CN112836744A - Multi-model false positive attenuation disease classification method and device based on CT slices - Google Patents

Multi-model false positive attenuation disease classification method and device based on CT slices Download PDF

Info

Publication number
CN112836744A
CN112836744A CN202110144190.4A CN202110144190A CN112836744A CN 112836744 A CN112836744 A CN 112836744A CN 202110144190 A CN202110144190 A CN 202110144190A CN 112836744 A CN112836744 A CN 112836744A
Authority
CN
China
Prior art keywords
model
false positive
data
screening model
false
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110144190.4A
Other languages
Chinese (zh)
Inventor
杜强
李德轩
郭雨晨
聂方兴
唐超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Xbentury Network Technology Co ltd
Original Assignee
Beijing Xbentury Network Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Xbentury Network Technology Co ltd filed Critical Beijing Xbentury Network Technology Co ltd
Priority to CN202110144190.4A priority Critical patent/CN112836744A/en
Publication of CN112836744A publication Critical patent/CN112836744A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30008—Bone
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30036—Dental; Teeth
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Abstract

The invention discloses a multi-model false positive attenuation disease classification method and a multi-model false positive attenuation disease classification device based on CT slices, wherein the method comprises the following steps: s1, a primary screening model automatically processes an input CT sequence image, namely, a convolutional neural network is used for carrying out feature extraction on the CT sequence image, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result; s2, independently taking out the part with the confidence coefficient lower than the minimum confidence coefficient preset value in the primary screening result to obtain false data; s3, setting parameters of the selected secondary screening model, and selecting a loss function and an optimizer; and S4, adding a label to the false data in the step S2 as a training label, inputting the training label to the selected secondary screening model for training, and carrying out accurate classification to obtain false positive data and true positive data.

