CN109727251A - The system that lung conditions are divided a kind of quantitatively, method and apparatus - Google Patents
The system that lung conditions are divided a kind of quantitatively, method and apparatus Download PDFInfo
- Publication number
- CN109727251A CN109727251A CN201811632060.XA CN201811632060A CN109727251A CN 109727251 A CN109727251 A CN 109727251A CN 201811632060 A CN201811632060 A CN 201811632060A CN 109727251 A CN109727251 A CN 109727251A
- Authority
- CN
- China
- Prior art keywords
- lung
- segmentation
- lobe
- conditions
- lung conditions
- 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
Links
Landscapes
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
This disclosure relates to a kind of system of quantitative lung conditions segmentation, including lung conditions parted pattern, it is configured to carry out lung conditions segmentation to the rabat image based on neural network；Lobe of the lung parted pattern is configured to carry out lobe of the lung segmentation to the rabat image based on neural network；And computation module, it is configured to calculate pneumothorax volume based on the result of lung conditions segmentation and lobe of the lung segmentation.Present disclosure also relates to quantify method, equipment and the computer-readable medium of lung conditions segmentation accordingly.
Description
Technical field
The present disclosure relates generally to Medical Image Processings, more particularly to realize quantitative lung conditions segmentation and inspection based on technology
It surveys.
Background technique
Common lung conditions include pneumothorax, and wet lung etc. can seriously endanger human health.Pneumothorax (pneumothorax)
Refer to that gas enters pleural cavity, causes the state of lung's pneumatosis.Pneumothorax generally can be divided into traumatic pneumothorax, spontaneous pneumothorax and people
Work pneumothorax etc..Traumatic pneumothorax is generally caused by the wall of the chest or pulmonary trauma, and spontaneous pneumothorax is then since pulmonary disease causes
Lung tissue, which voluntarily ruptures, to be caused, and artificial pneumothorax then refers to air is actively injected into pleural cavity during treatment and diagnosis
In.Spontaneous pneumothorax is more common in that male is between twenty and fifty or suffer with chronic bronchitis, pulmonary emphysema and pulmonary tuberculosis etc., this disease belongs to
One of lung section acute disease, serious person can critical life.Wet lung is medically generally termed " pleural effusion ", it is generally felt by lung
Hair dyeing is scorching or autoimmune disease causes, and then accumulation outside lungs is caused to have large quantity of moisture.X-ray chest radiograph inspection is to cure at present
To the main detection methods of the pulmonary diseases such as pneumothorax and wet lung on.
There are specific pneumothorax line, i.e. gas boundary line in atrophy lung tissue and pleural cavity on pneumothorax rabat mostly, generally
In evagination lines shadow.It is the transparent area of no lung marking outside pneumothorax line, is the lung tissue of compression in line.And wet lung is in x-ray chest radiograph
On reflection be easier to observe relative to pneumothorax.
Traditional quantitative pulmonary disease segmentation and detection are the routine works of x-ray doctor, all the time, doctor according to
Range estimation is carried out by pen, ruler and X-ray film to lobe of the lung profile and with profile where the illnesss such as pneumothorax or wet lung
Measurement and calculating.On this basis, some traditional machine learning methods based on active shape model (ASM) also can be to this
Certain help is played in the segmentation of class medical image, but since computational efficiency and segmentation accuracy are lower, cannot achieve and face
Bed application.
In recent years, the Image Segmentation Model based on machine learning is constantly suggested, but it is directed to common on x-ray chest radiograph
The Medical Image Segmentation Techniques of the illnesss such as pulmonary disease still have blind area.Due to lung conditions (e.g., including pneumothorax and lung product
Water) particularity, it is desirable to really realize quantitative lung conditions segmentation and detection, need a large amount of data and iteration tests.
Therefore, this field needs to realize the skill of quantitative the lung conditions segmentation and detection to rabat based on nerual network technique
Art, to accelerate diagnosis of the radiologist for lung conditions case, and quantifying by the lung conditions volume to rabat
Analysis, for the rapid screening of case, the monitoring of patient condition's parameter provides auxiliary.
Summary of the invention
A kind of system for relating in one aspect to quantitative lung conditions segmentation of the disclosure, including lung conditions parted pattern, match
It is set to and lung conditions segmentation is carried out to rabat image based on neural network；Lobe of the lung parted pattern is configured to based on neural network pair
The rabat image carries out lobe of the lung segmentation；And computation module, it is configured to based on lung conditions segmentation and the lobe of the lung point
The result cut calculates lung conditions volume.
It is exemplary according to one and nonlimiting examples, the system may also include pre-processing assembly, it is configured in the lung
Before the segmentation of portion's illness and lobe of the lung segmentation, the rabat image is pre-processed, the pretreatment includes with the next item down
Or multinomial combination: size adjustment cuts, rotation, normalizes and standardize.
