CN109785300A - A kind of cancer medical image processing method, system, device and storage medium - Google Patents
A kind of cancer medical image processing method, system, device and storage medium Download PDFInfo
- Publication number
- CN109785300A CN109785300A CN201811611511.1A CN201811611511A CN109785300A CN 109785300 A CN109785300 A CN 109785300A CN 201811611511 A CN201811611511 A CN 201811611511A CN 109785300 A CN109785300 A CN 109785300A
- Authority
- CN
- China
- Prior art keywords
- medical image
- focal area
- cancer
- pathological characters
- module
- 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
Abstract
The invention discloses a kind of cancer medical image processing methods, system, device and storage medium, the method includes cancer medical image to be processed is input to the focal area parted pattern pre-established, obtain the focal area image identification result of the focal area parted pattern output, the filtering processing of multiple various criterion is carried out to the focal area image cut out from cancer medical image, carry out channel stacking, the result that channel stacks is input to the Feature Selection Model pre-established, target signature is filtered out from pathological characters vector obtained using Feature Selection algorithm；Target signature is input in multi-layer perception (MLP), the cancer biomarkers state index value of multi-layer perception (MLP) output is obtained.The present invention can be with the focal area on automatic identification cancer medical image, and focal area image cropping is come out, and identification greatly improved and cut efficiency and accuracy.The present invention is widely used in technical field of image processing.
Description
Technical field
The present invention relates to technical field of image processing, especially a kind of cancer medical image processing method, system, dress
It sets and storage medium.
Background technique
In cancer diagnosis and therapy field, often will the medical images such as CT to patient and MRI analyze, to lesion into
Row positioning and the feature for extracting lesion.But the existing generally existing following problems of medical image analysis technology:
Before extracting feature to focal area, need to identify and cut out from medical image the focal area of Small object,
To exclude the interference of irrelevant information, this process is that manual operation is needed to complete at present, and efficiency is extremely low, it is difficult to be realized to high-volume
The processing of medical image, while the precision for cutting focal area is lower；
In characteristic extraction procedure, that extracted using prior arts such as gray level co-occurrence matrixes is manual feature, this kind of spy
The quality that extraction algorithm relies on data set is levied, very sensitive to noise, generalization ability is poor, and needs to occupy and largely calculate money
Source；
When carrying out image procossing using neural network, single pass medical image is input in neural network, nerve
The detailed information that network can be got is less；
When carrying out image procossing using traditional networks such as unimproved UNet, it is easily lost during continuous sampling
Information.
The above problem causes prior art analysis precision and efficiency remains at low levels.
Term is explained:
Multi-layer perception (MLP): a kind of common neural network model includes input layer, hidden layer and output layer, and can make
Nonlinear change is carried out with activation primitive；
Convolutional neural networks: a kind of feedforward neural network with depth structure comprising several convolution kernels is commonly used in
The data with structural information are handled, can be used for carrying out the tasks such as feature extraction, target identification, classification to picture；
Characteristic pattern: the characteristic information obtained after convolution kernel carries out feature extraction is usually stacked by multiple two-dimensional matrixes
It forms；
Keras: one deep learning frame based on python carries out height to frames such as Tensorflow and Theano
Encapsulation, provides for user and is more convenient succinct interface；
Biomarker: a kind of marker related with growth and proliferation of cell can mark human tissue structure or function
The biochemical indicator of change can provide the foundation of auxiliary diagnosis for clinician；
KI-67: human cancer cell is proliferated active degree standards of measurement, is divided into 0-100 grade, high level Ki67
Value indicates that cancer cell increment is fast, grade of malignancy is high；
Lesion: the position of lesion, i.e., one pathological tissues limiting to, with pathogenic microorganism occur on body；
Attention mechanism: i.e. attention mechanism, can to source data sequence carry out data weighting transformation come allow model more
A certain partial data is paid close attention to, the predictive ability of model is improved.
Full convolutional neural networks: the convolutional neural networks without full articulamentum are mainly used for carrying out pixel scale to image
Classification；UNet: a kind of parted pattern of common medical image lesion is divided into two stages of coding and decoding, and passes through cross
The combination of different scale characteristic pattern is realized to connection, is divided pixel-by-pixel to realize；
Image filtering: the noise of target image is inhibited under conditions of retaining image minutia as far as possible, can be used
In the characteristic information of the prominent a certain type of image；
CLAHE: the limited self-adapting histogram equilibrium algorithm of contrast, by the contrast to image carry out transformation and
Local contrast is limited, to realize enhancing picture contrast；
SOBEL_BLUR: enhance the marginal information of image by seeking the first derivative of digital picture, be mainly used for edge
Detection；
ReLU: i.e. amendment linear unit, is common activation primitive in a kind of neural network, and expression formula is f (x)=max
(0,x)；
LeakyRelu: the special version of amendment linear unit (ReLU) still suffers from non-zero output value when inactivated, from
And a small gradient is obtained, avoid gradient disappearance problem；
VGG16: a kind of convolutional neural networks model can be used for figure wherein including 13 convolutional layers and 3 full articulamentums
As identification.
