CN109671076A - Blood vessel segmentation method, apparatus, electronic equipment and storage medium - Google Patents
Blood vessel segmentation method, apparatus, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN109671076A CN109671076A CN201811577437.6A CN201811577437A CN109671076A CN 109671076 A CN109671076 A CN 109671076A CN 201811577437 A CN201811577437 A CN 201811577437A CN 109671076 A CN109671076 A CN 109671076A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network model
- image
- setting
- segmentation
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Abstract
The embodiment of the invention discloses a kind of blood vessel segmentation method, apparatus, electronic equipment and storage mediums.This method comprises: the setting neural network model for respectively training image to be split input at least two in advance, generates initial segmentation result corresponding with each setting neural network model;Using the corresponding object module weighted value of each setting neural network model, processing is weighted to each initial segmentation result, generates weighting segmentation result, wherein the object module weighted value determination when training setting neural network model;Post processing of image is carried out to the weighting segmentation result, generates segmentation blood-vessel image.Boosting idea about modeling is applied to blood vessel segmentation field for the first time through the above technical solutions, realizing, it is lower to solve the problems, such as that single Neural model obtains vessel segmentation precision, has achieved the effect that improve blood vessel segmentation precision.
Description
Technical field
The present embodiments relate to Medical Image Processings more particularly to a kind of blood vessel segmentation method, apparatus, electronics to set
Standby and storage medium.
Background technique
Blood vessel segmentation in medical image is a basic problem, such as the blood vessel segmentation of liver is widely used in liver
Diagnosis, treatment and the planning of operation on liver of lesion.
A kind of common method of blood vessel segmentation is traditional image processing method, such as threshold segmentation method or region at present
Growing method etc., but the blood vessel segmentation precision of these methods is limited, it is difficult to meet clinical demand.Another kind of common method is
Image processing method based on deep learning, such as full convolutional neural networks model (Fully convolutional
Networks, FCN), the full convolutional neural networks model U-net based on two-dimensional medical images and based on the complete of 3 d medical images
Convolutional neural networks model V-net etc., these methods outclass traditional image processing method in accuracy rate and robustness,
But the complexity of limitation and blood vessel itself morphosis due to imaging device, utilize any of the above-described a convolutional neural networks mould
The problems such as resulting vessel segmentation of type is easy to appear less vessel branch, vessel borders mistake and segmentation are blocky.
Summary of the invention
The embodiment of the present invention provides a kind of blood vessel segmentation method, apparatus, electronic equipment and storage medium, to improve blood vessel point
The precision cut.
In a first aspect, the embodiment of the invention provides a kind of blood vessel segmentation methods, comprising:
The setting neural network model that image to be split input at least two is trained in advance respectively, generates and each setting
The corresponding initial segmentation result of neural network model;
Using the corresponding object module weighted value of each setting neural network model, to each initial segmentation result
It is weighted processing, generates weighting segmentation result, wherein the object module weighted value is in the training setting neural network mould
It is determined when type;
Post processing of image is carried out to the weighting segmentation result, generates segmentation blood-vessel image.
Second aspect, the embodiment of the invention also provides a kind of blood vessel segmentation device, which includes:
Initial segmentation result generation module, the setting mind for respectively training image to be split input at least two in advance
Through network model, at least two initial segmentation results are generated;
Segmentation result generation module is weighted, for utilizing the corresponding object module of each setting neural network model to weigh
Weight values are weighted processing to each initial segmentation result, generate weighting segmentation result, wherein the object module weight
It is worth and is determined when the training setting neural network model;
Divide blood-vessel image generation module, for carrying out post processing of image to the weighting segmentation result, generates segmentation blood
Pipe image.
The third aspect, the embodiment of the invention also provides a kind of electronic equipment, which includes:
One or more processors;
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes blood vessel segmentation method provided by any embodiment of the invention.
Fourth aspect, the embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer
Program, the computer program realize blood vessel segmentation method provided by any embodiment of the invention when being executed by processor.
The embodiment of the present invention is by inputting at least two setting neural network moulds trained in advance for image to be split respectively
Type generates initial segmentation result corresponding with each setting neural network model;Utilize each setting neural network model
Corresponding object module weighted value is weighted processing to each initial segmentation result, generates weighting segmentation result;To described
It weights segmentation result and carries out post processing of image, generate segmentation blood-vessel image.It realizes based on Boosting idea about modeling, using more
A setting neural network model weighted value corresponding with its constructs reinforced neural network model, to carry out to blood-vessel image
It is lower to solve the problems, such as that single Neural model obtains vessel segmentation precision, has reached raising blood vessel for blood vessel segmentation
The effect of segmentation precision.
Detailed description of the invention
Fig. 1 is the flow chart of one of embodiment of the present invention one blood vessel segmentation method;
Fig. 2 is the implementation procedure schematic diagram of one of embodiment of the present invention one blood vessel segmentation method;
Fig. 3 is the flow chart of the model training method of one of the embodiment of the present invention two blood vessel segmentation;
Fig. 4 is the structural schematic diagram of one of embodiment of the present invention three blood vessel segmentation device;
Fig. 5 is the structural schematic diagram of one of the embodiment of the present invention four electronic equipment.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limiting the invention.It also should be noted that in order to just
Only the parts related to the present invention are shown in description, attached drawing rather than entire infrastructure.
Embodiment one
A kind of blood vessel segmentation method provided in this embodiment is applicable to the blood vessel segmentation of medical image, is particularly suitable for multiple
Blood vessel segmentation in miscellaneous organ, such as liver vessel segmentation.This method can be executed by blood vessel segmentation device, which can be with
It is realized by the mode of software and/or hardware, which can integrate in the electronic equipment with image processing function, such as platform
Formula computer or server etc..Referring to Fig. 1, the method for the present embodiment specifically comprises the following steps:
S110, the setting neural network model for respectively training image to be split input at least two in advance, generate and every
The corresponding initial segmentation result of a setting neural network model.