The nonlimiting examples according to another exemplary, computation module are further configured to through the lung conditions
Divide lobe of the lung pixel quantity that obtained lung conditions pixel quantity and the lobe of the lung are divided to calculate the lung
Illness volume.
The nonlimiting examples according to another exemplary, computation module are further configured to based on the lung conditions
Volume alerts to issue classification.
The other aspects of the disclosure further relate to accordingly quantitatively method, the equipment and computer-readable of lung conditions segmentation
Medium.
Detailed description of the invention
Fig. 1, which is shown, quantifies lung conditions point according to the realizing based on nerual network technique for an illustrative aspect of the disclosure
Cut the block diagram with detection system.
Divide the side with detection Fig. 2 shows the quantitative lung conditions based on nerual network technique according to disclosure one side
The block diagram of method.
Fig. 3 shows the training according to disclosure one side based on the side of the lung conditions parted pattern of nerual network technique
The block diagram of method.
Fig. 4 shows the quantitative lung conditions segmentation based on nerual network technique according to one exemplary embodiment of the disclosure
With the result example of detection.
Specific embodiment
Neural network is one of machine learning and the recent tendency of artificial intelligence study.Neural network method is computer view
Feel and machine learning brings revolutionary progress.Branch of the neural network as machine learning, is by using comprising complexity
Structure or the multiple process layers being made of multiple nonlinear transformation carry out the algorithm of higher level of abstraction to data.
Fig. 1, which is shown, quantifies lung conditions point according to the realizing based on nerual network technique for an illustrative aspect of the disclosure
Cut the block diagram with detection system 100.System 100 may include processor 102 and memory 104.It is single that system 100 may also include input
Member 110, parted pattern 112 and output unit 114.System 100 may also include pre-processing assembly 116, training module 118, lung
Portion's illness computing module 120.Above-mentioned various assemblies can be communicated with one another by such as 106 coupling merga pass bus 106 of bus.
Input unit 110 can obtain initial data.Obtaining initial data may include inputting x-ray chest radiograph data from external source.
The nonlimiting examples according to another exemplary, obtaining initial data may also comprise directly acquisition x-ray chest radiograph data.Original number
According to may include case with pneumothorax or wet lung or other lung conditions and with pneumothorax or wet lung or other lungs
The case of illness.
Pre-processing assembly 116 can pre-process initial data.And nonlimiting examples exemplary according to one, to original
It may include carrying out scaled, trimming and/or rotation or above-mentioned any combination to image that beginning data, which carry out pretreatment,.For example,
To image carry out scaled may include scale the images to parted pattern being capable of received size.For another example, image is carried out
Trimming may include being trimmed using random trimming strategy to image.For another example, carrying out rotation to image may include using Random-Rotation
Strategy rotates image.Such as institute it is found that the present invention is not limited by the every pretreated order of progress.Due to original X-ray
The contrast of rabat image may be lower, therefore the nonlimiting examples according to another exemplary, and pretreatment may also include pair
Operation and/or normalizing operation is normalized in image.The scheme of the disclosure can carry out above description and/or well known in the art
Various other pretreated various combinations.The nonlimiting examples according to another exemplary, for being intended as training image
The pretreatment of the initial data of data may also include artificial mark.For example, artificial mark may include marking out lung conditions (example
Such as, pneumothorax, wet lung etc.) and the lobe of the lung etc..Annotation results can be known as covering by the nonlimiting examples according to another exemplary
Film.Exposure mask can be used as goldstandard in supervised learning.
Parted pattern 112 can carry out the segmentation of lung conditions and the lobe of the lung.According to an exemplary and non-limiting realization, divide
Cutting model 112 can further comprise lung conditions segmentation submodel 122 and lobe of the lung segmentation submodel 124.In this case, it can incite somebody to action
Pretreated image data is respectively supplied to lung conditions segmentation submodel and lobe of the lung segmentation submodel, and by lung conditions point
Cut the segmentation that submodel carries out the segmentation of lung conditions and carries out the lobe of the lung by lobe of the lung segmentation submodel.It is exemplary according to one rather than
Limited embodiment, parted pattern 112 can use full convolutional network, v-net, u-net, link-net etc., but this public affairs
It opens and is not limited to this.For different illnesss, different neural network models can be used (for example, full convolutional network, v-
Net, u-net, link-net etc.) to reach ideal pneumothorax segmentation effect.And nonlimiting examples, lung exemplary according to one
Portion's illness segmentation submodel and lobe of the lung segmentation submodel can be identical or different neural network structure, such as be respectively
Two kinds in v-net, u-net, link-net etc., or be it is same, or any combination thereof.Moreover, for difference
Lung conditions (for example, pneumothorax, wet lung etc.), can also have different models.For example, lung conditions divide submodel 122
It may further include pneumothorax segmentation submodel, wet lung segmentation submodel etc..Every kind of different lung conditions divide submodule
Type is also possible to identical or different neural network structure, for example, be respectively in v-net, u-net, link-net etc. two
Kind, or be it is same, or any combination thereof.