Summary of the invention
In order to solve the above-mentioned technical problem, the object of the invention is that provide a kind of cancer medical image processing method,
System, device and storage medium.
On the one hand, the present invention includes a kind of cancer medical image processing method, comprising the following steps:
Cancer medical image to be processed is input to the focal area parted pattern pre-established, obtains the focal zone
The focal area image identification result of regional partition model output；
The filtering processing of multiple various criterion is carried out to the focal area image cut out from cancer medical image；
The result by the focal area image of non-filtered processing and being repeatedly filtered carries out channel stacking；
The result that channel stacks is input to the Feature Selection Model pre-established, obtains the Feature Selection Model output
Pathological characters vector；The depth characteristic information for the result that the pathological characters vector is stacked with channel；
Target signature is filtered out from pathological characters vector obtained using Feature Selection algorithm；
Target signature is input in multi-layer perception (MLP), the cancer biomarkers state of the multi-layer perception (MLP) output is obtained
Index value.
Further, cancer medical image to be processed is input to the focal area parted pattern that pre-establishes this
Before step, 0-1 normalized is carried out to the cancer medical image.
Further, the filtering processing of the multiple various criterion specifically includes contrast enhancing filtering processing and edge letter
Breath enhancing filtering processing.
Further, the Feature Selection algorithm specifically includes the following steps:
Utilize formulaCalculate each pathology included in pathological characters vector
The score of feature；In formula, fisher_score (fi) it is i-th of pathological characters fiScore, njIndicate the sample of j-th of classification
Number, μiIndicate pathological characters fiMean value, μijIndicate pathological characters f in the sample of j-th of classificationiMean value,It indicates j-th
Pathological characters f in the sample of classificationiVariance, c indicate label classification sum；
Each pathological characters are arranged by the descending for the score being calculated；
The pathological characters stood out are exported as target signature.
Further, the focal area parted pattern is through the following steps that pre-establish:
Analysis pixel-by-pixel is carried out to multiple standard medical images, to construct the corresponding label of each standard medical image；
The label is the focal area being labeled on each standard medical image and biomarker state；
A part of standard medical image and its corresponding label are set up into training set, by another part standard medical image and
Its corresponding label sets up test set；
The segmentation network based on UNet is trained and is tested using the training set and test set, to obtain lesion
Region segmentation model.
Further, the quantity for the standard medical image that the standard medical image and test set that the training set includes include
The ratio between be 2:1.
Further, the segmentation network based on UNet includes the coding stage composed in series by multiple convolution modules；
Each convolution module is corresponding with respective Attention module；Each Attention module is sequentially connected in series；Respectively
Attention module be used for receive an Attention module output result and correspondence convolution module up-sampling as a result,
And it is exported after carrying out pixel weight adjustment to the received content of institute.
On the other hand, the invention also includes a kind of cancer medical image processing systems, comprising:
Module is divided in focal area, for cancer medical image to be processed to be input to the focal area pre-established point
Model is cut, the focal area image identification result of the focal area parted pattern output is obtained；
Filter module, for carrying out the filter of multiple various criterion to the focal area image cut out from cancer medical image
Wave processing；
Channel stack module, the result for the focal area image to non-filtered processing and being repeatedly filtered carry out
Channel stacks；
Characteristic extracting module, the result stacked for receiving channel simultaneously export pathological characters vector；The pathological characters to
The depth characteristic information for the result that amount is stacked with channel；
Feature vector screening module, for filtering out target signature from pathological characters vector obtained；
Index value output module, for exporting cancer biomarkers state index value according to target signature.
On the other hand, the invention also includes a kind of cancer medical image processing unit, including memory and processor,
The memory is for storing at least one program, and the processor is for loading at least one described program to execute the present invention
The cancer medical image processing method.
On the other hand, the invention also includes a kind of storage mediums, wherein being stored with the executable instruction of processor, the place
The executable instruction of reason device is used to execute when executed by the processor cancer medical image processing method of the present invention.
The beneficial effects of the present invention are: by using improved focal area parted pattern, it can be with automatic identification cancer
Focal area on disease medical image, and focal area image cropping is come out, identification greatly improved and cut efficiency and
Accuracy；By using focal area parted pattern, Feature Selection Model and multi-layer perception (MLP), existing feature extraction skill is overcome
The disadvantages of art is very sensitive, generalization ability is poor to noise；By filtering and the processing of channel stacking, by single pass cancer medical image
Become multichannel, subsequent processes is enabled to obtain more detailed information.
