicating that the picture contains pulmonary embolism. In the same way as described above,Tags for the type of pulmonary embolism in the CT picture, i.eIndicating no pulmonary embolism, chronic pulmonary embolism and acute pulmonary embolism, respectively.Tags for the location of pulmonary embolism in the CT picture, i.eIndicating no pulmonary embolism, pulmonary embolism on the left, middle and right, respectively. The final task of the model after training based on the data set is to give any CT pictureIt is predicted whether it contains an embolism, if so, also the specific category (acute/chronic) and location (left/middle/right) should be predicted. Based on the above problem-related definitions and model descriptions, the present solution is implemented through the related steps in fig. 1. In the actual pulmonary embolism recognition process, the processing process of the image to be detected is basically similar to the processing process of the sample image, the training of each model is completed based on the sample image, after the output of the model is stabilized, the image to be detected is input into each model to complete corresponding prediction, in the following description, the training process of each model is mainly described, and the process of the image to be detected for prediction recognition in each trained model refers to the corresponding training process.
Specifically, referring to fig. 1, the pulmonary embolism identifying method based on parameter sharing specifically includes:
Step S101, inputting an image to be tested into a pre-trained first feature extractor to obtain first features.
The data flow change during the detailed process of this embodiment is shown in fig. 2. As shown in fig. 2, there is、...、N samples total, first using a first feature extractorExtracting input samplesFeatures of (2),Representing the feature set extracted from all samples, i.e. the first feature set obtained in the training phase.
The CT pictures in the training set are 2D images of standard sizes (e.g., 512×512, 256×256) and are input to a first feature extractor as shown in fig. 3, where the first feature extractor shown in fig. 3 includes 5 convolution layers, the first convolution layer being a convolution layer with a convolution kernel size of 3×3, a channel number of 24, and a step size of 1×1. The second convolution layer comprises two parts, namely a convolution block with a convolution kernel size of 3×3, a channel number of 32, a step size of 1×1, and a maximum pooling layer with a convolution kernel size of 3×3 and a step size of 2×2. The third to fifth layers are similar in structure to the second layer except that the number of channels of the third and fourth layers is 64 and the number of channels of the fifth layer is 32. Finally outputting the feature vector corresponding to each sample through the networkThe size is。
Step S102, inputting the first characteristic into a pre-trained first classification network to obtain the probability that the image to be detected contains pulmonary embolism.
With further reference to FIG. 2, features are describedThrough a first classification networkThe classification network is used for calculating the probability of pulmonary embolism in each CT picture. If it isIs greater than a first threshold value (assuming a first threshold value beta) The sample is considered to contain embolism and is identified as having no pulmonary embolism and having pulmonary embolism with 0 or 1 as described above, and withRepresenting a first classification networkThe corresponding features of the samples containing pulmonary embolism are predicted. In a subsequent step, the screened features are subjected toThrough a second classification networkAnd a third classification networkThe category of pulmonary embolism and the location of pulmonary embolism are predicted separately.
The first classification network for calculating the probability of pulmonary embolism in each CT picture is obtained by training based on the following minimum loss function:
Wherein the method comprises the steps of Representing samples in a training setWhether a tag for a pulmonary embolism is present,Representing the total number of samples in the training set,Representing a sampleContaining the probability of pulmonary embolism.
The network structure of the first classification network is shown in fig. 4, which consists of 2 convolutional layers and 1 fully-connected layer. The first convolution layer comprises two parts, namely a convolution block with a convolution kernel size of 3×3, a channel number of 64, a step size of 1×1, and a maximum pooling layer with a convolution kernel size of 3×3, a step size of 2×2. The second convolution layer is similar to the first convolution layer in structure, except that the number of channels is 32, and finally the full connection layer is connected. And (3) passing the first characteristic obtained in the step (S101) through a first classification network to finally obtain a prediction result of the probability that each CT picture contains pulmonary embolism.
Step S103, inputting the image to be detected, the probability of which reaches a first threshold value, into a pre-trained area detection network so as to detect the position of the pulmonary embolism and intercept the corresponding area image.