Training module 118 can train the parted pattern based on nerual network technique to carry out lung conditions based on training data
With the segmentation of the lobe of the lung.For example, including lung conditions segmentation submodel 122 and lobe of the lung segmentation submodel 124 in parted pattern 112
In embodiment, lung conditions segmentation submodel 122 and lobe of the lung segmentation submodel 124 can be respectively trained based on training data.Lung
Portion's illness divides submodel 122 and lobe of the lung segmentation submodel 124 can be by joint training, or can be respectively trained.Training may include
Supervised learning, semi-supervised learning, and/or unsupervised learning.For different lung conditions (for example, pneumothorax, wet lung etc.),
Different models can be trained.For example, pneumothorax can be trained to divide submodel for pneumothorax illness, and can be instructed for wet lung
Practice wet lung segmentation submodel etc..
Lung conditions computing module 120 can calculate lung conditions based on lung conditions segmentation result and lobe of the lung segmentation result
Volume.And nonlimiting examples exemplary according to one, calculating lung conditions volume may include the lung's disease obtained based on segmentation
Disease pixel/voxel quantity and the obtained lobe of the lung pixel/voxel quantity of segmentation calculate lung conditions volume.For example, depending on
Targeted lung conditions, lung conditions computing module 120 can correspondingly calculate pneumothorax volume, wet lung volume etc..Lung
Illness computing module 120 can also judge the presence or absence and classification warning of particular condition based on threshold value.For example, threshold value can wrap
Include the respective threshold based on various lung conditions.For example, threshold value may include for pneumothorax volume, for wet lung volume,
And the volume etc. for other illnesss.The nonlimiting examples according to further exemplary, for each specific disease
Disease may also include one group of threshold value.By being compared with each threshold value among one group of threshold value, it may be determined that the journey of corresponding illness
Degree and classification warning.For example, the lobe of the lung pixel that pneumothorax pixel/voxel quantity and segmentation obtain in pneumothorax illness segmentation result/
The ratio of number of voxel is that a is determined as highly dangerous type pneumothorax as a>40%, and as a<10%, being determined as can self-cure type gas
Chest.
The exportable segmentation result of output unit 114, the lung conditions volume being calculated and to be also possible to output associated
Information.For example, associated information may include such as presence or absence of specified lung illness, point based on lung conditions volume
Grade warning etc., to be used for computer-aided diagnosis and treatment.
Above-mentioned input unit 110, parted pattern 112 and output unit 114.System 100 may also include pre-processing assembly
116, training module 118, lung conditions computing module 120 and its corresponding sub-component etc. can be realized by various modes.
For example, it is exemplary according to one and nonlimiting examples, said modules can be realized by hardware, including for example pass through general place
Manage device, digital signal processor (DSP), specific integrated circuit (ASIC), field programmable gate array (FPGA), discrete electricity
Road, discrete logical device etc. are realized.Said modules can also be realized by software or firmware.For example, said modules can quilt
It is stored in memory 104 and is executed by processor 102 to realize the respective function of each component.Said modules can also be with
It is realized using the various combinations of software described above, firmware and/or hardware.
Divide the side with detection Fig. 2 shows the quantitative lung conditions based on nerual network technique according to disclosure one side
The block diagram of method 200.Method 200 may include obtaining initial data in frame 202.And nonlimiting examples exemplary according to one, are obtained
Taking initial data may include inputting x-ray chest radiograph data from external source.The nonlimiting examples according to another exemplary obtain former
Beginning data may also comprise directly acquisition x-ray chest radiograph.Obtaining initial data can be for example, by above in conjunction with the input unit of Fig. 1 description
110 execute.
After frame 202 obtains initial data, method 200 may include pre-processing in frame 204 to initial data.
And nonlimiting examples exemplary according to one, carrying out pretreatment to initial data may include carrying out scaled to image, repairing
It cuts and/or rotates or above-mentioned any combination.For example, carrying out scaled to image may include scaling the images to segmentation mould
Type being capable of received size.For another example, carrying out trimming to image may include being trimmed using random trimming strategy to image.Again
Such as, carrying out rotation to image may include being rotated with Random-Rotation strategy to image.Such as institute it is found that the present invention is not carried out respectively
The pretreated order of item is limited.Since the contrast of original x-ray chest radiograph image may be lower, according to another exemplary
Nonlimiting examples, pretreatment, which may also include, is normalized operation and/or normalizing operation to image.To image data
Pretreatment can be executed for example, by the pre-processing assembly 116 described above in conjunction with Fig. 1.More than the scheme of the disclosure can carry out
Description and/or various other pretreated various combinations well known in the art.