Detailed description of the invention
Fig. 1 is a kind of flow chart of cancer medical image processing method of the present invention；
Fig. 2 is the network structure of Feature Selection Model in the embodiment of the present invention；
Fig. 3 is the network structure of multi-layer perception (MLP) in the embodiment of the present invention；
Fig. 4 is the schematic diagram of Attention module in the embodiment of the present invention；
Fig. 5 is the schematic diagram of focal area parted pattern in the embodiment of the present invention；
Fig. 6 is the cancer medical image being input in the parted pattern of focal area in the embodiment of the present invention；
Fig. 7 is the focal area image of parted pattern output in focal area in the embodiment of the present invention.
Specific embodiment
A kind of cancer medical image processing method, referring to Fig.1, comprising the following steps:
S1., cancer medical image to be processed is input to the focal area parted pattern pre-established, obtains the disease
The focal area image identification result of stove region segmentation model output；
S2. the filtering processing of multiple various criterion is carried out to the focal area image cut out from cancer medical image；
S3. the result being filtered by the focal area image of non-filtered processing and repeatedly carries out channel stacking；
S4. the result that channel stacks is input to the Feature Selection Model pre-established, obtains the Feature Selection Model
The pathological characters vector of output；The depth characteristic information for the result that the pathological characters vector is stacked with channel；
S5. target signature is filtered out from pathological characters vector obtained using Feature Selection algorithm；
S6. target signature is input in multi-layer perception (MLP), obtains the cancer biomarkers of the multi-layer perception (MLP) output
State index value.
Cancer medical image processing method of the present invention divides stage and biological markers detection stage two by lesion image
Part forms, and wherein the lesion image segmentation stage includes step S1, and the biological markers detection stage includes step S2-S6.
Cancer medical image in the present embodiment refers to the images such as CT, the MRI of patient shot by Medical Devices.
In step sl, trained focal area parted pattern can analyze cancer medical image, and will
Focal area part in cancer medical image is identified from cancer medical image, and output identification result.Focal area
Parted pattern can replace the manual operation relied in the prior art, greatly improve the efficiency and precision of focal area identification.
Preferably, in step S1, that is, the cancer medical image that will test is input to the focal area pre-established
Before the step for parted pattern, 0-1 normalized is carried out to the cancer medical image.It, can be with by 0-1 normalized
The value of all pixels of cancer medical image is normalized into [0,1] section, can to reach in the implementation procedure of step S1-S6
To higher convergence rate and precision.
In step S2, according to the focal area image identification of step S1 output as a result, focal area image is cured from cancer
It learns to cut out in image and, then carry out the filtering processing of multiple various criterion.It is further used as preferred embodiment, step
The filtering processing of multiple various criterion described in S2 specifically includes at contrast enhancing filtering processing and marginal information enhancing filtering
Reason.Preferably, focal area image degree of comparing is enhanced using CLAHE filter in the present embodiment and is filtered, used
SOBEL_BLUR filter carries out marginal information enhancing filtering processing to focal area image.
Enhanced by contrast and be filtered, the contrast of focal area image is higher, and important foreground area can be more
It is distinguished from background well.Enhanced by marginal information and be filtered, the edge contour of focal area image becomes apparent from, can
By protrude the organ or tissue in lesion area image position and in the form of.
In step S3, by the focal area image of non-filtered processing, by the focal zone of contrast enhancing filtering processing
Area image and the focal area image progress channel stacking being filtered by marginal information enhancing, form triple channel image,
Focal area image itself can be overcome only to have the shortcomings that single channel, subsequent processing steps include that detailed information is less.
In step S4, Feature Selection Model can be the networks such as VGG16, GoogleNet or ResNet, and the present embodiment uses
VGG16 network as shown in Figure 2 includes 13 convolutional layers and 5 pond layers as Feature Selection Model, the VGG16 network, is made
With the good model of pre-training in Keras and weight, the characteristic pattern that the last one pond layer exports, which is carried out flattening, can obtain one
The feature vector of dimension, the depth characteristic information for the result that this feature vector is stacked with channel, is that step S4 is to be output
Pathological characters vector.
In step S5, it is special that qualified target is filtered out from multiple pathological characters vectors using Feature Selection algorithm
Sign.Feature Selection algorithm used in the present embodiment be Fisher Score algorithm, the algorithm the following steps are included:
S501. formula is utilizedIt calculates included in pathological characters vector
The score of each pathological characters；In formula, fisher_score (fi) it is i-th of pathological characters fiScore, njIndicate j-th of classification
Sample number, μiIndicate pathological characters fiMean value, μijIndicate pathological characters f in the sample of j-th of classificationiMean value,Table
Show pathological characters f in the sample of j-th of classificationiVariance, c indicate label classification sum；
S502. each pathological characters are arranged by the descending for the score being calculated；
S503 exports the pathological characters stood out as target signature.