Description

Multi-model false positive attenuation disease classification method and device based on CT slices
Technical Field
The invention relates to the technical field of deep learning and medical image processing in the field of artificial intelligence, in particular to a multi-model false positive attenuation disease classification method and device based on CT slices.
Background
At present, the progress of the AI technology is rapid, and particularly, the application in deep learning is changed day by day, for example, in thyroid screening, lung cancer investigation and other cases, the performance of the AI model can reach or even exceed the performance of human beings, which is not secret. The same is true in the medical field.
At present, CT pictures are mainly used in the application of deep learning in medical treatment. When the existing deep learning-based training model is used for checking the CT image, because the resolution ratio of the CT image is relatively low and is only hundreds times hundreds, although the number can reflect many conditions of internal diseases of a human body, the effect of pixels is obviously insufficient compared with millions and millions of pixels of an existing common mobile phone; furthermore, CT images are noisy to visualize. Such as contrast agents that patients need to drink in order to better show images of certain organs within the body, which are not in the body itself and are inherently noisy; in addition, due to brand differences in the quality of the machine or the careless shaking of the body of a patient during scanning on the machine, the generated CT image has interferences such as artifacts, and the fine noise or artifacts can confuse the characteristics of the image, thereby destroying the training and the use of the final deep learning model. When a doctor manually observes a CT image for diagnosis, the doctor observes and judges whether diseases exist or not through a scanned CT image sequence, the diagnosis level of the doctor depends greatly, the diagnosis of the doctor at the normal level is possibly not accurate enough, and the time of the doctor at the high level does not have too much time, so that the doctor often has a difficulty in piling up the famous doctors.
Disclosure of Invention
The invention aims to provide a multi-model false positive attenuation disease classification method and device based on CT slices, and aims to solve the problem that the existing model is not accurate enough in diagnosing some diseases.
The invention provides a multi-model false positive attenuation disease classification method based on CT slices, which comprises the following steps:
s1, inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
s2, independently taking out the part with the confidence coefficient lower than the minimum confidence coefficient preset value in the primary screening result to obtain false data;
s3, selecting a secondary screening model, setting parameters of the selected secondary screening model, and selecting a loss function and an optimizer;
and S4, adding the label of the secondary screening model to the false data in the step S2 as a training label, inputting the training label into the selected secondary screening model for training, and carrying out accurate classification to obtain false positive data and true positive data.
The invention provides a multi-model false positive attenuation disease classification device based on CT slices, which comprises:
a primary screening module: the CT sequence image processing system is used for inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
false positive data selection module: and the part with the confidence coefficient lower than the preset value of the lowest confidence coefficient in the primary screening result is independently taken out to obtain the false data.
A parameter setting module: and the method is used for setting the parameters of the selected secondary screening model and selecting the loss function and the optimizer.
And a secondary screening module: and the secondary screening model label is added to the false data in the step S2 to serve as a training label, and the training label is input to the selected secondary screening model for training and accurate classification to obtain false positive data and true positive data.
The embodiment of the invention also provides a multi-model false positive attenuation disease classification device based on CT slices, which comprises: a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of a method for multi-model false positive attenuation disease classification based on CT slices as described above.
The embodiment of the invention also provides a computer readable storage medium, which stores an implementation program for information transmission, and when the program is executed by a processor, the steps of the multi-model false positive attenuation disease classification method based on CT slices are implemented.
By adopting the embodiment of the invention, the classification model is specifically trained aiming at specific characteristics, and the targeted false positive attenuation model is used for secondary screening, so that the defect of overhigh false positive of the rough classification model is overcome. The false positive attenuation model does not need additional artificial labels, so that the acquisition cost of the artificial labels is saved, and the investment of manpower and material resources is reduced; in addition, the design and training of the false positive attenuation model are much simpler than the model used in one-time screening, and the training of the network is easier, so that the network can be converged more quickly and more efficiently.
The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a multi-model false positive attenuation disease classification method based on CT slices according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a DenseNet50 model used in one embodiment of the present invention;
FIG. 3 is a schematic diagram of a three-dimensional convolution structure in a 3D-ResNet18 model according to a second embodiment of the present invention;
FIG. 4 is a schematic diagram of a CT slice-based multi-model false positive attenuation disease classification device according to a first embodiment of the present invention;
fig. 5 is a schematic diagram of a multi-model false positive attenuation disease classification device based on CT slices according to a second embodiment of the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the following embodiments, and it should be understood that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it is to be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", and the like, indicate orientations and positional relationships based on those shown in the drawings, and are used only for convenience of description and simplicity of description, and do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be considered as limiting the present invention.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise. Furthermore, the terms "mounted," "connected," and "connected" are to be construed broadly and may, for example, be fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Method embodiment
According to an embodiment of the present invention, a multi-model false positive attenuation disease classification method based on CT slices is provided, fig. 1 is a flowchart of the multi-model false positive attenuation disease classification method based on CT slices according to the embodiment of the present invention, as shown in fig. 1, the multi-model false positive attenuation disease classification method based on CT slices according to the embodiment of the present invention specifically includes:
s1, inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
s2, independently taking out the part with the confidence coefficient lower than the minimum confidence coefficient preset value in the primary screening result to obtain false data;
in the results output by the primary screening model, each result corresponds to a position, namely a suspected lesion area, and a score, namely a confidence level, of the position, the confidence level is between 0 and 1, the confidence level can also be understood as the confidence level of the primary screening model, namely the WNet model in the embodiment, on the detected position, the nodule, and the partial data with the confidence value lower than the preset lowest confidence level is regarded as data with extremely uncertain model identification, and needs to be taken out separately as false data to be input into the secondary screening model.