After frame 204 pre-processes image data, method 200 may include will be pretreated in frame 206
Image data is supplied to housebroken parted pattern to carry out the segmentation of lung conditions and the lobe of the lung.For example, including in parted pattern
Lung conditions are divided in the embodiment of submodel and lobe of the lung segmentation submodel, pretreated image data are supplied to trained
Parted pattern with the segmentation for carrying out lung conditions and the lobe of the lung may include that pretreated image data is respectively supplied to lung
Illness divides submodel and the lobe of the lung and divides submodel, and by lung conditions segmentation submodel carry out lung conditions segmentation and by
The lobe of the lung divides the segmentation that submodel carries out the lobe of the lung.Parted pattern can for example by above in conjunction with Fig. 1 describe parted pattern 112 come
It executes.For different illnesss, can be used different neural network models (for example, full convolutional network, v-net, u-net,
Link-net etc.) to reach ideal pneumothorax segmentation effect.
After having carried out the segmentation of lung conditions and the lobe of the lung in frame 206, method 200 may include being based on lung in frame 208
Illness segmentation result and lobe of the lung segmentation result calculate lung conditions volume.And nonlimiting examples exemplary according to one, meter
Calculating lung conditions volume may include the lung conditions pixel/voxel quantity obtained based on segmentation and the lobe of the lung picture that segmentation obtains
Element/number of voxel calculates lung conditions volume.Calculating for lung conditions volume can be for example by above in conjunction with the lung of Fig. 1 description
Portion's illness computing module 120 executes.For example, depending on targeted lung conditions, pneumothorax volume, lung product can be correspondingly calculated
Water capacity etc..The presence or absence and classification warning of particular condition can be also judged based on threshold value.For example, threshold value may include base
In the respective threshold of various lung conditions.For example, threshold value may include for pneumothorax volume, for wet lung volume and
For the volume etc. of other illnesss.The nonlimiting examples according to further exemplary, for each particular condition, also
It may include one group of threshold value.By being compared with each threshold value among one group of threshold value, it may be determined that the degree of corresponding illness and
Classification warning.For example, the lobe of the lung pixel/voxel that pneumothorax pixel/voxel quantity and segmentation obtain in pneumothorax illness segmentation result
The ratio of quantity is that a is determined as highly dangerous type pneumothorax as a>40%, and as a<10%, being determined as can self-cure type pneumothorax.
After frame 208 calculates lung conditions volume, method 200 may include the lung conditions volume that output is calculated and go back
Associated information may be exported.For example, associated information may include such as presence or absence of specified lung illness, based on lung
Classification warning of portion's illness volume etc., to be used for computer-aided diagnosis and treatment.Export lung conditions volume and associated
Information can be executed for example, by the output unit 114 described above in conjunction with Fig. 1.
Fig. 3 shows the training according to disclosure one side based on the side of the lung conditions parted pattern of nerual network technique
The block diagram of method 300.Method 300 may include that training image data are supplied to parted pattern in frame 302.Training image data can
For example by choosing at least part in acquired initial data (for example, x-ray chest radiograph) data and being located in advance to it
Reason is to obtain.Training image data may include suffering from the case of pneumothorax or wet lung or other lung conditions and not with gas
The case of chest or wet lung or other lung conditions.To the initial data for being intended as training image data, before being carried out to it
The pretreatment stated, such as scaled, trimming and/or rotation or above-mentioned any combination can be carried out to image.For example, to figure
As carry out scaled may include scale the images to parted pattern being capable of received size.For another example, image is trimmed
It may include being trimmed using random trimming strategy to image.For another example, carrying out rotation to image may include with Random-Rotation strategy
Image is rotated.Such as institute it is found that the present invention is not limited by the every pretreated order of progress.Due to pair of initial data
May be lower than degree, therefore the nonlimiting examples according to another exemplary, pretreatment, which may also include, carries out normalizing to image
Change operation and/or normalizing operation.Pretreatment for the initial data for being intended as training image data may also include manually
Mark.For example, artificial mark may include marking out corresponding lung conditions and lobe of the lung etc..It is non-limiting according to another exemplary
Annotation results can be known as exposure mask by embodiment.Exposure mask can be used as goldstandard in supervised learning.To being intended as trained figure
As image data carry out pretreatment can by using above in conjunction with Fig. 1 described in pre-processing assembly 116 execute.There is mark
Training data can also be obtained directly from external source, such as be obtained via the input unit 110 described above in conjunction with Fig. 1.
After training image data are supplied to parted pattern by frame 302, method 300 may include being based on instruction in frame 304
Practice data to train the parted pattern based on nerual network technique to carry out the segmentation of lung conditions and the lobe of the lung.For example, in segmentation mould
Type includes that lung can be respectively trained based on training data in the embodiment of lung conditions segmentation submodel and lobe of the lung segmentation submodel
Portion's illness divides submodel and the lobe of the lung divides submodel.Lung conditions, which divide submodel and lobe of the lung segmentation submodel, to be instructed by joint
Practice, or can be respectively trained.According to further embodiments, for different lung conditions (for example, pneumothorax, wet lung etc.),
Different models can be trained.For example, pneumothorax can be trained to divide submodel for pneumothorax illness, and can be instructed for wet lung
Practice wet lung segmentation submodel etc..The parted pattern based on nerual network technique is trained to carry out lung's disease based on training data
The segmentation of disease and the lobe of the lung can be executed by the training module 118 for example described with reference to Fig. 1.