Wherein, image to be split refers to the medical image of the blood vessel comprising needing to divide, and can be two-dimensional medical images,
It is also possible to 3 d medical images.Setting neural network model refers to a training in advance, for carrying out the base of blood vessel segmentation
In the neural network model of deep learning, such as it can be full convolutional neural networks module FCN, Mask-RCNN, DeepLab, U-
Net, V-Net or SegNet etc..Setting neural network model is to be constructed relative to the application based on Boosting idea about modeling
For reinforced neural network model (referred to as strong neural network model), one strong neural network model at least needs to combine 2
Neural network model is set to construct.Each setting in multiple setting neural network models involved in the embodiment of the present invention
Neural network model can be identical neural network model, be also possible to different neural network models.Initial segmentation result
Refer to that treating segmented image using blood vessel segmentation model carries out the result that blood vessel segmentation directly obtains.In the initial segmentation result
Each gray value characterizes the probability that its corresponding image is blood vessel.
Specifically, referring to fig. 2, image 201 to be split is inputted into each trained setting neural network model in advance
202, obtain the corresponding initial segmentation result 203 of each setting neural network model 202.Above-mentioned image input process can be string
Row input, is also possible to input parallel.The quantity of initial segmentation result is corresponding with the setting quantity of neural network model consistent.
Illustratively, S110 includes: and treats segmented image to carry out pretreatment generation pretreatment image, wherein pretreatment packet
Include at least one of resolution ratio resampling, gray scale normalization and piecemeal cutting;Pretreatment image is inputted at least two respectively
Trained setting neural network model in advance generates initial segmentation result corresponding with each setting neural network model.
Specifically, it before carrying out blood vessel segmentation, needs first to treat segmented image and carries out image preprocessing, pretreatment is obtained
The image obtained is pretreatment image.Here pretreatment operation may include resolution ratio resampling, gray scale normalization and piecemeal
At least one of cut.Wherein, resolution ratio resampling is by the resolution adjustment of image to be split to specified resolution ratio, example
If image to be split is liver three-dimensional CT image, the image resolution ratio of initial x-axis direction, y-axis direction and z-axis direction is distinguished
It, can be by the corresponding equal resampling of image resolution ratio of three change in coordinate axis direction for 0.5mm~0.7mm, 0.5mm~0.7mm and 1mm
To 0.5mm.Gray scale normalization is to adjust the gray value of image to specified numberical range according to certain normalization rule,
Such as [0,1], normalization rule here can be linear programming or Non―linear programming.Piecemeal cutting be by piece image according to
Specified size is cut to multiple subgraphs, to achieve the purpose that reduce the data volume of an image procossing.Obtain pretreatment figure
As after, which is inputted into each setting neural network model and is handled, obtained and setting neural network model
The initial segmentation result of respective numbers.
Illustratively, it when pretreatment includes resolution ratio resampling and gray scale normalization, treats segmented image and is located in advance
Reason includes: to treat segmented image to carry out resolution ratio resampling, obtains resampling image;According to window width and default window position is preset, really
Fixed first cut off value and the second cut off value, the first cut off value is less than the second cut off value;If the gray value of resampling image is greater than the
Gray value is then determined as normalizing the right interval value in section by two cut off value;If the gray value of resampling image is less than first point
Gray value is then determined as normalizing the left interval value in section by dividing value;If the gray value of resampling image is greater than or equal to first
Cut off value, and be less than or equal to the second cut off value, then gray value is determined as normalizing the interval value in section.
Wherein, it presets window width and default window position refers to preset window width and window position when display medical image, such as is right
In CT image for liver, default window width and default window position can be respectively set to 400 and 40.First cut off value and the second cut off value point
Be not the boundary value for carrying out gray value of image classification, the purpose of gray value classification be will the image-region comprising blood vessel with
Other image-regions distinguish, to filter out the target organ region comprising blood vessel from image to be split, such as from wait divide
It cuts and filters out liver area in image.Normalization section refers to gray scale normalization treated numberical range, such as [0,1].It is right
Interval value and left interval value refer respectively to the right boundary value and left boundary value in normalization section, such as 1 and 0.
Specifically, it by taking liver vessel segmentation as an example, first treats segmented image and carries out resolution ratio resampling, obtain resampling figure
Picture.Then, gray scale normalization processing is carried out to the resampling image.Gray scale normalization process are as follows: first according to default window width and window
Position calculates according to the first cut off value=(window position+window width/2) and the second cut off value=(window position-window width/2) and obtains the first cut off value
With the second cut off value.Then, the gray value that the gray value of resampling image is greater than the second cut off value is all set to normalize
The gray value of resampling image is all set to normalization section less than the gray value of the first cut off value by the right interval value in section
Left interval value.Finally, by gray value in resampling image between the first cut off value and the second cut off value (comprising the first cut off value
And second cut off value) between gray value, be calculated as the numerical value in normalization section according to normalization rule.The knot being arranged in this way
Fruit is that the image-region that gray value is located in normalization section in pretreatment image corresponds to liver area, other image-regions are equal
For background area, to further increase the processing speed of Subsequent vessel segmentation.
S120, using the corresponding object module weighted value of each setting neural network model, to each initial segmentation result into
Row weighting processing generates weighting segmentation result.
Wherein, object module weighted value refers to the corresponding weighted value of setting neural network model, is used to set this
The result for determining neural network model input is weighted processing.Object module weighted value is that the Boosting of the embodiment of the present invention is built
The factor that each setting neural network model is joined together in mould thought, thus constructs strong neural network model.The target mould
Type weighted value determines that one setting neural network model of every training will correspondingly really when training sets neural network model
The setting neural network model of the fixed training is constructing the object module weighted value in strong neural network model.