In the specific implementation of the detection of the area image, the detection of the area image can be implemented by an R-CNN (Region-Convolutional Neural Networks, area-based convolutional neural network) system algorithm (such as R-CNN, fast R-CNN, etc.), or can be implemented by other detection methods, for example, CENTERNET is adopted as a detection network of the area image.
If CENTERNET is used as the area detection network, it is specifically trained by the following loss function:
Where N represents the number of pulmonary embolism, M is the number of center points of the predicted output region image, Representing the probability that each center point prediction belongs to a pulmonary embolism,AndWeight coefficients respectively representing the weight sparsity of the offset of the center point and the target size loss of the region image,AndRepresenting the predicted and actual center point offsets respectively,AndRepresenting the target size of the predicted region image and the target size of the real region image,Representing the super parameter.
In the area detection network using CENTERNET, the training input is a sample picture containing pulmonary embolism, and as a detection algorithm of an anchor-free, CENTERNET outputs probability prediction P for each point as a center point of a target (pulmonary embolism), and directly predicts a center point offset and a target size. The formula for the loss function shown above, the classification loss uses the original focal loss, the first term to the right in the formula above,The value of the super parameter is generally 2. The regression loss takes the L1 loss, the second and third term on the right in the above equation.
As shown in fig. 2, assuming that the screening by the first classification network is performed, there are n samplesAndAnd (3) reaching a first threshold value, respectively passing CT pictures of the two samples through a pre-trained area detection network to obtain the center position (x, y) of the suspected pulmonary embolism, namely, the coordinate of the center position of the suspected embolism in each CT picture, and taking the coordinate as a center point to intercept area images with preset sizes (such as 48 multiplied by 48, 60 multiplied by 60 and the like) from the corresponding CT pictures to obtain an Instance set. In one CT picture, the number of possible detected center positions may be more than one, and thus, there may be a plurality of area images captured in each CT picture corresponding to each center position.
Step S104, inputting the regional image into a pre-trained second feature extractor to obtain a second feature.
The region image captured in step S103 is a 2D image of standard size, and the second feature extractor for identifying the region image has the same structure as the first feature extractor, that is, also includes 5 convolution layers, where the first convolution layer is a convolution layer with a convolution kernel size of 3×3, a channel number of 24, and a step size of 1×1. The second convolution layer comprises two parts, namely a convolution block with a convolution kernel size of 3×3, a channel number of 32, a step size of 1×1, and a maximum pooling layer with a convolution kernel size of 3×3 and a step size of 2×2. The third to fifth layers are similar in structure to the second layer except that the number of channels is 64. Finally outputting the feature vector corresponding to each sample through the networkThe size is。
In a specific implementation, the second feature extractor may be trained based on the following minimization loss function:
wherein, the AndRespectively represent samples in training setIs a combination of the nature tag and the location tag of (c),Representing the number of samples screened according to the first threshold,Representing a first characteristic of the screened ith sample,Representing a second characteristic of the screened ith sample,Representing the blended characteristics of the i-th sample that was screened,A second classification network is used to classify the data,Representing a third classification network. In this embodiment, for the object distinguished from the feature extraction, the sample image screened out is expressed asRepresenting the region image cut out from the screened sample image as. It should be noted that, the feature extraction of the region image achieved by the above second feature extractor is obtained by an optional training method, and in a specific implementation process, other existing image feature extraction schemes may also be used to perform feature extraction on the region image, for example SIFT (Scale-INVARIANT FEATURE TRANSFORM ), HOG (Histogram of Oriented Gradient, direction gradient histogram), and the like.
Step S105, fusing the first feature and the second feature to obtain a mixed feature.
The first feature and the second feature are mixed by:
wherein, the The characteristics of the mixture are represented and,A first characteristic is indicated by the fact that,A second characteristic is indicated by the fact that,Representing a mixture of two feature vectors. The specific mixing strategy can be splicing, adding and fusing of two feature vectors, corresponding multiplication of elements and the like, and in the actually adopted mixing strategy, only the two feature vectors are fused into a feature description through a group of data through a certain calculation mode.
Step S106, the mixed characteristics are respectively input into a second classification network and a third classification network which are trained in advance so as to confirm the property type and the position type of the pulmonary embolism.