It is that each pathological characters vector calculates corresponding point by Fisher Score algorithmic formula in step S501
Number, Fisher Score algorithm is a kind of feature selecting algorithm for having supervision, and target is so that identical after feature selecting
The sample of classification is similar as much as possible in new feature space, and different classes of sample is separated as much as possible.In the present embodiment,
Fisher Score algorithm is realized using the library scikit-feature of python, and uses the biological aspect mark of each sample
Label save the index of k feature after the completion of training, and retain primitive character as supervision message training this feature preference pattern
The feature of corresponding index position is exported as the feature vector after Feature Selection in vector.
By the sequence of step S502, the higher pathological characters sequence of score is more forward.
In step S503, it can choose fixed preceding k pathological characters and exported as target signature.
By the processing of step S5, the pathological characters comprising excessive irrelevant information can be eliminated, avoid due to
The dimension of pathological characters vector is higher and causes purity lower, Time & Space Complexity needed for can reducing subsequent process.
In step S6, the target signature screened is handled using multi-layer perception (MLP), exports cancer biomarkers
State index value is as final result.The present embodiment uses network as shown in Figure 3 as multi-layer perception (MLP), which includes 5
A full articulamentum, other than the last layer output layer, other layers use LeakyRelu to disappear as activation primitive to avoid gradient
It loses.Multi-layer perception (MLP) exports Ki67 value as final result by the last one neuron, and uses mean square error mean_
Squared_error carries out weight optimization as objective function.In the training process to the multi-layer perception (MLP), it can be used
Keras is built, and storage configuration and weight parameter can be obtained a reusable biological markers detection mould after training
Type.
It is further used as preferred embodiment, the focal area parted pattern used in step S1 is by following
What step pre-established:
S100. analysis pixel-by-pixel is carried out to multiple standard medical images, so that it is corresponding to construct each standard medical image
Label；The label is the focal area being labeled on each standard medical image and biomarker state；
S200. a part of standard medical image and its corresponding label are set up into training set, by another part standard medical
Image and its corresponding label set up test set；
S300. the segmentation network based on Attention mechanism and UNet is carried out using the training set and test set
Training and test, to obtain focal area parted pattern.
Step S100 and S200 are for constructing training set and test set.The material of training set and test set comes from standard medical
Image.In the present embodiment, standard medical image and cancer medical image belong to same type, standard medical image for training and
Focal area parted pattern is tested, and cancer medical image is the process object of the focal area parted pattern after training.
In step S100, multiple standard medical images can be analyzed by medical staff, in each standard medical image
On mark out focal area and biomarker state as label, so that each standard medical image has its corresponding label.
In step S200, by way of randomly selecting, from multiple standard medical image selection a part, by what is be selected
Standard medical image and its corresponding label set up training set；By not selected standard medical image and its corresponding set of tags
Build test set.Preferably, the ratio of number for the standard medical image that the standard medical image and test set that training set includes include
For 2:1.
In step S300, network model is divided for improved UNet, it is trained using training set and test set,
Thus focal area parted pattern needed for obtaining step S1.
It is further used as preferred embodiment, in step S300, the segmentation network based on UNet includes by more
The coding stage that a convolution module composes in series；Each convolution module is corresponding with respective Attention module；It is each described
Attention module is sequentially connected in series；Each Attention module be used for receive an Attention module output result and
The up-sampling of corresponding convolution module is as a result, and export after carrying out pixel weight adjustment to the received content of institute.
Segmentation network used in step S300 based on UNet as shown in figure 4, its by five convolution modules, that is, module 1-
Module 5 form, each convolution module include at least two convolution kernel sizes be 3 × 3, the convolutional layer that activation primitive is ReLu, no
Step-length is used the size of characteristic pattern to be reduced to original half for 2 × 2 maximization pond between module, and in each module
The quantity of characteristic pattern is all twice of a upper module.
It is being every provided with Attention module based on the up-sampling stage of the segmentation network of UNet in the present embodiment
A pixel assigns different weights.
The principle of Attention module is as shown in Figure 4: there are two input terminal and one are defeated for each Attention module tool
Outlet, two of them input terminal are respectively used to receive original characteristic pattern F1With the characteristic pattern F obtained by up-sampling2, output end
For exporting the characteristic pattern F for passing through weight and adjusting3, F3Size and F2Unanimously.For characteristic pattern F1, continuous using two
, step-length greater than 13 × 3 convolutional layers carry out down-sampling, make F1Size become 8 × 8 × 64 fixed (length × wide × channels
Number), the one-dimensional vector C that length is 4096 is obtained after flattening.For characteristic pattern F2, take the information of corresponding position on its different channel
Form one-dimensional characteristic vector Vi,j(wherein i and j respectively indicates the row and column where the pixel).Then, to F2In each pixel
Calculate its weighted value Wi,j=P1·Vi,j+P2C+b, wherein P1And P2It is trainable one-dimensional vector, b is that can train ginseng
Number is optimized when training by back-propagation algorithm, and all pixels share P1、P2With tri- parameters of b.All Wi,jComposition one
A weight matrix W is for indicating that model to the degree of concern of characteristic pattern different pixels position, is worth the important of the bigger expression region
Degree is higher.Finally, F2In each pixel value can be obtained with the multiplied by weight of W corresponding position it is adjusted by weight
Characteristic pattern F3。
By Attention module application into the segmentation network based on UNet, principle is as shown in figure 5, wherein trapezoidal frame portion
Dividing indicates Attention module.In Fig. 5, deconvolution is respectively adopted in module 2 to the characteristic pattern that module 5 obtains of coding stage
Mode is up-sampled, and one times of its expanded in size is made；The characteristic pattern M that module 5 is obtained5With the feature obtained after up-sampling
Scheme M '5It is input to update Pixel Information in Attention module and obtains M4, new characteristic pattern M4It is obtained with module 4 by up-sampling
Characteristic pattern M '4Attention operation is continued to execute, this process is repeated and obtains M1, the characteristic pattern obtained with module 1 executes
Last time Attention operation, so that obtaining one group contains predictive information pixel-by-pixel and size is equal with original image
Heating power prognostic chart.