Specifically, referring to fig. 2, neural network model 202 is set for each, by the setting neural network model 202
Corresponding object module weighted value 204 is configured to the weight of the initial segmentation result 203 of the setting neural network model 202 generation
Value.Later, the corresponding object module weighted value 204 of each initial segmentation result 203 is multiplied, and each product that adds up
Value just obtains a weighting segmentation result 205.It is appreciated that the weighting segmentation result is still a probabilistic image.
S130, post processing of image is carried out to weighting segmentation result, generates segmentation blood-vessel image.
Specifically, weighting segmentation result 205 is carried out in post processing of image, such as binaryzation, smooth and denoising at least
A kind of processing generates segmentation blood-vessel image 206.
Illustratively, S130 includes: and carries out binary conversion treatment using predetermined probabilities threshold value to weighting segmentation result, generate
Binarization segmentation result;Denoising is carried out to binarization segmentation result, generates segmentation blood-vessel image.
Wherein, predetermined probabilities threshold value is a preset probability value, is used to carry out at binaryzation probabilistic image
Reason, such as can be 0.5.
Specifically, will weight segmentation result in gray value (being also probability value) be greater than or equal to predetermined probabilities threshold value gray scale
Value is set as 1, sets 0 for the gray value that gray value is less than predetermined probabilities threshold value, just obtains a binarization segmentation result.It
Afterwards, largest connected region is taken to the binarization segmentation result to filter the picture noise in binarization segmentation result, obtains one
Divide blood-vessel image.
The technical solution of the present embodiment passes through the setting nerve for respectively training image to be split input at least two in advance
Network model generates initial segmentation result corresponding with each setting neural network model;Utilize each setting neural network mould
The corresponding object module weighted value of type is weighted processing to each initial segmentation result, generates weighting segmentation result;To weighting point
It cuts result and carries out post processing of image, generate segmentation blood-vessel image.It realizes and Boosting idea about modeling is applied to blood vessel for the first time
Segmentation field is formed one strong by the way that multiple setting neural network models are weighted reconstruct according to its corresponding weighted value
Neural network model --- blood vessel segmentation model solves single Neural mould to carry out blood vessel segmentation to blood-vessel image
The lower problem of vessel segmentation precision obtained by type has achieved the effect that improve blood vessel segmentation precision.
Embodiment two
The present embodiment illustrates the training method of blood vessel segmentation model based on above-described embodiment.Wherein with it is above-mentioned each
Embodiment is identical or the explanation of corresponding term details are not described herein.Referring to Fig. 3, the model of blood vessel segmentation provided in this embodiment
Training method includes:
S310, the first training sample set that first setting neural network model is generated according to sample image, and utilize the
One training sample set is trained convolutional neural networks model, obtains first setting neural network model.
Wherein, sample image refers to the medical image comprising blood vessel for carrying out blood vessel segmentation model training, in order to mention
The applicability of high blood vessel segmentation model, can choose the medical image of each organ, such as can be brain medical image, liver
Medical image and four limbs medical image etc..First training sample set refers to sample used in first setting neural network model of training
This collection.
Specifically, a certain number of image datas are selected for first setting neural network model from multiple sample images
As the first training sample set.Later, which is set into quantity according to third, sequentially inputs to batch type choosing
Fixed convolutional neural networks model (such as V-net model), and using can quickly position the Dice loss functions of angiosomes into
Row model training obtains first setting neural network model.Here third setting quantity refers to preset quantitative value,
The quantity for concentrating the training sample of primary parallel input setting neural network model for characterizing training sample.The third sets number
The value of amount is related with the configuration hardware of electronic equipment of blood vessel segmentation device, such as the GPU of electronic equipment is 12G, then third
Setting quantity can be with value as 8, then inputting 8 training samples when model training parallel into model every time.
S320, neural network model is set for any of remaining each setting neural network model, according to sample
This image the second training sample set of segmentation errors Area generation corresponding with previous setting neural network model, and utilize second
Training sample set is trained convolutional neural networks model, obtains remaining setting neural network model.
Wherein, the second training sample set refers to each setting nerve after first setting neural network model of training
Sample set used in network model.
Specifically, since the embodiment of the present invention is the structure for carrying out strong neural network model based on Boosting idea about modeling
It builds, in the hope of improving blood vessel segmentation precision, therefore each setting nerve in the present embodiment after first setting neural network model
The training of network model is just trained specifically for the part of previous setting neural network model medium vessels segmentation errors,
To improve the precision of blood vessel segmentation step by step.
Since the selection of the training sample of the latter setting neural network model depends on previous setting neural network mould
The training result of type, therefore for each setting neural network model after first setting neural network model, according to serial
Mode trains a setting neural network model every time.Wherein, each setting neural network model training process is equal are as follows: root
According to the previous setting neural network model of current setting neural network model, previous setting is determined in sample image region
The region (i.e. segmentation errors region) of neural network model blood vessel segmentation mistake.Later, in the segmentation errors of multiple sample images
In region, select a certain number of image datas as the second training sample set for current setting neural network model.Finally, will
Second training sample set sequentially inputs current setting neural network model (such as V-net according to third setting quantity batch type
Model), and the Focal loss function for being able to solve difficult sample training problem is combined to carry out model training, obtain a setting
Neural network model.
It should be noted that each setting neural network model training sample image used can be phase in the present embodiment
With, it is also possible to different;In addition, the sample size that the first training sample set and the second training sample are concentrated can be equal,
It can not also wait.
Illustratively, segmentation errors region is determined as follows: using previous setting neural network model to preceding
One corresponding training sample set of setting neural network model is tested, and model test results are generated;Comparison model test knot
Fruit and goldstandard generate segmentation errors region.