When training parted pattern based on nerual network technique to carry out lung conditions and lung based on training data in frame 304
After the segmentation of leaf, method 300 may include the lung conditions exported parted pattern in frame 306 and lobe of the lung segmentation result with it is right
The goldstandard through marking training image data answered makes comparisons and calculates loss.For example, including lung conditions point in parted pattern
In the embodiment for cutting submodel and lobe of the lung segmentation submodel, the loss and lobe of the lung segmentation of lung conditions segmentation result can be calculated separately
As a result loss.Calculating loss can for example be executed by the training module 118 described in conjunction with Fig. 1.
After frame 306 calculates loss, method 300 may include determining whether parted pattern can put into frame 308
It uses.Whether determine parted pattern and can come into operation may include for example determining whether the calculated loss of institute is sufficiently small.For example,
And nonlimiting examples exemplary according to one can determine point of lung conditions segmentation submodel and lobe of the lung segmentation submodel respectively
Whether the loss for cutting result is less than or equal to threshold value.If so, method 300 may include in frame 310 by housebroken parted pattern
It comes into operation.If it is not, then method 300 can go to frame 312.
In frame 312, method 300 may include that will lose backpropagation to update the weight of parted pattern.For example, in segmentation mould
Type includes that can divide calculated lung conditions in the embodiment of lung conditions segmentation submodel and lobe of the lung segmentation submodel and tie
The loss of fruit propagates backward to lung conditions segmentation submodel to update its weight, and the calculated lobe of the lung is divided submodel
Loss propagates backward to lobe of the lung segmentation submodel to update its weight.Backpropagation will be lost can be for example by combining Fig. 1 description
Training module 118 executes.
After 312 backpropagation of frame loss, method 300 may include returning to frame 304 with further progress training, calculating
And update, until the parted pattern trained can come into operation.
Fig. 4 shows the quantitative lung conditions segmentation based on nerual network technique according to one exemplary embodiment of the disclosure
With the result example 400 of detection.As shown in the example 1 in Fig. 4, left side is original x-ray image, and centre is the goldstandard of mark,
And the right is then the lobe of the lung segmentation result obtained using the technology of the disclosure.Similarly, as shown in the example 2 in Fig. 4, left side is
Original x-ray image, centre is the goldstandard of mark, and the right is then the lobe of the lung segmentation result obtained using the technology of the disclosure.
As it can be seen that the technology of the disclosure can understand and relatively accurately mark off lung conditions line, i.e., by lung caused by disease
The compressed practical lung outlines line in portion.
Nerual network technique of the application based on image area realizes the quantitative lung conditions in chest film inspection (for example, gas
Chest, wet lung ...) divide and detects.The technology makes doctor when diagnosing patient lungs' ongoing disease degree, can be direct
The lung conditions volume quantitative in x-ray chest radiograph is obtained as a result, accelerating diagnosis speed, and facilitate quantitative information typing patient
Information system.
The application uses mind on the basis of great amount of samples as full automatic quantitative lung conditions segmentation detection method
Model training is split through network technology.The parted pattern energy sufficiently learning sample feature of the application, obtains high accuracy
The segmentation index of Testing index and degree of precision, so as to realize the lung conditions of high accuracy to common x-ray chest radiograph image
Detection, and lung conditions position and the shape segmentations of degree of precision can be realized to the case with lung conditions, and
Realize automatic calculating to lung conditions volume on the basis of segmentation, and can by connecting computer (PACS) system of X-ray equipment,
Realize the intelligent pre-sifted, read tablet auxiliary and/or the record function of diagnosis report etc. of x-ray chest radiograph disease.Although more than the disclosure
It is to be described in conjunction with pneumothorax and wet lung illness, but the technology of the disclosure can be applied equally to other lung conditions
Segmentation and detection, or even the segmentation and detection of other illnesss other than lung conditions can be applied to.
It will be recognized by one of ordinary skill in the art that the beneficial effect of the disclosure is not by any single embodiment Lai all real
It is existing.Various combinations, modification and replacement are that those of ordinary skill in the art are illustrated on the basis of the disclosure.
In addition, term "or" is intended to indicate that inclusive "or" and nonexcludability "or".That is, unless otherwise specified or from upper and lower
Text can be clearly seen, otherwise phrase " X " use " A " or " B " be intended to indicate that it is any naturally can and arrangement.That is, phrase " X " is adopted
Met with " A " or " B " by example any in following instance: X uses A；X uses B；Or X uses both A and B.Belong to
" connection " can indicate identical meanings with " coupling ", indicate the electrical connection of two devices.In addition, disclosure and the accompanying claims book
Used in the article " one " and " certain " be generally to be understood as indicating " one or more ", unless stated otherwise or can be from upper
It is hereinafter apparent from and refers to singular.