Finally, the present embodiment carries out information integration using two convolutional layers, convolution kernel size is respectively 3 × 3 and 1 × 1, and
Classified by softmax on each pixel to judge whether the pixel starts from focal area, here using cross entropy as
Loss function.Above-mentioned Web vector graphic open source deep learning frame Keras is built, and using pretreated image as input, disease
The label in stove region is trained as supervision message, and can be obtained one after the completion of training can carry out pixel-by-pixel focal area
The lesion segmentation model of prediction.
The Attention module being arranged in the present embodiment will calculate the feature vector in each pixel, or more
It states focal area parted pattern and compares consuming computing resource.It, can not be right in the limited situation of hardware resource in practical application
The bigger module of characteristic pattern size uses Attention mechanism, using the method for original UNet, the feature by up-sampling
The characteristic pattern of figure and lateral connection carries out splicing laggard row information integration, finally exports the prediction result of focal area.
Cancer medical image shown in fig. 6 is input in focal area parted pattern as shown in Figure 5 in the present embodiment,
Focal area image shown in Fig. 7 can be exported.
The invention also includes a kind of cancer medical image processing systems, comprising:
Module is divided in focal area, for cancer medical image to be processed to be input to the focal area pre-established point
Model is cut, the focal area image identification result of the focal area parted pattern output is obtained；
Filter module, for carrying out the filter of multiple various criterion to the focal area image cut out from cancer medical image
Wave processing；
Channel stack module, the result for the focal area image to non-filtered processing and being repeatedly filtered carry out
Channel stacks；
Characteristic extracting module, the result stacked for receiving channel simultaneously export pathological characters vector；The pathological characters to
The depth characteristic information for the result that amount is stacked with channel；
Feature vector screening module, for being sieved from multiple pathological characters vectors when obtaining multiple pathological characters vectors
Select target signature；
Index value output module, for exporting cancer biomarkers state index value according to target signature.
Any combination implementation steps of cancer medical image processing system executing method embodiment of the present invention, have
The corresponding function of this method and beneficial effect.
The invention also includes a kind of cancer medical image processing unit, including memory and processor, the storages
Device is for storing at least one program, and the processor is for loading at least one described program to execute cancer of the present invention
Medical image processing method.
Any combination implementation steps of cancer medical image processing unit executing method embodiment of the present invention, have
The corresponding function of this method and beneficial effect.
The invention also includes a kind of storage mediums, wherein being stored with the executable instruction of processor, the processor can be held
Capable instruction is used to execute when executed by the processor cancer medical image processing method of the present invention.
To sum up, the invention has the advantages that
It, can be with the focal zone on automatic identification cancer medical image by using improved focal area parted pattern
Domain, and focal area image cropping is come out, identification greatly improved and cuts efficiency and accuracy；
By using focal area parted pattern, Feature Selection Model and multi-layer perception (MLP), existing feature extraction is overcome
The disadvantages of technology is very sensitive, generalization ability is poor to noise；
By filtering and the processing of channel stacking, single pass cancer medical image is become into multichannel, so that subsequent processing
Process can obtain more detailed information；
Further, it is provided with Attention module on the basis of segmentation network of the tradition based on UNet, is continuously adopting
Retain more information during sample；The pathological characters comprising excessive irrelevant information can be screened out by Fisher Score algorithm,
Improve computational efficiency.
It is to be illustrated to preferable implementation of the invention, but the implementation is not limited to the invention above
Example, those skilled in the art can also make various equivalent variations on the premise of without prejudice to spirit of the invention or replace
It changes, these equivalent deformations or replacement are all included in the scope defined by the claims of the present application.