The second classification network is trained based on the following minimization loss function:
wherein, the Representing the number of samples screened according to the first threshold,Representing samples in a training setIs a property of the tag of (a),A second classification network is represented as such,Representing the mixing characteristics of the screened samples.
The third classification network is trained based on the following minimization loss function:
wherein, the Representing the number of samples screened according to the first threshold,Representing samples in a training setIs provided with a position tag of (a),A third classification network is represented as such,Representing the mixing characteristics of the screened samples.
The first feature extractor is trained based on the following minimization loss function:
wherein, the Representing the number of samples in the training set,Representing the number of samples screened according to the first threshold,Representing a first characteristic of the screened ith sample,Representing a second characteristic of the screened ith sample,、AndRespectively represent samples in training setIs a disease label, a property label and a position label,A first classification network is represented and is shown,A second classification network is represented as such,A third classification network is represented as such,Representing the mixing characteristics of the screened samples.
The second and third classification networks are similar in structure to the first classification network, and each consists of 2 convolutional layers and 1 fully-connected layer. The first convolution layer comprises two parts, namely a convolution block with a convolution kernel size of 3×3, a channel number of 64, a step size of 1×1, and a maximum pooling layer with a convolution kernel size of 3×3, a step size of 2×2. The second convolution layer is similar to the first convolution layer in structure, except that the number of channels is 32, and finally the full connection layer is connected. The first classification network, the second classification network and the third classification network are mainly different in that the structure of the input layer and the output layer of each network is correspondingly adjusted according to the data input and the classification output.
As can be seen from the limitation of step S104 to step S106, the first classification network, the second classification network and the third classification network all share the parameters of the first feature extractor, the second classification network and the third classification network also share the parameters of the second feature extractor, the parameters of the first feature extractor and the parameters of the second feature extractor are synchronously updated according to the output result of each classifier, and each network loses learning of the shared features by utilizing multiple losses under different classification tasks, so that the association relationship between different classification tasks is established.
In the prior art, only a single judgment is made on the CT picture, if multiple models are needed for classification prediction for realizing multiple judgments, in the scheme, a multi-level classification scheme is simultaneously carried out by using one model through the association relation among various classification tasks, so that the mining of the hierarchical relation and shared characteristic information among different categories is facilitated.
In general, the method predicts whether the pulmonary embolism exists in the image to be measured in a grading manner, if the pulmonary embolism exists, the type and the position of the pulmonary embolism are further judged, usually, the lower-level features (such as the shape) of the image to be measured can be captured at the lower layer of the neural network, so that the feature extractor is utilized to initially extract the features, and the higher-level features of the image can be extracted at the higher layer of the neural network, so that different branches are connected to the feature extractor to respectively and specifically classify the tasks. By embedding the hierarchical structure of categories into the network model, the hierarchical network can be predicted with more interpretability, and the precision of the final classification result is enhanced. And the parameter sharing of the feature extractor at the bottom layer is equivalent to that the loss obtained by each task updates the parameters of the feature extractor, so that the feature extractor can learn the common features among different tasks, the classification networks of different branches are used for learning the characteristic features of the respective tasks, the common features and the characteristics are combined, and the accuracy of the final classification result is enhanced.
The method comprises the steps of inputting an image to be detected into a first feature extractor which is trained in advance to obtain first features, inputting the first features into a first classification network which is trained in advance to obtain the probability that the image to be detected contains pulmonary embolism, inputting the image to be detected, of which the probability reaches a first threshold value, into a region detection network which is trained in advance to detect the position of the pulmonary embolism and intercept a corresponding region image, inputting the region image into a second feature extractor which is trained in advance to obtain second features, fusing the first features and the second features to obtain mixed features, and inputting the mixed features into a second classification network and a third classification network which are trained in advance to confirm the property type and the position type of the pulmonary embolism. The type and the position of a focus area and a focus in an image to be detected are confirmed in a grading detection and feature fusion mode, hierarchical relations and shared feature information among different focus information dimensions are mined by a multi-layer grading scheme, the commonality and the feature of a feature recognition result are combined, and the accuracy of the recognition and detection result is improved.