Various aspects or feature by by may include several equipment, component, module, and the like system in the form of be in
It is existing.It should be understood that and understand, various systems may include optional equipment, component, module etc., and/or can not include in conjunction with attached drawing
Armamentarium, component, module for being discussed etc..Also the combination of these methods can be used.
It can be with general in conjunction with various illustrative logicals, logical block, module and the circuit that presently disclosed embodiment describes
Processor, digital signal processor (DSP), specific integrated circuit (ASIC), field programmable gate array (FPGA) or it is other can
Programmed logic device, discrete door or transistor logic, discrete hardware component or its be designed to carry out function described herein
Any combination realize or execute.General processor can be microprocessor, but in alternative, and processor, which can be, appoints
What conventional processor, controller, microcontroller or state machine.Processor is also implemented as calculating the combination of equipment, example
As DSP and the combination of microprocessor, multi-microprocessor, the one or more microprocessors cooperateed with DSP core or it is any its
Its such configuration.In addition, at least one processor may include may act on execute one or more steps described above and/or
One or more modules of movement.For example, above can be by processor and being coupled in conjunction with the embodiment of each method description
The memory of processor realizes, wherein the processor can be configured to execute any step of aforementioned any method or its is any
Combination.
In addition, the method that describes in conjunction with aspect disclosed herein or the step of algorithm and/or movement can be directly hard
Implement in part, in the software module executed by processor or in combination of the two.For example, combining each method above
The embodiment of description can realize by being stored with the computer-readable medium of computer program code, wherein the computer journey
Sequence code executes any step or any combination thereof of aforementioned any method when being executed by processor/computer.
The element of the various aspects described in the whole text in the disclosure is that those of ordinary skill in the art are currently or hereafter known
Equivalent scheme in all structures and functionally is clearly included in this by citation, and is intended to be intended to be encompassed by the claims.
In addition, any content disclosed herein is all not intended to contribute to the public --- it is no matter such open whether in claim
It is explicitly recited in book.
Claims (10)
1. a kind of system of quantitative lung conditions segmentation characterized by comprising
Lung conditions parted pattern is configured to carry out lung conditions segmentation to rabat image based on neural network；
Lobe of the lung parted pattern is configured to carry out lobe of the lung segmentation to the rabat image based on neural network；And
Computation module is configured to be divided based on the lung conditions and the result of lobe of the lung segmentation is held to calculate lung conditions
Product.
2. the system as claimed in claim 1, which is characterized in that further include:
Pre-processing assembly is configured to divide in the lung conditions with before lobe of the lung segmentation, carry out to the rabat image
Pretreatment, the pretreatment include following one or more combinations: size adjustment, cutting, rotation, normalization and standardization.
3. the system as claimed in claim 1, which is characterized in that the computation module is further configured to through the lung
The lobe of the lung pixel quantity that the lung conditions pixel quantity and the lobe of the lung that illness is divided are divided is come described to calculate
Lung conditions volume.
4. the system as claimed in claim 1, which is characterized in that the computation module is further configured to based on the lung
Illness volume alerts to issue classification.
5. a kind of method of quantitative lung conditions segmentation characterized by comprising
Lung conditions segmentation is carried out to rabat image using lung conditions parted pattern neural network based；
Lobe of the lung segmentation is carried out to the rabat image using lobe of the lung parted pattern neural network based；And it is based on the lung
Illness, which is divided with the result of lobe of the lung segmentation, calculates lung conditions volume.
6. method as claimed in claim 5, which is characterized in that further include:
Before lung conditions segmentation and lobe of the lung segmentation, the rabat image is pre-processed, the pretreatment
It includes at least one of the following: size adjustment, cuts, rotation, normalizes and standardize.
7. method as claimed in claim 5, which is characterized in that the knot based on lung conditions segmentation and lobe of the lung segmentation
Fruit further comprises to calculate lung conditions volume:
The lobe of the lung pixel that the lung conditions pixel quantity and the lobe of the lung divided by the lung conditions are divided
Quantity calculates the lung conditions volume.
8. a kind of equipment for the segmentation of quantitative lung conditions characterized by comprising
Memory；And
It is coupled to the processor of the memory, the processor is configured to:
Lung conditions segmentation is carried out to rabat image using lung conditions parted pattern neural network based；
Lobe of the lung segmentation is carried out to the rabat image using lobe of the lung parted pattern neural network based；And
Lung conditions volume is calculated based on the result of lung conditions segmentation and lobe of the lung segmentation.
9. a kind of device for the segmentation of quantitative lung conditions characterized by comprising
For using lung conditions parted pattern neural network based to carry out the device of lung conditions segmentation to rabat image；
For using lobe of the lung parted pattern neural network based to carry out the device of lobe of the lung segmentation to the rabat image；And
For calculating the device of lung conditions volume based on the result of lung conditions segmentation and lobe of the lung segmentation.