Claims (10)
1. a kind of cancer medical image processing method, which comprises the following steps:
Cancer medical image to be processed is input to the focal area parted pattern pre-established, obtains the focal area point
Cut the focal area image identification result of model output；
The filtering processing of multiple various criterion is carried out to the focal area image cut out from cancer medical image；
The result by the focal area image of non-filtered processing and being repeatedly filtered carries out channel stacking；
The result that channel stacks is input to the Feature Selection Model pre-established, obtains the disease of the Feature Selection Model output
Manage feature vector；The depth characteristic information for the result that the pathological characters vector is stacked with channel；
Target signature is filtered out from pathological characters vector obtained using Feature Selection algorithm；
Target signature is input in multi-layer perception (MLP), the cancer biomarkers state index of the multi-layer perception (MLP) output is obtained
Value.
2. a kind of cancer medical image processing method according to claim 1, which is characterized in that will be to be processed
Before cancer medical image is input to the step for focal area parted pattern pre-established, the cancer medical image is carried out
0-1 normalized.
3. a kind of cancer medical image processing method according to claim 1, which is characterized in that the multiple difference
The filtering processing of standard specifically includes contrast enhancing filtering processing and marginal information enhancing filtering processing.
4. a kind of cancer medical image processing method according to claim 1, which is characterized in that the Feature Selection
Algorithm specifically includes the following steps:
Utilize formulaCalculate each pathological characters included in pathological characters vector
Score；In formula, fisher_score (fi) it is i-th of pathological characters fiScore, njIndicate the sample number of j-th of classification, μi
Indicate pathological characters fiMean value, μijIndicate pathological characters f in the sample of j-th of classificationiMean value,Indicate j-th of classification
Sample in pathological characters fiVariance, c indicate label classification sum；
Each pathological characters are arranged by the descending for the score being calculated；
The pathological characters stood out are exported as target signature.
5. a kind of cancer medical image processing method according to claim 1, which is characterized in that the focal area
Parted pattern is through the following steps that pre-establish:
Analysis pixel-by-pixel is carried out to multiple standard medical images, to construct the corresponding label of each standard medical image；It is described
Label is the focal area being labeled on each standard medical image and biomarker state；
A part of standard medical image and its corresponding label are set up into training set, by another part standard medical image and its right
The label answered sets up test set；
The segmentation network based on UNet is trained and is tested using the training set and test set, to obtain focal area
Parted pattern.
6. a kind of cancer medical image processing method according to claim 5, which is characterized in that the training set packet
The ratio of number for the standard medical image that the standard medical image and test set contained includes is 2:1.
7. a kind of cancer medical image processing method according to claim 5, which is characterized in that described to be based on UNet
Segmentation network include the coding stage composed in series by multiple convolution modules；Each convolution module is corresponding with respective
Attention module；Each Attention module is sequentially connected in series；Each Attention module is for receiving one
The up-sampling of the output result of Attention module and corresponding convolution module is as a result, and carry out pixel to the received content of institute
It is exported after weight adjustment.
8. a kind of cancer medical image processing system characterized by comprising
Module is divided in focal area, divides mould for cancer medical image to be processed to be input to the focal area pre-established
Type obtains the focal area image identification result of the focal area parted pattern output；
Filter module, for carrying out filtering place of multiple various criterion to the focal area image cut out from cancer medical image
Reason；
Channel stack module, the result for the focal area image to non-filtered processing and being repeatedly filtered carry out channel
It stacks；
Characteristic extracting module, the result stacked for receiving channel simultaneously export pathological characters vector；The pathological characters vector band
The depth characteristic information for the result for thering is channel to stack；
Feature Selection module, for filtering out target signature from pathological characters vector obtained；
Index value output module, for exporting cancer biomarkers state index value according to target signature.
9. a kind of cancer medical image processing unit, which is characterized in that including memory and processor, the memory is used
In storing at least one program, the processor requires any one of 1-7 for loading at least one described program with perform claim
The method.