Fig. 5 is a schematic structural diagram of a pulmonary embolism recognition device based on parameter sharing according to an embodiment of the present invention. Referring to fig. 5, the pulmonary embolism recognition apparatus based on parameter sharing includes a first extraction unit 210, a probability calculation unit 220, a region screenshot unit 230, a second extraction unit 240, a feature fusion unit 250, and a comprehensive judgment unit 260.
The device comprises a first extraction unit 210, a probability calculation unit 220, a region screenshot unit 230, a second extraction unit 240, a feature fusion unit 250 and a comprehensive judgment unit 260, wherein the first extraction unit is used for inputting an image to be detected into a pre-trained first feature extractor to obtain a first feature, the probability calculation unit 220 is used for inputting the first feature into a pre-trained first classification network to obtain the probability that the image to be detected contains a pulmonary embolism, the region screenshot unit 230 is used for inputting the image to be detected, the probability of which reaches a first threshold value, into a pre-trained region detection network to detect the position of the pulmonary embolism and intercept a corresponding region image, the second extraction unit 240 is used for inputting the region image into a pre-trained second feature extractor to obtain a second feature, the feature fusion unit 250 is used for fusing the first feature and the second feature to obtain a mixed feature, and the comprehensive judgment unit 260 is used for respectively inputting the mixed feature into a pre-trained second classification network and a third classification network to confirm the property type and the position type of the pulmonary embolism.
On the basis of the above embodiment, the first classification network is trained based on the following minimization loss function:
wherein, the Representing samples in a training setWhether a tag for a pulmonary embolism is present,Representing the total number of samples in the training set,Representing a sampleContaining the probability of pulmonary embolism.
On the basis of the above embodiment, the area detection network is trained by the following loss function:
Where N represents the number of pulmonary embolism, M is the number of center points of the predicted output region image, Representing the probability that each center point prediction belongs to a pulmonary embolism,AndWeight coefficients respectively representing the weight sparsity of the offset of the center point and the target size loss of the region image,AndRepresenting the predicted and actual center point offsets respectively,AndRepresenting the target size of the predicted region image and the target size of the real region image,Representing the super parameter.
On the basis of the above embodiment, the first feature and the second feature are mixed by:
wherein, the The characteristics of the mixture are represented and,A first characteristic is indicated by the fact that,A second characteristic is indicated by the fact that,Representing a mixture of two feature vectors.
On the basis of the above embodiment, the second feature extractor is trained based on the following minimization loss function:
wherein, the AndRespectively represent samples in training setIs a combination of the nature tag and the location tag of (c),Representing the number of samples screened according to the first threshold,Representing a first characteristic of the screened ith sample,Representing a second characteristic of the screened ith sample,Representing the blended characteristics of the i-th sample that was screened,A second classification network is used to classify the data,Representing a third classification network.
Based on the above embodiment, the second classification network is trained based on the following minimization loss function:
wherein, the Representing the number of samples screened according to the first threshold,Representing samples in a training setIs a property of the tag of (a),A second classification network is represented as such,Representing the mixing characteristics of the screened samples.
On the basis of the above embodiment, the third classification network is trained based on the following minimization loss function:
wherein, the Representing the number of samples screened according to the first threshold,Representing samples in a training setIs provided with a position tag of (a),A third classification network is represented as such,Representing the mixing characteristics of the screened samples.
On the basis of the above embodiment, the first feature extractor is trained based on the following minimization loss function:
wherein, the Representing the number of samples in the training set,Representing the number of samples screened according to the first threshold,Representing a first characteristic of the screened ith sample,Representing a second characteristic of the screened ith sample,、AndRespectively represent samples in training setIs a disease label, a property label and a position label,A first classification network is represented and is shown,A second classification network is represented as such,A third classification network is represented as such,Representing the mixing characteristics of the screened samples.
The pulmonary embolism recognition device based on parameter sharing provided by the embodiment of the invention is contained in the terminal equipment, can be used for executing any pulmonary embolism recognition method based on parameter sharing provided by the embodiment, and has corresponding functions and beneficial effects.
It should be noted that, in the above embodiment of the pulmonary embolism identifying device based on parameter sharing, each unit and module included are only divided according to the functional logic, but not limited to the above division, as long as the corresponding functions can be implemented, and the specific names of the functional units are only for convenience of distinguishing each other, and are not used for limiting the protection scope of the present invention.