10. a kind of computer-readable medium for being stored with processor-executable instruction, the processor-executable instruction by
When managing device execution, make the processor:
Lung conditions segmentation is carried out to rabat image using lung conditions parted pattern neural network based；
Lobe of the lung segmentation is carried out to the rabat image using lobe of the lung parted pattern neural network based；And
Lung conditions volume is calculated based on the result of lung conditions segmentation and lobe of the lung segmentation.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811632060.XA CN109727251A (en) | 2018-12-29 | 2018-12-29 | The system that lung conditions are divided a kind of quantitatively, method and apparatus |
| US16/729,249 US11436720B2 (en) | 2018-12-28 | 2019-12-27 | Systems and methods for generating image metric |
| PCT/CN2019/129553 WO2020135792A1 (en) | 2018-12-28 | 2019-12-28 | Systems and methods for generating image metric |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811632060.XA CN109727251A (en) | 2018-12-29 | 2018-12-29 | The system that lung conditions are divided a kind of quantitatively, method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109727251A true CN109727251A (en) | 2019-05-07 |
Family
ID=66296067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811632060.XA Pending CN109727251A (en) | 2018-12-28 | 2018-12-29 | The system that lung conditions are divided a kind of quantitatively, method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109727251A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211134A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
| CN110400299A (en) * | 2019-07-23 | 2019-11-01 | 杭州依图医疗技术有限公司 | A kind of method and device of lung's pleural effusion detection |
| CN110706207A (en) * | 2019-09-12 | 2020-01-17 | 上海联影智能医疗科技有限公司 | Image quantization method, image quantization device, computer equipment and storage medium |
| CN110766652A (en) * | 2019-09-06 | 2020-02-07 | 上海联影智能医疗科技有限公司 | Network training method, device, segmentation method, computer equipment and storage medium |
| WO2020135792A1 (en) * | 2018-12-28 | 2020-07-02 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for generating image metric |
| CN111489335A (en) * | 2020-04-07 | 2020-08-04 | 广州柏视医疗科技有限公司 | Quantitative analysis system and method based on novel coronavirus pneumonia X-ray image |
| CN111724360A (en) * | 2020-06-12 | 2020-09-29 | 深圳技术大学 | Lung lobe segmentation method and device and storage medium |
| CN114332023A (en) * | 2021-12-30 | 2022-04-12 | 上海市嘉定区中心医院 | Pneumothorax automatic diagnosis and crisis early warning method, device, equipment and storage medium |
| CN115100179A (en) * | 2022-07-15 | 2022-09-23 | 北京医准智能科技有限公司 | Image processing method, device, equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105997119A (en) * | 2014-09-09 | 2016-10-12 | 西门子保健有限责任公司 | Method and system for simultaneously evaluating airway wall density and inflammation |
| CN104424629B (en) * | 2013-08-19 | 2018-01-26 | 深圳先进技术研究院 | A kind of x-ray chest radiograph lung segmentation method and apparatus |
| CN107909581A (en) * | 2017-11-03 | 2018-04-13 | 杭州依图医疗技术有限公司 | Lobe of the lung section dividing method, device, system, storage medium and the equipment of CT images |
| CN108288271A (en) * | 2018-02-06 | 2018-07-17 | 上海交通大学 | Image detecting system and method based on three-dimensional residual error network |
| CN108537784A (en) * | 2018-03-30 | 2018-09-14 | 四川元匠科技有限公司 | A kind of CT figure pulmonary nodule detection methods based on deep learning |
| CN108648172A (en) * | 2018-03-30 | 2018-10-12 | 四川元匠科技有限公司 | A kind of CT figure Lung neoplasm detecting systems based on 3D-Unet |
| CN109035283A (en) * | 2018-09-05 | 2018-12-18 | 中国医科大学附属盛京医院 | It is a kind of precisely to be detected and quantitative analysis method based on the pulmonary emphysema for randomly selecting subregion |
-
2018
- 2018-12-29 CN CN201811632060.XA patent/CN109727251A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104424629B (en) * | 2013-08-19 | 2018-01-26 | 深圳先进技术研究院 | A kind of x-ray chest radiograph lung segmentation method and apparatus |
| CN105997119A (en) * | 2014-09-09 | 2016-10-12 | 西门子保健有限责任公司 | Method and system for simultaneously evaluating airway wall density and inflammation |
| CN107909581A (en) * | 2017-11-03 | 2018-04-13 | 杭州依图医疗技术有限公司 | Lobe of the lung section dividing method, device, system, storage medium and the equipment of CT images |
| CN108288271A (en) * | 2018-02-06 | 2018-07-17 | 上海交通大学 | Image detecting system and method based on three-dimensional residual error network |
| CN108537784A (en) * | 2018-03-30 | 2018-09-14 | 四川元匠科技有限公司 | A kind of CT figure pulmonary nodule detection methods based on deep learning |
| CN108648172A (en) * | 2018-03-30 | 2018-10-12 | 四川元匠科技有限公司 | A kind of CT figure Lung neoplasm detecting systems based on 3D-Unet |