S3, selecting a secondary screening model, setting parameters of the selected secondary screening model, and selecting a loss function and an optimizer;
specifically, the secondary screening model selected in the embodiment of the present invention selects a shallow network such as ResNet18, and the parameters set for the selected secondary screening model specifically include: the learning rate, which in this embodiment is fixedly set to 0.001, BCEloss is set as a loss function of the secondary screening model, and the SGD optimizer is selected as the optimizer of the secondary screening model.
And S4, adding the label of the secondary screening model to the false data in the step S2 as a training label, inputting the training label into the selected secondary screening model for training, and carrying out accurate classification to obtain false positive data and true positive data.
The following three embodiments are used to further explain the specific operation of the multi-model false positive attenuation disease classification method based on CT slices in the method embodiment:
example one
Classified localization of benign and malignant disease on pulmonary nodule detection:
primary screening: after the CT sequence image is input into the WNet model, the model automatically processes the image, namely, the CT sequence image is subjected to feature extraction through a convolutional neural network, a response is generated on a feature region of a suspected nodule, segmentation and positioning are carried out, and then a detected lesion region is output.
The WNet model is a model developed inside a company, the structure of the WNet model is similar to a Unet model + pyramid structure, and the WNet model is similar to the connection of a front neural network and a rear neural network of W, so that feature information in an image can better circulate in different neural network layers.
The reason for the large number of false positives in lung nodule detection is because the nodules have very similar cross-sections in CT images due to their characteristics, such as size, color, shape, and other common features, such as blood vessels, trachea. In a normal CT image of a human body, the number of blood vessels and air ducts is very large. When the model judges the lesion area, if the judgment is based on the attributes of time, size, color, shape and the like, the section of the blood vessel and the trachea can be easily judged to be the lesion area by mistake, and the result output by the screening is required to be screened secondarily.
Secondary screening: the false positive data supplementary labels in the primary screening output result are used as data to be input into a secondary screening model DenseNet or Resnet18, in the embodiment, the model learning rate is fixedly set to be 0.001, the loss function and the optimizer are respectively set to be BCEloss and SGD for false positive attenuation training, the false positive data are screened once again, accurate classification is achieved, and true positive data are output.
The Resnet18 model is a small-scale model, and its input can be small pieces, such as nodules or slices of the vascular trachea, so that it is better to perform fine feature extraction on data with small shape and size than a screening model. The Resnet18 model fits much more closely with the compact data than the WNet model used for one-time screening.
As shown in fig. 2, the DenseNet50 model is a deep learning basic network that is most famous after ResNet, and its specific dense link structure is widely used in industry.
The lung CT sequence image used in the embodiment is provided by Beijing 301 hospital, and the embodiment has better effect by experiments on 638 cases of data in 301 hospital.
Example two
On oral edentulous detection:
primary screening: after the CT sequence image is input into the EfficientDet model, the model automatically carries out processing such as convolution and the like on the image, extracts the characteristics of the image, responds to the characteristic region of suspected missing teeth, and carries out segmentation and positioning, thereby realizing the missing tooth detection.
The EfficientDet model for positioning is a mainstream model in the current detection network, the EfficientDet-d2 model is specifically selected for positioning, the structure of the model comprises a conventional convolution network and a specially-customized BiFPN network, the high-efficiency flow of information in the image can be realized, the model detection effect is good, and the model omission ratio is low in general cases.
The reason for the large number of false positives in missing tooth detection is due to some characteristics of missing teeth on CT images. For example, if the teeth of many people are askew from east to west and the gap is large, the position of the gap can be judged as a missing tooth; or the difference in length is large, so that the model can consider that a part of the short teeth is the missing teeth. There are also many patients who have the outermost of the upper and lower rows of teeth, i.e., the wisdom teeth, and are likely to be determined to be missing because there are no teeth adjacent to the wisdom teeth.
Secondary screening: the false positive data output by the primary screening model are taken out independently, the labels are supplemented, the original true positive labels of the model are added for processing, the data become a 3D data block, the 3D data block is input into the secondary screening model 3D-ResNet18, in the embodiment, the learning rate of the model is initially set to be 0.001 and is continuously decreased along with the iteration times, the loss function and the optimizer are respectively set to carry out false positive attenuation training for BCEloss and SGD, the false positive data are screened once again, and the true positive data are output.
Compared with the primary screening model EfficientDet-D2, the secondary screening model 3D-ResNet18 has the advantages that attention can be focused on the 3D data block, judgment is carried out only according to information in 3D, interference of information of the position of the tooth is avoided, and accuracy is improved.
The 3D-ResNet18 model contains jump links and is a classic model of deep learning, wherein the convolution part is three-dimensional convolution, and the structure is shown in FIG. 3.
The oral CT sequence data of this embodiment is provided by oral hospitals in Jiangsu province, and includes 1100 CT sequences.
Example three
On rib fracture detection:
primary screening: after the CT sequence image is input into the nnUNet model, the model automatically carries out processing such as convolution and the like on the image, extracts the characteristics of the image, responds to the characteristic region of suspected fracture, and carries out segmentation and positioning, thereby realizing the detection of the rib fracture.
The nnUNet model also comprises a large number of settings for automatic parameter search besides the structure of the Unet, including the parameters of the structure of each convolution layer, so that the model can adapt to various different data sets and has strong robustness.
The reason why the rib fracture detection generates a large number of false positives is because of some characteristics of the fracture on the CT image. The fracture is not caused by the fact that one bone is broken and then is split into two parts, mainly the bone is extruded after being subjected to severe impact, so that the bones are overlapped in a staggered mode, the two bones are connected together more closely in the shape, and a plurality of parts are connected in the normal bone structure. Erroneous judgment due to the similarity on the image is generated.
Secondary screening: the false positive data output by the primary screening model are taken out independently, the labels are supplemented, the original true positive labels of the model are added for processing, the data become a 3D data block, the 3D data block is input into the secondary screening model 3D-ResNet18, in the embodiment, the learning rate of the model is initially set to be 0.001 and is continuously decreased along with the iteration times, the loss function and the optimizer are respectively set to carry out false positive attenuation training for BCEloss and SGD, the false positive data are screened once again, and the true positive data are output.