Wherein, model test results refer to the output of a setting neural network model as a result, it is a binary picture
Picture, use 0 indicate that non-vascular, use 1 indicate blood vessel.Goldstandard is one accurately referring to binary image, also uses 0 to indicate non-blood
Pipe, use 1 indicate blood vessel.
Specifically, the corresponding training sample set of previous setting neural network model is inputted into the previous setting nerve net
Network model obtains and concentrates the consistent model test results of sample size with training sample.Using goldstandard create-rule, to each
The corresponding training sample of a model test results is handled, and the goldstandard of each training sample is obtained.Later, by each
The corresponding model test results of training sample generate each model test results compared with goldstandard carries out gray value one by one
Corresponding model inspection image.Such as the gray value of model test results is consistent with the gray value of goldstandard, then model accuracy rate
The gray value of image is 0;, whereas if the gray value of model test results and the gray value of goldstandard are inconsistent, then model is examined
The gray value of altimetric image is 1.Previous setting neural network is generated according to the part that gray value in each model inspection image is 1
The corresponding segmentation errors region of model.The advantages of this arrangement are as follows the accuracy in segmentation errors region is improved, thus into one
Step improves the training precision of each setting neural network model after first setting neural network model.
S330, according to each setting neural network model, with setting the one-to-one training sample set of neural network model
And goldstandard, determine the object module weighted value of each setting neural network model.
Specifically, neural network model is set for each, generates model measurement using its corresponding training sample set
As a result.Accuracy is split to model test results using goldstandard to judge, and determines that this sets according to accuracy judging result
Determine the object module weighted value of neural network model.It should be appreciated that the segmentation accuracy of a setting neural network model is higher,
Its corresponding object module weighted value is just higher.
Illustratively, S330 includes: to set neural network model for any one, using setting neural network model to setting
Determine the corresponding training sample set of neural network model to be tested, generate and each setting neural network model each mould correspondingly
Type test result;Respectively according to each model test results and goldstandard, determine that the segmentation of each setting neural network model is quasi-
True rate, and the initial model weighted value of each setting neural network model is determined according to each segmentation accuracy rate respectively;To each first
Beginning Model Weight value is normalized, and obtains the object module weighted value of each setting neural network model.
Specifically, neural network model is set for any of all setting neural network models, utilizes the setting
The sample set of neural network model generates its corresponding each model test results.Then, compare each model test results
And its corresponding goldstandard, and the ratio for dividing correct number of gray values and whole number of gray values is calculated, obtain the setting
The segmentation accuracy rate of neural network model.Later, determine that the model of the setting neural network model is weighed according to the segmentation accuracy rate
Weight values, as initial model weighted value.The initial model weighted value of each setting neural network model is determined according to the above process.
Finally, each initial model weighted value is normalized, so that the Model Weight value of each setting neural network model
The sum of be 1, the weighted value of each setting neural network model thereby determined that is just its object module weighted value.It is arranged in this way
It is advantageous in that, further increases the reasonability of object module weighted value setting.
The technical solution of the present embodiment, by raw according to the corresponding segmentation errors region of previous setting neural network model
At the second training sample set, so that currently setting neural network model is specifically for previous setting neural network model medium vessels
The part of segmentation errors is trained, and can step up the precision of blood vessel segmentation, and then is further increased based on Boosting
The segmentation precision of the blood vessel segmentation model of idea about modeling building.Pass through each setting neural network model and setting neural network
The one-to-one training sample set of model and goldstandard determine object module weighted value, so that object module weighted value is really
Surely there is theoretical foundation, improve the setting reasonability of object module weighted value.
Based on the above technical solution, the model training method of the blood vessel segmentation further include: to initial sample image
Carry out pretreatment and generate sample image, wherein pretreatment include during resolution ratio resampling, gray scale normalization and piecemeal are cut extremely
Few one kind.
Wherein, initial sample image refers to the medical image of the reconstruction initially obtained.
Specifically, pretreatment operation is carried out to every initial sample image in multiple initial sample images of acquisition,
Obtain multiple sample images.It should be noted that the step, which can generate the first training sample set and S320 in S310, generates the
It is to carry out pretreatment to all initial sample images to generate structure after whole sample images in this way before two training sample sets
Build different sample sets.The step be also possible to before S310 and S320 before execute respectively, be to handle the respectively in this way
One training sample set and the corresponding initial sample image of the second training sample set.Specific execution sequence can be according to needing to carry out
Setting.
Based on the above technical solution, first instruction of first setting neural network model is generated according to sample image
Practicing sample set includes: the stochastical sampling in sample image with preset image sizes, determines the subsample figure of the first setting quantity
Picture, as the first training sample set;
According to sample image and the corresponding training of segmentation errors Area generation second sample of previous setting neural network model
This collection includes: the segmentation errors region that sample image is determined according to previous setting neural network model;With preset image sizes,
The stochastical sampling in the segmentation errors region of sample image determines the subsample image of the second setting quantity, as the second training
Sample set.
Wherein, preset image sizes refer to preset image size, the setting of the preset image sizes and configuration blood
The hardware of the electronic equipment of pipe segmenting device is related.For example, for hardware GPU, if GPU performance is high, pre-set image ruler
It is very little to take a biggish numerical value;Conversely, preset image sizes take a lesser numerical value if GPU performance is low.Such as electricity
The GPU of sub- equipment is 12G, and sample image is 3 d medical images, then can set 96*96*96 for preset image sizes.
Subsample image refers to be obtained from sample image, and image that picture size is smaller than the picture size of sample image, the son
The image data of sample image is consistent with the image data of corresponding region in sample image.First setting quantity and the second setting number
Amount is the sample size of preset first training sample set and the second training sample set respectively, the first setting quantity and the
Two setting quantity can be equal, can not also wait.