Fig. 6 is a schematic structural diagram of a terminal device according to an embodiment of the present invention, where the terminal device is a specific hardware presentation scheme of the foregoing operation device of the intelligent interaction tablet. As shown in fig. 6, the terminal device includes a processor 310, a memory 320, an input device 330, an output device 340 and a communication device 350, where the number of processors 310 in the terminal device may be one or more, and one processor 310 is illustrated in fig. 6, and the processor 310, the memory 320, the input device 330, the output device 340 and the communication device 350 in the terminal device may be connected by a bus or other means, and in fig. 6, the connection is illustrated by a bus.
The memory 320 is a computer readable storage medium, and may be used to store a software program, a computer executable program, and a module, such as program instructions/modules corresponding to the parameter sharing-based pulmonary embolism recognition method in the embodiment of the present invention (for example, the first extraction unit 210, the probability calculation unit 220, the region screenshot unit 230, the second extraction unit 240, the feature fusion unit 250, and the comprehensive judgment unit 260 in the parameter sharing-based pulmonary embolism recognition device). The processor 310 executes various functional applications of the terminal device and data processing by running software programs, instructions and modules stored in the memory 320, i.e. implements the above-described parameter sharing-based pulmonary embolism recognition method.
The memory 320 may mainly include a storage program area which may store an operating system, application programs required for at least one function, and a storage data area which may store data created according to the use of the terminal device, etc. In addition, memory 320 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples, memory 320 may further include memory located remotely from processor 310, which may be connected to the terminal device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input means 330 may be used to receive input numeric or character information and to generate key signal inputs related to user settings and function control of the terminal device. The output device 340 may include a display device such as a display screen.
The terminal equipment comprises the pulmonary embolism recognition device based on parameter sharing, can be used for executing any pulmonary embolism recognition method based on parameter sharing, and has corresponding functions and beneficial effects.
Embodiments of the present application also provide a storage medium containing computer-executable instructions, which when executed by a computer processor, are configured to perform the relevant operations in the parameter sharing-based pulmonary embolism recognition method provided in any embodiment of the present application, and have corresponding functions and advantages.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product.
Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory. The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, etc., such as Read Only Memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
Note that the above is only a preferred embodiment of the present invention and the technical principle applied. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, while the invention has been described in connection with the above embodiments, the invention is not limited to the embodiments, but may be embodied in many other equivalent forms without departing from the spirit or scope of the invention, which is set forth in the following claims.
Claims (7)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110114241.9A CN114818844B (en) | 2021-01-27 | 2021-01-27 | Pulmonary embolism identification device, terminal device and storage medium based on parameter sharing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110114241.9A CN114818844B (en) | 2021-01-27 | 2021-01-27 | Pulmonary embolism identification device, terminal device and storage medium based on parameter sharing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114818844A CN114818844A (en) | 2022-07-29 |
| CN114818844B true CN114818844B (en) | 2025-06-24 |
Family
ID=82523742
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110114241.9A Active CN114818844B (en) | 2021-01-27 | 2021-01-27 | Pulmonary embolism identification device, terminal device and storage medium based on parameter sharing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114818844B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116029967B (en) * | 2022-09-08 | 2024-06-11 | 浙江医准智能科技有限公司 | Pulmonary embolism image processing method, device, equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778856A (en) * | 2016-12-08 | 2017-05-31 | 深圳大学 | A kind of object identification method and device |
| CN111259823A (en) * | 2020-01-19 | 2020-06-09 | 人民中科(山东)智能技术有限公司 | Pornographic image identification method based on convolutional neural network |