The advantage of the secondary screening model 3D-ResNet18 over the primary screening model nnUNet is that it has a greater focusing power and is less disturbed because there is only one 3D data block input rather than the entire sequence, making it easier to determine whether the data is simply a fractured connection due to compression or a normal connection due to natural bone growth.
The rib fracture data used in this example is provided by the public data set RibFrac, which contains 170 CT sequences. The automatic examination of fracture has very important application in the fields of forensic identification and the like.
In summary, by adopting the embodiments of the present invention, the classification model is specifically trained for specific features. And a targeted false positive attenuation model is used for secondary screening, so that the defect of overhigh false positive of a rough classification model is overcome. The false positive attenuation model does not need additional artificial labels, so that the acquisition cost of the artificial labels is saved, and the investment of manpower and material resources is reduced; in addition, the design and training of the false positive attenuation model are much simpler than the model used in one-time screening, and the training of the network is easier, so that the network can be converged more quickly and more efficiently.
Apparatus embodiment one
According to an embodiment of the present invention, a multi-model false positive attenuation disease classification device based on CT slices is provided, fig. 4 is a schematic diagram of the multi-model false positive attenuation disease classification device based on CT slices according to the embodiment of the present invention, as shown in fig. 4, the multi-model false positive attenuation disease classification device based on CT slices according to the embodiment of the present invention specifically includes:
primary screening module 40: the CT sequence image processing system is used for inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
false positive data selection module 42: and the part with the confidence coefficient lower than the preset value of the lowest confidence coefficient in the primary screening result is independently taken out to obtain the false data.
The parameter setting module 44: and the method is used for setting the parameters of the selected secondary screening model and selecting the loss function and the optimizer.
The parameter setting module is specifically used for setting a learning rate, a loss function and an optimizer for the selected secondary screening model.
Secondary screening module 46: and the secondary screening model label is added to the false data in the step S2 to serve as a training label, and the training label is input to the selected secondary screening model for training and accurate classification to obtain false positive data and true positive data.
The embodiment of the present invention is an apparatus embodiment corresponding to the above method embodiment, and specific operations of each module may be understood with reference to the description of the method embodiment, which is not described herein again.
Device embodiment II
The embodiment of the invention provides a multi-model false positive attenuation disease classification device based on CT slices, as shown in figure 5, comprising: a memory 50, a processor 52 and a computer program stored on the memory 50 and executable on the processor 52, which computer program, when executed by the processor 52, carries out the following method steps:
s1, inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
s2, independently taking out the part with the confidence coefficient lower than the minimum confidence coefficient preset value in the primary screening result to obtain false data;
in the results output by the primary screening model, each result corresponds to a position, namely a suspected lesion area, and a score, namely a confidence level, of the position, the confidence level is between 0 and 1, the confidence level can also be understood as the confidence level of the primary screening model, namely the WNet model in the embodiment, on the detected position, the nodule, and the partial data with the confidence value lower than the preset lowest confidence level is regarded as data with extremely uncertain model identification, and needs to be taken out separately as false data to be input into the secondary screening model.
S3, selecting a secondary screening model, setting parameters of the selected secondary screening model, and selecting a loss function and an optimizer;
specifically, the secondary screening model selected in the embodiment of the present invention is a shallow network model such as ResNet18, and the parameters set for the selected secondary screening model specifically include: the learning rate is fixedly set to be 0.001 in the embodiment of the invention, BCEloss is set as a loss function of the secondary screening model, and an SGD optimizer is selected as an optimizer of the secondary screening model.
And S4, adding the label of the secondary screening model to the false data in the step S2 as a training label, inputting the training label into the selected secondary screening model for training, and carrying out accurate classification to obtain false positive data and true positive data.
Device embodiment III
The embodiment of the present invention provides a computer-readable storage medium, on which an implementation program for information transmission is stored, and when being executed by the processor 52, the implementation program implements the following method steps:
s1, inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
s2, independently taking out the part with the confidence coefficient lower than the minimum confidence coefficient preset value in the primary screening result to obtain false data;
in the results output by the primary screening model, each result corresponds to a position, namely a suspected lesion area, and a score, namely a confidence level, of the position, the confidence level is between 0 and 1, the confidence level can also be understood as the confidence level of the primary screening model, namely the WNet model in the embodiment, on the detected position, the nodule, and the partial data with the confidence value lower than the preset lowest confidence level is regarded as data with extremely uncertain model identification, and needs to be taken out separately as false data to be input into the secondary screening model.
S3, selecting a secondary screening model, setting parameters of the selected secondary screening model, and selecting a loss function and an optimizer;
specifically, the secondary screening model selected in the embodiment of the present invention is a shallow network model such as ResNet18, and the parameters set for the selected secondary screening model specifically include: the learning rate is fixedly set to be 0.001 in the embodiment of the invention, BCEloss is set as a loss function of the secondary screening model, and an SGD optimizer is selected as an optimizer of the secondary screening model.
And S4, adding the label of the secondary screening model to the false data in the step S2 as a training label, inputting the training label into the selected secondary screening model for training, and carrying out accurate classification to obtain false positive data and true positive data.
The computer-readable storage medium of this embodiment includes, but is not limited to: ROM, RAM, magnetic or optical disks, and the like.
It will be apparent to those skilled in the art that the modules or steps of the present invention described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and alternatively, they may be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described may be performed in an order different than that described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple ones of them may be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (9)