Specifically, by multiple sample images generate the first training sample set when, and it is indirect using whole picture sample image as
The training sample that training sample is concentrated, but the first setting is randomly choosed from each width sample image according to preset image sizes
The subsample image of quantity, to form the first training sample set.Similarly, for the generation of the second training sample set, be by
The subsample figure of the second setting quantity is randomly choosed from the segmentation errors region of each width sample image according to preset image sizes
Picture, to form the second training sample set.The advantages of this arrangement are as follows the video memory GPU performance of electronic equipment can be taken into account, together
The diversity of Shi Zengjia training sample, so that the generalization ability of Subsequent vessel parted pattern is improved, so that the blood vessel that training obtains
Parted pattern is more excellent.
It is the embodiment of blood vessel segmentation device provided in an embodiment of the present invention, the blood of the device and the various embodiments described above below
Pipe dividing method belongs to the same inventive concept, the detail content of not detailed description in the embodiment of blood vessel segmentation device, can
With the embodiment with reference to above-mentioned blood vessel segmentation method.
Embodiment three
The present embodiment provides a kind of blood vessel segmentation devices, and referring to fig. 4, which specifically includes:
Initial segmentation result generation module 410, for setting the training in advance of image to be split input at least two respectively
Determine neural network model, generates at least two initial segmentation results;
Segmentation result generation module 420 is weighted, for weighing using the corresponding object module of each setting neural network model
Weight values are weighted processing to each initial segmentation result, generate weighting segmentation result, wherein object module weighted value is in training
It is determined when setting neural network model;
Divide blood-vessel image generation module 430, for carrying out post processing of image to weighting segmentation result, generates segmentation blood vessel
Image.
Optionally, initial segmentation result generation module 410 is specifically used for:
It treats segmented image and carries out pretreatment generation pretreatment image, wherein pretreatment includes resolution ratio resampling, gray scale
At least one of normalization and piecemeal cutting;
The setting neural network model that pretreatment image input at least two is trained in advance respectively, generates and each setting
The corresponding initial segmentation result of neural network model.
Optionally, segmentation blood-vessel image generation module 430 is specifically used for:
Using predetermined probabilities threshold value, binary conversion treatment is carried out to weighting segmentation result, generates binarization segmentation result;
Denoising is carried out to binarization segmentation result, generates segmentation blood-vessel image.
On the basis of above-mentioned apparatus, which further includes model training module 440, including the first model training submodule
Block, the second model training submodule and weighted value determine submodule;
Wherein, the first model training submodule is used to generate the of first setting neural network model according to sample image
One training sample set, and convolutional neural networks model is trained using the first training sample set, obtain first setting mind
Through network model;
Second model training submodule is used for for any of remaining each setting neural network model setting mind
Through network model, according to sample image and the corresponding training of segmentation errors Area generation second of previous setting neural network model
Sample set, and convolutional neural networks model is trained using the second training sample set, obtain remaining setting neural network
Model;
Weighted value determines submodule for a pair of according to each setting neural network model and setting neural network model one
The training sample set and goldstandard answered determine the object module weighted value of each setting neural network model.
Further, the first model training submodule is specifically used for:
With preset image sizes, the stochastical sampling in sample image determines the subsample image of the first setting quantity, as
First training sample set;
Second model training submodule is specifically used for:
The segmentation errors region of sample image is determined according to previous setting neural network model;
With preset image sizes, the stochastical sampling in the segmentation errors region of sample image determines the second setting quantity
Subsample image, as the second training sample set.
Further, the second model training submodule also particularly useful for:
Using previous setting neural network model to the corresponding training sample set of previous setting neural network model into
Row test, generates model test results;
Comparison model test result and goldstandard generate segmentation errors region.
Further, weighted value determines that submodule is specifically used for:
Neural network model is set for any one, it is corresponding to setting neural network model using setting neural network model
Training sample set tested, generate and each setting neural network model each model test results correspondingly;
Respectively according to each model test results and goldstandard, determine that the segmentation of each setting neural network model is accurate
Rate, and the initial model weighted value of each setting neural network model is determined according to each segmentation accuracy rate respectively;
Each initial model weighted value is normalized, the object module weight of each setting neural network model is obtained
Value.
Three a kind of blood vessel segmentation device through the embodiment of the present invention realizes Boosting idea about modeling application for the first time
It is formed to blood vessel segmentation field by the way that multiple setting neural network models are weighted reconstruct according to its corresponding weighted value
One strong neural network model --- blood vessel segmentation model solves single setting to carry out blood vessel segmentation to blood-vessel image
The lower problem of vessel segmentation precision obtained by neural network model has achieved the effect that improve blood vessel segmentation precision.
Blood vessel provided by any embodiment of the invention point can be performed in blood vessel segmentation device provided by the embodiment of the present invention
Segmentation method has the corresponding functional module of execution method and beneficial effect.
It is worth noting that, included each unit and module are only pressed in the embodiment of above-mentioned blood vessel segmentation device
It is divided, but is not limited to the above division according to function logic, as long as corresponding functions can be realized;In addition,
The specific name of each functional unit is also only for convenience of distinguishing each other, the protection scope being not intended to restrict the invention.
Example IV
Referring to Fig. 5, a kind of electronic equipment 500 is present embodiments provided comprising: one or more processors 520;Storage
Device 510, for storing one or more programs, when one or more programs are executed by one or more processors 520, so that
One or more processors 520 realize blood vessel segmentation method provided by the embodiment of the present invention, comprising:
The setting neural network model that image to be split input at least two is trained in advance respectively, generates and each setting
The corresponding initial segmentation result of neural network model;
Using the corresponding object module weighted value of each setting neural network model, each initial segmentation result is weighted
Processing generates weighting segmentation result, wherein object module weighted value determines when training sets neural network model;
Post processing of image is carried out to weighting segmentation result, generates segmentation blood-vessel image.
Certainly, it will be understood by those skilled in the art that processor 520 can also realize that any embodiment of that present invention is provided
Blood vessel segmentation method technical solution.