| CN109035283A (en) * | 2018-09-05 | 2018-12-18 | 中国医科大学附属盛京医院 | It is a kind of precisely to be detected and quantitative analysis method based on the pulmonary emphysema for randomly selecting subregion |
Non-Patent Citations (1)
| Title |
|---|
| SERKANKIRANYAZ等: "《多维粒子群优化在机器学习与模式识别中的应用》", 30 September 2017 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11436720B2 (en) | 2018-12-28 | 2022-09-06 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for generating image metric |
| WO2020135792A1 (en) * | 2018-12-28 | 2020-07-02 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for generating image metric |
| CN110211134B (en) * | 2019-05-30 | 2021-11-05 | 上海商汤智能科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
| CN110211134A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
| CN110400299A (en) * | 2019-07-23 | 2019-11-01 | 杭州依图医疗技术有限公司 | A kind of method and device of lung's pleural effusion detection |
| CN110766652A (en) * | 2019-09-06 | 2020-02-07 | 上海联影智能医疗科技有限公司 | Network training method, device, segmentation method, computer equipment and storage medium |
| CN110766652B (en) * | 2019-09-06 | 2022-12-20 | 上海联影智能医疗科技有限公司 | Network training method, device, segmentation method, computer equipment and storage medium |
| CN110706207A (en) * | 2019-09-12 | 2020-01-17 | 上海联影智能医疗科技有限公司 | Image quantization method, image quantization device, computer equipment and storage medium |
| CN111489335A (en) * | 2020-04-07 | 2020-08-04 | 广州柏视医疗科技有限公司 | Quantitative analysis system and method based on novel coronavirus pneumonia X-ray image |
| CN111489335B (en) * | 2020-04-07 | 2023-08-15 | 广州柏视医疗科技有限公司 | Quantitative analysis system and method based on novel coronavirus pneumonia X-ray image |
| CN111724360A (en) * | 2020-06-12 | 2020-09-29 | 深圳技术大学 | Lung lobe segmentation method and device and storage medium |
| CN111724360B (en) * | 2020-06-12 | 2023-06-02 | 深圳技术大学 | Lung lobe segmentation method, device and storage medium |
| CN114332023A (en) * | 2021-12-30 | 2022-04-12 | 上海市嘉定区中心医院 | Pneumothorax automatic diagnosis and crisis early warning method, device, equipment and storage medium |
| CN115100179A (en) * | 2022-07-15 | 2022-09-23 | 北京医准智能科技有限公司 | Image processing method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109727251A (en) | The system that lung conditions are divided a kind of quantitatively, method and apparatus | |
| Omonigho et al. | Breast cancer: tumor detection in mammogram images using modified alexnet deep convolution neural network | |
| US8126242B2 (en) | Computer program products and methods for detection and tracking of rheumatoid arthritis | |
| CN112365464B (en) | GAN-based medical image lesion area weak supervision positioning method | |
| US11386553B2 (en) | Medical image data | |
| CN114565613B (en) | Pancreas postoperative diabetes prediction system based on there is study of supervision degree of depth subspace | |
| CN113781640A (en) | Three-dimensional face reconstruction model establishing method based on weak supervised learning and application thereof | |
| CN109919254B (en) | Breast density classification method, system, readable storage medium and computer device | |
| CN112465905A (en) | Characteristic brain region positioning method of magnetic resonance imaging data based on deep learning | |
| Heine et al. | A statistical methodology for mammographic density detection | |
| CN109598730A (en) | Rabat segmentation and processing method, system and electronic equipment | |
| Gao et al. | Accurate lung segmentation for X-ray CT images | |
| CN114450716A (en) | Image processing for stroke characterization | |
| Tsap et al. | A vision-based technique for objective assessment of burn scars | |
| CN115661149A (en) | Lung image processing system based on lung tissue data | |
| CN116824173A (en) | Medical image processing method, medical image processing device and storage medium | |
| Thangam et al. | Skeletal Bone Age Assessment-Research Directions. | |
| CN111938655B (en) | Orbit soft tissue form evaluation method, system and equipment based on key point information | |
| KR102505614B1 (en) | Method and apparatus of providing osteoarthritis prediction information | |
| Furuo et al. | Automatic benign and malignant estimation of bone tumors using deep learning | |
| Hobbs et al. | Quad-mesh based radial distance biomarkers for Alzheimer's disease | |
| Landgren et al. | An automated system for the detection and diagnosis of kidney lesions in children from scintigraphy images | |
| CN113409447B (en) | Coronary artery segmentation method and device based on multi-slice combination | |
| CN112598669B (en) | Lung lobe segmentation method based on digital human technology | |
| Tong et al. | Abdominal adiposity quantification at MRI via fuzzy model-based anatomy recognition |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190507 |