10. a kind of storage medium, wherein being stored with the executable instruction of processor, which is characterized in that the processor is executable
Instruction be used to execute such as any one of claim 1-7 the method when executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811611511.1A CN109785300A (en) | 2018-12-27 | 2018-12-27 | A kind of cancer medical image processing method, system, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811611511.1A CN109785300A (en) | 2018-12-27 | 2018-12-27 | A kind of cancer medical image processing method, system, device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109785300A true CN109785300A (en) | 2019-05-21 |
Family
ID=66498537
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811611511.1A Pending CN109785300A (en) | 2018-12-27 | 2018-12-27 | A kind of cancer medical image processing method, system, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785300A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110298345A (en) * | 2019-07-05 | 2019-10-01 | 福州大学 | A kind of area-of-interest automatic marking method of medical images data sets |
| CN110600122A (en) * | 2019-08-23 | 2019-12-20 | 腾讯医疗健康(深圳)有限公司 | Digestive tract image processing method and device and medical system |
| CN110648331A (en) * | 2019-09-02 | 2020-01-03 | 众安信息技术服务有限公司 | Detection method for medical image segmentation, medical image segmentation method and device |
| CN110648311A (en) * | 2019-09-03 | 2020-01-03 | 南开大学 | Acne image focus segmentation and counting network model based on multitask learning |
| CN111180011A (en) * | 2019-12-31 | 2020-05-19 | 上海依智医疗技术有限公司 | Method and device for detecting gene mutation of focus |
| CN111210449A (en) * | 2019-12-23 | 2020-05-29 | 深圳市华嘉生物智能科技有限公司 | Automatic segmentation method for gland cavity in prostate cancer pathological image |
| CN111276254A (en) * | 2020-01-13 | 2020-06-12 | 印迹信息科技（北京）有限公司 | Medical open platform system and diagnosis and treatment data processing method |
| CN111739640A (en) * | 2020-06-22 | 2020-10-02 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Risk prediction system based on mammary gland molybdenum target and MR image imaging omics |
| CN112116074A (en) * | 2020-09-18 | 2020-12-22 | 西北工业大学 | Image description method based on two-dimensional space coding |
| CN113362350A (en) * | 2021-07-26 | 2021-09-07 | 海南大学 | Segmentation method and device for cancer medical record image, terminal device and storage medium |
| CN113421274A (en) * | 2021-07-13 | 2021-09-21 | 上海工程技术大学 | Intelligent stomach cancer staging method based on CT (computed tomography) images |
| CN114782432A (en) * | 2022-06-20 | 2022-07-22 | 南通电博士自动化设备有限公司 | Edge detection method of improved canny operator based on textural features |
| CN115359325A (en) * | 2022-10-19 | 2022-11-18 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and medium of image recognition model |
| WO2022247007A1 (en) * | 2021-05-25 | 2022-12-01 | 平安科技（深圳）有限公司 | Medical image grading method and apparatus, electronic device, and readable storage medium |
| CN116258717A (en) * | 2023-05-15 | 2023-06-13 | 广州思德医疗科技有限公司 | Lesion recognition method, device, apparatus and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105893946A (en) * | 2016-03-29 | 2016-08-24 | 中国科学院上海高等研究院 | Front face image detection method |
| CN108257135A (en) * | 2018-02-01 | 2018-07-06 | 浙江德尚韵兴图像科技有限公司 | The assistant diagnosis system of medical image features is understood based on deep learning method |
| CN108257134A (en) * | 2017-12-21 | 2018-07-06 | 深圳大学 | Nasopharyngeal Carcinoma Lesions automatic division method and system based on deep learning |
| WO2018189550A1 (en) * | 2017-04-12 | 2018-10-18 | Kheiron Medical Technologies Ltd | Segmentation of anatomical regions and lesions |
| CN109003280A (en) * | 2018-07-06 | 2018-12-14 | 华南理工大学 | Inner membrance dividing method in a kind of blood vessel of binary channels intravascular ultrasound image |
-
2018
- 2018-12-27 CN CN201811611511.1A patent/CN109785300A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105893946A (en) * | 2016-03-29 | 2016-08-24 | 中国科学院上海高等研究院 | Front face image detection method |
| WO2018189550A1 (en) * | 2017-04-12 | 2018-10-18 | Kheiron Medical Technologies Ltd | Segmentation of anatomical regions and lesions |
| CN108257134A (en) * | 2017-12-21 | 2018-07-06 | 深圳大学 | Nasopharyngeal Carcinoma Lesions automatic division method and system based on deep learning |
| CN108257135A (en) * | 2018-02-01 | 2018-07-06 | 浙江德尚韵兴图像科技有限公司 | The assistant diagnosis system of medical image features is understood based on deep learning method |
| CN109003280A (en) * | 2018-07-06 | 2018-12-14 | 华南理工大学 | Inner membrance dividing method in a kind of blood vessel of binary channels intravascular ultrasound image |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110298345A (en) * | 2019-07-05 | 2019-10-01 | 福州大学 | A kind of area-of-interest automatic marking method of medical images data sets |
| CN110600122A (en) * | 2019-08-23 | 2019-12-20 | 腾讯医疗健康(深圳)有限公司 | Digestive tract image processing method and device and medical system |
| CN110600122B (en) * | 2019-08-23 | 2023-08-29 | 腾讯医疗健康(深圳)有限公司 | Digestive tract image processing method and device and medical system |
| CN110648331B (en) * | 2019-09-02 | 2023-01-17 | 众安信息技术服务有限公司 | Detection method for medical image segmentation, medical image segmentation method and device |