| CN112085714A (en) * | 2020-08-31 | 2020-12-15 | 广州视源电子科技股份有限公司 | Pulmonary nodule detection method, model training method, device, equipment and medium |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107092918B (en) * | 2017-03-29 | 2020-10-30 | 太原理工大学 | An Image Retrieval Method Based on Semantic Features and Supervised Hashing |
| US10531807B2 (en) * | 2017-12-20 | 2020-01-14 | International Business Machines Corporation | Automated extraction of echocardiograph measurements from medical images |
| CN108961273B (en) * | 2018-07-03 | 2021-10-29 | 东北大学 | A method and system for segmenting pulmonary arteries and pulmonary veins from CT images |
| US10861183B2 (en) * | 2019-01-31 | 2020-12-08 | StradVision, Inc. | Method and device for short-term path planning of autonomous driving through information fusion by using V2X communication and image processing |
| CN110084237A (en) * | 2019-05-09 | 2019-08-02 | 北京化工大学 | Detection model construction method, detection method and the device of Lung neoplasm |
| CN110766051A (en) * | 2019-09-20 | 2020-02-07 | 四川大学华西医院 | Lung nodule morphological classification method based on neural network |
| CN111325739B (en) * | 2020-02-28 | 2020-12-29 | 推想医疗科技股份有限公司 | Method and device for detecting lung focus and training method of image detection model |
| CN112116004B (en) * | 2020-09-18 | 2021-11-02 | 推想医疗科技股份有限公司 | Focus classification method and device and focus classification model training method |
-
2021
- 2021-01-27 CN CN202110114241.9A patent/CN114818844B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778856A (en) * | 2016-12-08 | 2017-05-31 | 深圳大学 | A kind of object identification method and device |
| CN111259823A (en) * | 2020-01-19 | 2020-06-09 | 人民中科(山东)智能技术有限公司 | Pornographic image identification method based on convolutional neural network |
| CN112085714A (en) * | 2020-08-31 | 2020-12-15 | 广州视源电子科技股份有限公司 | Pulmonary nodule detection method, model training method, device, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114818844A (en) | 2022-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Alghamdi et al. | Deep learning approaches for detecting COVID-19 from chest X-ray images: A survey | |
| CN110807788B (en) | Medical image processing method, device, electronic equipment and computer storage medium | |
| Sander et al. | Automatic segmentation with detection of local segmentation failures in cardiac MRI | |
| US20220198230A1 (en) | Auxiliary detection method and image recognition method for rib fractures based on deep learning | |
| Guo et al. | SCU‐Net: A deep learning method for segmentation and quantification of breast arterial calcifications on mammograms | |
| Tian et al. | Few-shot anomaly detection for polyp frames from colonoscopy | |
| santha Kumar | Automatic segmentation of liver and tumor for CAD of liver | |
| Chen et al. | VertXNet: an ensemble method for vertebral body segmentation and identification from cervical and lumbar spinal X-rays | |
| US20250173874A1 (en) | Method for detecting white matter lesions based on medical image | |
| Behrendt et al. | A systematic approach to deep learning-based nodule detection in chest radiographs | |
| Ahmed et al. | Exploring deep learning and machine learning approaches for brain hemorrhage detection | |
| Wang et al. | False positive reduction in pulmonary nodule classification using 3D texture and edge feature in CT images | |
| Zhang et al. | A fine-grain error map prediction and segmentation quality assessment framework for whole-heart segmentation | |
| Al-Antari et al. | A hybrid segmentation and classification CAD framework for automated myocardial infarction prediction from MRI images | |
| CN114818844B (en) | Pulmonary embolism identification device, terminal device and storage medium based on parameter sharing | |
| He et al. | A two-stage approach solo_GAN for overlapping cervical cell segmentation based on single-cell identification and boundary generation | |
| AU2025230736A1 (en) | Systems and Methods of Generating Medical Concordance Scores | |
| Hao et al. | Automatic sequence-based network for lung diseases detection in chest CT | |
| Dua et al. | Classifying pulmonary embolism cases in chest ct scans using vgg16 and xgboost | |
| Zeng et al. | Recognition of rare antinuclear antibody patterns based on a novel attention-based enhancement framework | |
| Sourget et al. | Mask of truth: model sensitivity to unexpected regions of medical images | |
| CN114187240A (en) | A multi-lesion image processing method, device, electronic device, and computer-readable medium | |
| Mummaneni et al. | A comprehensive study: intracranial aneurysm detection via vgg16-densenet hybrid deep learning on dsa images | |
| Arora et al. | Enhancing teeth segmentation using multifusion deep neural net in panoramic X-ray images | |
| Lenin Marksia et al. | Accurate segmentation of COVID-19 infected regions in lung CT scans with deep learning |
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 |