The electronic equipment 500 that Fig. 5 is shown is only an example, should not function and use scope to the embodiment of the present invention
Bring any restrictions.As shown in figure 5, the electronic equipment 500 includes processor 520, storage device 510, input unit 530 and defeated
Device 540 out;The quantity of processor 520 can be one or more in electronic equipment, in Fig. 5 by taking a processor 520 as an example;
Processor 520, storage device 510, input unit 530 and output device 540 in electronic equipment can by bus or other
Mode connects, in Fig. 5 for being connected by bus 550.
Storage device 510 is used as a kind of computer readable storage medium, and it is executable to can be used for storing software program, computer
Program and module, if the corresponding program instruction/module of blood vessel segmentation method in the embodiment of the present invention is (for example, blood vessel segmentation
Initial segmentation result generation module, weighting segmentation result generation module and segmentation blood-vessel image generation module in device).
Storage device 510 can mainly include storing program area and storage data area, wherein storing program area can store operation
Application program needed for system, at least one function;Storage data area, which can be stored, uses created data etc. according to terminal.
In addition, storage device 510 may include high-speed random access memory, it can also include nonvolatile memory, for example, at least
One disk memory, flush memory device or other non-volatile solid state memory parts.In some instances, storage device 510
It can further comprise the memory remotely located relative to processor 520, these remote memories can be by being connected to the network extremely
Electronic equipment.The example of above-mentioned network includes but is not limited to internet, intranet, local area network, mobile radio communication and its group
It closes.
Input unit 530 can be used for receiving number, image or the character information of input, and generate the use with electronic equipment
Family setting and the related key signals input of function control.Output device 540 may include that display screen etc. shows equipment.
Embodiment five
The present embodiment provides a kind of storage mediums comprising computer executable instructions, and computer executable instructions are by counting
For executing a kind of blood vessel segmentation method when calculation machine processor executes, this method comprises:
The setting neural network model that image to be split input at least two is trained in advance respectively, generates and each setting
The corresponding initial segmentation result of neural network model;
Using the corresponding object module weighted value of each setting neural network model, each initial segmentation result is weighted
Processing generates weighting segmentation result, wherein object module weighted value determines when training sets neural network model;
Post processing of image is carried out to weighting segmentation result, generates segmentation blood-vessel image.
Certainly, a kind of storage medium comprising computer executable instructions, computer provided by the embodiment of the present invention
Executable instruction is not limited to method operation as above, and blood vessel segmentation method provided by any embodiment of the invention can also be performed
In relevant operation.
By the description above with respect to embodiment, it is apparent to those skilled in the art that, the present invention
It can be realized by software and required common hardware, naturally it is also possible to which by hardware realization, but in many cases, the former is more
Good embodiment.Based on this understanding, technical solution of the present invention substantially in other words contributes to the prior art
Part can be embodied in the form of software products, which can store in computer readable storage medium
In, floppy disk, read-only memory (Read-Only Memory, ROM), random access memory (Random such as computer
Access Memory, RAM), flash memory (FLASH), hard disk or CD etc., including some instructions use so that an electronic equipment
(can be personal computer, server or the network equipment etc.) executes blood vessel segmentation provided by each embodiment of the present invention
Method.
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The invention is not limited to the specific embodiments described herein, be able to carry out for a person skilled in the art it is various it is apparent variation,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above embodiments to the present invention
It is described in further detail, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, also
It may include more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (10)
1. a kind of blood vessel segmentation method characterized by comprising
The setting neural network model that image to be split input at least two is trained in advance respectively generates and each setting nerve
The corresponding initial segmentation result of network model;
Using the corresponding object module weighted value of each setting neural network model, each initial segmentation result is carried out
Weighting processing generates weighting segmentation result, wherein the object module weighted value is when the training setting neural network model
It determines;
Post processing of image is carried out to the weighting segmentation result, generates segmentation blood-vessel image.
2. the method according to claim 1, wherein respectively by the training in advance of image to be split input at least two
Setting neural network model, generating corresponding with each setting neural network model initial segmentation result includes:
Pretreatment is carried out to the image to be split and generates pretreatment image, wherein pretreatment includes resolution ratio resampling, gray scale
At least one of normalization and piecemeal cutting;
The setting neural network model that pretreatment image input at least two is trained in advance respectively, generates and each setting
The corresponding initial segmentation result of neural network model.
3. the method according to claim 1, wherein carrying out post processing of image, life to the weighting segmentation result
Include: at segmentation blood-vessel image
Using predetermined probabilities threshold value, binary conversion treatment is carried out to the weighting segmentation result, generates binarization segmentation result;
Denoising is carried out to the binarization segmentation result, generates the segmentation blood-vessel image.
4. the method according to claim 1, wherein each setting neural network model and corresponding target mould
Type weighted value is trained in advance in the following way:
The first training sample set of first setting neural network model is generated according to sample image, and utilizes described first
Training sample set is trained convolutional neural networks model, obtains first setting neural network model;
Neural network model is set for any of remaining each described setting neural network model, according to sample image
The second training sample set of segmentation errors Area generation corresponding with previous setting neural network model, and utilize second instruction
Practice sample set to be trained convolutional neural networks model, obtains the remaining setting neural network model;
According to each setting neural network model, with the one-to-one training sample set of the setting neural network model with
And goldstandard, determine the object module weighted value of each setting neural network model.
5. according to the method described in claim 4, it is characterized in that, generating first setting nerve net according to sample image
First training sample set of network model includes:
With preset image sizes, the stochastical sampling in the sample image determines the subsample image of the first setting quantity, as
First training sample set;
According to sample image and corresponding the second training sample set of segmentation errors Area generation of previous setting neural network model
Include:
The segmentation errors region of the sample image is determined according to previous setting neural network model;
With the preset image sizes, the stochastical sampling in the segmentation errors region of the sample image determines the second setting number
The subsample image of amount, as second training sample set.