| CN110648331A (en) * | 2019-09-02 | 2020-01-03 | 众安信息技术服务有限公司 | Detection method for medical image segmentation, medical image segmentation method and device |
| CN110648311A (en) * | 2019-09-03 | 2020-01-03 | 南开大学 | Acne image focus segmentation and counting network model based on multitask learning |
| CN110648311B (en) * | 2019-09-03 | 2023-04-18 | 南开大学 | Acne image focus segmentation and counting network model based on multitask learning |
| CN111210449B (en) * | 2019-12-23 | 2024-01-23 | 深圳市华嘉生物智能科技有限公司 | Automatic segmentation method for gland cavity in prostate cancer pathological image |
| CN111210449A (en) * | 2019-12-23 | 2020-05-29 | 深圳市华嘉生物智能科技有限公司 | Automatic segmentation method for gland cavity in prostate cancer pathological image |
| CN111180011A (en) * | 2019-12-31 | 2020-05-19 | 上海依智医疗技术有限公司 | Method and device for detecting gene mutation of focus |
| CN111276254A (en) * | 2020-01-13 | 2020-06-12 | 印迹信息科技（北京）有限公司 | Medical open platform system and diagnosis and treatment data processing method |
| CN111739640A (en) * | 2020-06-22 | 2020-10-02 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Risk prediction system based on mammary gland molybdenum target and MR image imaging omics |
| CN112116074A (en) * | 2020-09-18 | 2020-12-22 | 西北工业大学 | Image description method based on two-dimensional space coding |
| CN112116074B (en) * | 2020-09-18 | 2022-04-15 | 西北工业大学 | Image description method based on two-dimensional space coding |
| WO2022247007A1 (en) * | 2021-05-25 | 2022-12-01 | 平安科技（深圳）有限公司 | Medical image grading method and apparatus, electronic device, and readable storage medium |
| CN113421274A (en) * | 2021-07-13 | 2021-09-21 | 上海工程技术大学 | Intelligent stomach cancer staging method based on CT (computed tomography) images |
| CN113362350A (en) * | 2021-07-26 | 2021-09-07 | 海南大学 | Segmentation method and device for cancer medical record image, terminal device and storage medium |
| CN113362350B (en) * | 2021-07-26 | 2024-04-02 | 海南大学 | Method, device, terminal equipment and storage medium for segmenting cancer medical record image |
| CN114782432A (en) * | 2022-06-20 | 2022-07-22 | 南通电博士自动化设备有限公司 | Edge detection method of improved canny operator based on textural features |
| CN115359325B (en) * | 2022-10-19 | 2023-01-10 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and medium for image recognition model |
| CN115359325A (en) * | 2022-10-19 | 2022-11-18 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and medium of image recognition model |
| CN116258717A (en) * | 2023-05-15 | 2023-06-13 | 广州思德医疗科技有限公司 | Lesion recognition method, device, apparatus and storage medium |
| CN116258717B (en) * | 2023-05-15 | 2023-09-08 | 广州思德医疗科技有限公司 | Lesion recognition method, device, apparatus and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109785300A (en) | A kind of cancer medical image processing method, system, device and storage medium | |
| CN106202997B (en) | A kind of cell division detection method based on deep learning | |
| CN110889853B (en) | Tumor segmentation method based on residual error-attention deep neural network | |
| Yang et al. | Class-balanced deep neural network for automatic ventricular structure segmentation | |
| CN110705555B (en) | Abdomen multi-organ nuclear magnetic resonance image segmentation method, system and medium based on FCN | |
| CN110097554B (en) | Retina blood vessel segmentation method based on dense convolution and depth separable convolution | |
| CN110930416B (en) | MRI image prostate segmentation method based on U-shaped network | |
| CN110889852B (en) | Liver segmentation method based on residual error-attention deep neural network | |
| CN108021916A (en) | Deep learning diabetic retinopathy sorting technique based on notice mechanism | |
| Klibisz et al. | Fast, simple calcium imaging segmentation with fully convolutional networks | |
| CN110675411B (en) | Cervical squamous intraepithelial lesion recognition algorithm based on deep learning | |
| CN107341518A (en) | A kind of image classification method based on convolutional neural networks | |
| CN114846507A (en) | Method and system for non-invasive gene detection using Artificial Intelligence (AI) models | |
| CN108492271A (en) | A kind of automated graphics enhancing system and method for fusion multi-scale information | |
| US11954593B2 (en) | Method to determine a degree of abnormality, a respective computer readable medium and a distributed cancer analysis system | |
| CN104484886B (en) | A kind of dividing method and device of MR images | |
| CN108416318A (en) | Diameter radar image target depth method of model identification based on data enhancing | |
| Cao et al. | Gastric cancer diagnosis with mask R-CNN | |
| CN112347977B (en) | Automatic detection method, storage medium and device for induced pluripotent stem cells | |
| CN114708255B (en) | Multi-center children X-ray chest image lung segmentation method based on TransUNet model | |
| CN108305253A (en) | A kind of pathology full slice diagnostic method based on more multiplying power deep learnings | |
| CN110287777A (en) | A kind of golden monkey body partitioning algorithm under natural scene | |
| CN113327258A (en) | Lung CT image identification method based on deep learning | |
| CN106845551A (en) | A kind of histopathology image-recognizing method | |
| CN111899259A (en) | Prostate cancer tissue microarray classification method based on convolutional 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190521 |
|
| RJ01 | Rejection of invention patent application after publication |