1. A multi-model false positive attenuation disease classification method based on CT slices is characterized by comprising the following steps:
s1, inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
s2, independently taking out the part with the confidence coefficient lower than the minimum confidence coefficient preset value in the primary screening result to obtain false data;
s3, selecting a secondary screening model, setting parameters of the selected secondary screening model, and selecting a loss function and an optimizer;
and S4, adding the label of the secondary screening model to the false data in the step S2 as a training label, inputting the training label into the selected secondary screening model for training, and carrying out accurate classification to obtain false positive data and true positive data.
2. The method for classifying diseases based on multi-model false positive fading of CT slices as claimed in claim 1, wherein the step S2 of obtaining false positive data specifically comprises: when the primary screening model result is output, each result corresponds to a position and a confidence coefficient, the confidence coefficient value is between 0 and 1, a numerical value is set as a minimum confidence coefficient preset value, data lower than the minimum confidence coefficient preset value are independently taken out and used as fake data for the next screening.
3. The method for classifying diseases based on multi-model false positive fading of CT slices as claimed in claim 1, wherein in step S3, a shallow network such as ResNet18 is selected as the secondary screening model.
4. The method for classifying diseases based on multi-model false positive attenuation of CT slices as claimed in claim 1, wherein the parameters in step S3 specifically include: the learning rate is set to be 0.001, the loss function is BCEloss, and the optimizer selects an SGD optimizer.
5. A multi-model false positive attenuation disease classification device based on CT slices is characterized by comprising:
a primary screening module: the CT sequence image processing system is used for inputting a CT sequence image into a primary screening model, automatically processing the CT sequence image through the primary screening model, namely performing feature extraction on the CT sequence image by using a convolutional neural network, segmenting and positioning a feature region of a suspected nodule, and outputting a primary screening model result;
false positive data selection module: and the part with the confidence coefficient lower than the preset value of the lowest confidence coefficient in the primary screening result is independently taken out to obtain the false data.
A parameter setting module: and the method is used for setting the parameters of the selected secondary screening model and selecting the loss function and the optimizer.
And a secondary screening module: and the secondary screening model label is added to the false data in the step S2 to serve as a training label, and the training label is input to the selected secondary screening model for training and accurate classification to obtain false positive data and true positive data.
6. The CT-slice-based multi-model false positive attenuation disease classification device as claimed in claim 5, wherein the false positive data selection module is specifically configured to select data with a confidence lower than a minimum confidence preset value as false data for a next screening.
7. The CT-slice-based multi-model false positive fading disease classification device of claim 5, wherein the parameter setting module is specifically configured to set a learning rate, a loss function and an optimizer for a selected shallow network model such as ResNet18, wherein the learning rate is fixedly set to 0.001, the loss function is BCEloss, and the optimizer is an SGD optimizer.
8. A multi-model false positive attenuation disease classification device based on CT slices is characterized by comprising: memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the CT slice-based multi-model false positive attenuation disease classification method according to any of claims 1 to 4.
9. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon an information-transfer implementing program which, when being executed by a processor, implements the steps of the CT-slice-based multi-model false positive attenuation disease classification method according to any one of claims 1 to 4.
CN202110144190.4A 2021-02-02 2021-02-02 Multi-model false positive attenuation disease classification method and device based on CT slices Pending CN112836744A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110144190.4A CN112836744A (en) 2021-02-02 2021-02-02 Multi-model false positive attenuation disease classification method and device based on CT slices