6. method according to claim 4 or 5, which is characterized in that the segmentation errors region is determined as follows:
The corresponding training sample set of previous setting neural network model is surveyed using previous setting neural network model
Examination generates model test results;
Compare the model test results and the goldstandard, generates the segmentation errors region.
7. according to the method described in claim 4, it is characterized in that, according to each setting neural network model, with it is described
The one-to-one training sample set of neural network model and goldstandard are set, determines the mould of each setting neural network model
Type weighted value includes:
For setting neural network model described in any one, using the setting neural network model to the setting neural network
The corresponding training sample set of model is tested, and is generated and each setting neural network model each model measurement correspondingly
As a result;
Respectively according to each model test results and the goldstandard, point of each setting neural network model is determined
Accuracy rate is cut, and determines the initial model power of each setting neural network model according to each segmentation accuracy rate respectively
Weight values;
Each initial model weighted value is normalized, the object module power of each setting neural network model is obtained
Weight values.
8. a kind of blood vessel segmentation device characterized by comprising
Initial segmentation result generation module, the setting nerve net for respectively training image to be split input at least two in advance
Network model generates at least two initial segmentation results;
Segmentation result generation module is weighted, for utilizing the corresponding object module weight of each setting neural network model
Value is weighted processing to each initial segmentation result, generates weighting segmentation result, wherein the object module weighted value
It is determined when the training setting neural network model;
Divide blood-vessel image generation module, for carrying out post processing of image to the weighting segmentation result, generates segmentation vessel graph
Picture.
9. a kind of electronic equipment, which is characterized in that the electronic equipment includes:
One or more processors;
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
The now blood vessel segmentation method as described in any in claim 1-7.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The blood vessel segmentation method as described in any in claim 1-7 is realized when being executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811577437.6A CN109671076A (en) | 2018-12-20 | 2018-12-20 | Blood vessel segmentation method, apparatus, electronic equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811577437.6A CN109671076A (en) | 2018-12-20 | 2018-12-20 | Blood vessel segmentation method, apparatus, electronic equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109671076A true CN109671076A (en) | 2019-04-23 |
Family
ID=66145837
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811577437.6A Pending CN109671076A (en) | 2018-12-20 | 2018-12-20 | Blood vessel segmentation method, apparatus, electronic equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109671076A (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110189306A (en) * | 2019-05-14 | 2019-08-30 | 上海联影智能医疗科技有限公司 | The corresponding compressing vessels of abnormal area determine method and apparatus in brain image |
CN110309855A (en) * | 2019-05-30 | 2019-10-08 | 上海联影智能医疗科技有限公司 | Training method, computer equipment and the storage medium of image segmentation |
CN110547869A (en) * | 2019-09-17 | 2019-12-10 | 上海交通大学 | Preoperative auxiliary planning device based on virtual reality |
CN110634144A (en) * | 2019-09-23 | 2019-12-31 | 武汉联影医疗科技有限公司 | Oval hole positioning method and device and storage medium |
CN110826563A (en) * | 2019-09-27 | 2020-02-21 | 五邑大学 | Finger vein segmentation method and device based on neural network and probability map model |
CN110969619A (en) * | 2019-12-19 | 2020-04-07 | 广州柏视医疗科技有限公司 | Method and device for automatically identifying primary tumor of nasopharyngeal carcinoma |
CN111062963A (en) * | 2019-12-16 | 2020-04-24 | 上海联影医疗科技有限公司 | Blood vessel extraction method, system, device and storage medium |
CN111161240A (en) * | 2019-12-27 | 2020-05-15 | 上海联影智能医疗科技有限公司 | Blood vessel classification method, computer device and readable storage medium |
CN111210431A (en) * | 2019-12-27 | 2020-05-29 | 上海联影智能医疗科技有限公司 | Blood vessel segmentation method, device, equipment and storage medium |
CN111462047A (en) * | 2020-03-06 | 2020-07-28 | 深圳睿心智能医疗科技有限公司 | Blood vessel parameter measuring method, blood vessel parameter measuring device, computer equipment and storage medium |
CN111539885A (en) * | 2020-04-21 | 2020-08-14 | 西安交通大学 | Image enhancement defogging method based on multi-scale network |
CN111640124A (en) * | 2020-05-25 | 2020-09-08 | 浙江同花顺智能科技有限公司 | Blood vessel extraction method, device, equipment and storage medium |
CN111681211A (en) * | 2020-05-18 | 2020-09-18 | 沈阳先进医疗设备技术孵化中心有限公司 | Blood vessel image processing method and device |
CN111724413A (en) * | 2020-06-19 | 2020-09-29 | 上海杏脉信息科技有限公司 | Blood vessel tracking method, coronary image segmentation method, medium and electronic equipment |
CN112734774A (en) * | 2021-01-28 | 2021-04-30 | 依未科技(北京)有限公司 | High-precision fundus blood vessel extraction method, device, medium, equipment and system |