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110144190.4A CN112836744A (en) 2021-02-02 2021-02-02 Multi-model false positive attenuation disease classification method and device based on CT slices

Publications (1)

Publication Number Publication Date
CN112836744A true CN112836744A (en) 2021-05-25

Family

ID=75931678

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110144190.4A Pending CN112836744A (en) 2021-02-02 2021-02-02 Multi-model false positive attenuation disease classification method and device based on CT slices

Country Status (1)

Country Link
CN (1) CN112836744A (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160335478A1 (en) * 2014-01-28 2016-11-17 Ventana Medical Systems, Inc. Adaptive classification for whole slide tissue segmentation
CN110458107A (en) * 2019-08-13 2019-11-15 北京百度网讯科技有限公司 Method and apparatus for image recognition
CN110489550A (en) * 2019-07-16 2019-11-22 招联消费金融有限公司 File classification method, device and computer equipment based on combination neural net
CN111369530A (en) * 2020-03-04 2020-07-03 浙江明峰智能医疗科技有限公司 CT image pulmonary nodule rapid screening method based on deep learning
CN111415352A (en) * 2020-04-03 2020-07-14 南京信息工程大学 Cancer metastasis panoramic pathological section analysis method based on deep cascade network
CN111461243A (en) * 2020-04-08 2020-07-28 中国医学科学院肿瘤医院 Classification method, classification device, electronic equipment and computer-readable storage medium
CN112116603A (en) * 2020-09-14 2020-12-22 中国科学院大学宁波华美医院 Pulmonary nodule false positive screening method based on multitask learning
CN112241948A (en) * 2020-09-23 2021-01-19 深圳视见医疗科技有限公司 Lung nodule detection and analysis method and system capable of adapting to thickness of layer

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160335478A1 (en) * 2014-01-28 2016-11-17 Ventana Medical Systems, Inc. Adaptive classification for whole slide tissue segmentation
CN110489550A (en) * 2019-07-16 2019-11-22 招联消费金融有限公司 File classification method, device and computer equipment based on combination neural net
CN110458107A (en) * 2019-08-13 2019-11-15 北京百度网讯科技有限公司 Method and apparatus for image recognition
CN111369530A (en) * 2020-03-04 2020-07-03 浙江明峰智能医疗科技有限公司 CT image pulmonary nodule rapid screening method based on deep learning
CN111415352A (en) * 2020-04-03 2020-07-14 南京信息工程大学 Cancer metastasis panoramic pathological section analysis method based on deep cascade network
CN111461243A (en) * 2020-04-08 2020-07-28 中国医学科学院肿瘤医院 Classification method, classification device, electronic equipment and computer-readable storage medium
CN112116603A (en) * 2020-09-14 2020-12-22 中国科学院大学宁波华美医院 Pulmonary nodule false positive screening method based on multitask learning
CN112241948A (en) * 2020-09-23 2021-01-19 深圳视见医疗科技有限公司 Lung nodule detection and analysis method and system capable of adapting to thickness of layer

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ARIA PEZESHK 等: "3-D Convolutional Neural Networks for Automatic Detection of Pulmonary Nodules in Chest CT", 《IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS》, vol. 23, no. 5, pages 2080 - 2090, XP011743833, DOI: 10.1109/JBHI.2018.2879449 *
HAO TANG 等: "AUTOMATED PULMONARY NODULE DETECTION USING 3D DEEP CONVOLUTIONAL NEURAL NETWORKS", 《2018 IEEE 15TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING》, pages 523 - 526 *
宋德娟;张承明;杨晓霞;李峰;韩颖娟;高帅;董海燕;: "高分二号遥感影像提取冬小麦空间分布", 《遥感学报》, vol. 24, no. 05, pages 596 - 608 *

Similar Documents

Publication Publication Date Title
CN111507381B (en) Image recognition method, related device and equipment
CN110796613B (en) Automatic identification method and device for image artifacts
JP4303598B2 (en) Pixel coding method, image processing method, and image processing method for qualitative recognition of an object reproduced by one or more pixels
JP5643304B2 (en) Computer-aided lung nodule detection system and method and chest image segmentation system and method in chest tomosynthesis imaging
CN112292691A (en) Methods and systems for improving cancer detection using deep learning
CN110506278A (en) Target detection in latent space
CN105678746B (en) The localization method and device of liver scope in a kind of medical image
JPH1156828A (en) Abnormal shadow candidate detecting method and its device
US9042611B2 (en) Automated vascular region separation in medical imaging
JP5364290B2 (en) Image display apparatus, image display control method, and program
US8385614B2 (en) Slice image display apparatus, method and recording-medium having stored therein program
EP3973539A1 (en) System and method for interpretation of multiple medical images using deep learning
US20210217166A1 (en) Automated screening of medical data
JP2006034585A (en) Picture display device and picture display method, and program thereof
CN112036506A (en) Image recognition method and related device and equipment
CN111904379A (en) Scanning method and device of multi-modal medical equipment
JP2014221163A (en) Image diagnosis support device and support method
CN112820399A (en) Method and device for automatically diagnosing benign and malignant thyroid nodules
CN110706236A (en) Three-dimensional reconstruction method and device of blood vessel image
CN111462139A (en) Medical image display method, medical image display device, computer equipment and readable storage medium
KR102257904B1 (en) Method for predicting pathological conditions and device for diagnosing pathological conditions using the same
CN112288752B (en) Full-automatic coronary calcified focus segmentation method based on chest flat scan CT
CN114202516A (en) Foreign matter detection method and device, electronic equipment and storage medium
US11475568B2 (en) Method for controlling display of abnormality in chest x-ray image, storage medium, abnormality display control apparatus, and server apparatus
CN113034522A (en) CT image segmentation method based on artificial neural network

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