CN113409447A (en) * | 2021-06-09 | 2021-09-17 | 广东工业大学 | Coronary artery segmentation method and device based on multi-slice combination |
CN113506302A (en) * | 2021-07-27 | 2021-10-15 | 四川九洲电器集团有限责任公司 | Interactive object updating method, device and processing system |
CN114937019A (en) * | 2022-05-30 | 2022-08-23 | 杭州健培科技有限公司 | Key point detection method and device based on self-adaptive local gray scale balance and application |
CN116071375A (en) * | 2023-03-10 | 2023-05-05 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Image segmentation method and device, storage medium and electronic equipment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101169867A (en) * | 2007-12-04 | 2008-04-30 | 北京中星微电子有限公司 | Image dividing method, image processing apparatus and system |
CN108596268A (en) * | 2018-05-03 | 2018-09-28 | 湖南大学 | A kind of data classification method |
CN108898069A (en) * | 2018-06-05 | 2018-11-27 | 辽宁石油化工大学 | Video flame detecting method based on multiple Classifiers Combination |
-
2018
- 2018-12-20 CN CN201811577437.6A patent/CN109671076A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101169867A (en) * | 2007-12-04 | 2008-04-30 | 北京中星微电子有限公司 | Image dividing method, image processing apparatus and system |
CN108596268A (en) * | 2018-05-03 | 2018-09-28 | 湖南大学 | A kind of data classification method |
CN108898069A (en) * | 2018-06-05 | 2018-11-27 | 辽宁石油化工大学 | Video flame detecting method based on multiple Classifiers Combination |
Non-Patent Citations (1)
Title |
---|
孙显: "基于对象的Boosting 方法自动提取高分辨率遥感图像中建筑物目标", 《电子与信息学报》 * |
Cited By (33)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110189306B (en) * | 2019-05-14 | 2021-02-19 | 上海联影智能医疗科技有限公司 | Method and device for determining responsible blood vessels corresponding to abnormal regions in brain image |
CN110189306A (en) * | 2019-05-14 | 2019-08-30 | 上海联影智能医疗科技有限公司 | The corresponding compressing vessels of abnormal area determine method and apparatus in brain image |
CN110309855A (en) * | 2019-05-30 | 2019-10-08 | 上海联影智能医疗科技有限公司 | Training method, computer equipment and the storage medium of image segmentation |
CN110547869A (en) * | 2019-09-17 | 2019-12-10 | 上海交通大学 | Preoperative auxiliary planning device based on virtual reality |
CN110634144A (en) * | 2019-09-23 | 2019-12-31 | 武汉联影医疗科技有限公司 | Oval hole positioning method and device and storage medium |
CN110634144B (en) * | 2019-09-23 | 2022-08-02 | 武汉联影医疗科技有限公司 | Oval hole positioning method and device and storage medium |
CN110826563A (en) * | 2019-09-27 | 2020-02-21 | 五邑大学 | Finger vein segmentation method and device based on neural network and probability map model |
CN110826563B (en) * | 2019-09-27 | 2023-11-24 | 五邑大学 | Finger vein segmentation method and device based on neural network and probability map model |
CN111062963A (en) * | 2019-12-16 | 2020-04-24 | 上海联影医疗科技有限公司 | Blood vessel extraction method, system, device and storage medium |
CN111062963B (en) * | 2019-12-16 | 2024-03-26 | 上海联影医疗科技股份有限公司 | Blood vessel extraction method, system, equipment and storage medium |
CN110969619A (en) * | 2019-12-19 | 2020-04-07 | 广州柏视医疗科技有限公司 | Method and device for automatically identifying primary tumor of nasopharyngeal carcinoma |
CN110969619B (en) * | 2019-12-19 | 2021-06-29 | 广州柏视医疗科技有限公司 | Method and device for automatically identifying primary tumor of nasopharyngeal carcinoma |
CN111210431A (en) * | 2019-12-27 | 2020-05-29 | 上海联影智能医疗科技有限公司 | Blood vessel segmentation method, device, equipment and storage medium |
CN111161240A (en) * | 2019-12-27 | 2020-05-15 | 上海联影智能医疗科技有限公司 | Blood vessel classification method, computer device and readable storage medium |
CN111161240B (en) * | 2019-12-27 | 2024-03-05 | 上海联影智能医疗科技有限公司 | Blood vessel classification method, apparatus, computer device, and readable storage medium |
CN111210431B (en) * | 2019-12-27 | 2023-06-16 | 上海联影智能医疗科技有限公司 | Blood vessel segmentation method, device, equipment and storage medium |
CN111462047A (en) * | 2020-03-06 | 2020-07-28 | 深圳睿心智能医疗科技有限公司 | Blood vessel parameter measuring method, blood vessel parameter measuring device, computer equipment and storage medium |
CN111462047B (en) * | 2020-03-06 | 2024-03-12 | 深圳睿心智能医疗科技有限公司 | Vascular parameter measurement method, vascular parameter measurement device, vascular parameter measurement computer device and vascular parameter measurement storage medium |
CN111539885B (en) * | 2020-04-21 | 2023-09-19 | 西安交通大学 | Image enhancement defogging method based on multi-scale network |
CN111539885A (en) * | 2020-04-21 | 2020-08-14 | 西安交通大学 | Image enhancement defogging method based on multi-scale network |
CN111681211B (en) * | 2020-05-18 | 2024-03-08 | 东软医疗系统股份有限公司 | Vascular image processing method and device |
CN111681211A (en) * | 2020-05-18 | 2020-09-18 | 沈阳先进医疗设备技术孵化中心有限公司 | Blood vessel image processing method and device |
CN111640124B (en) * | 2020-05-25 | 2023-06-02 | 浙江同花顺智能科技有限公司 | Blood vessel extraction method, device, equipment and storage medium |
CN111640124A (en) * | 2020-05-25 | 2020-09-08 | 浙江同花顺智能科技有限公司 | Blood vessel extraction method, device, equipment and storage medium |
CN111724413A (en) * | 2020-06-19 | 2020-09-29 | 上海杏脉信息科技有限公司 | Blood vessel tracking method, coronary image segmentation method, medium and electronic equipment |
CN112734774A (en) * | 2021-01-28 | 2021-04-30 | 依未科技(北京)有限公司 | High-precision fundus blood vessel extraction method, device, medium, equipment and system |
CN113409447A (en) * | 2021-06-09 | 2021-09-17 | 广东工业大学 | Coronary artery segmentation method and device based on multi-slice combination |
CN113506302A (en) * | 2021-07-27 | 2021-10-15 | 四川九洲电器集团有限责任公司 | Interactive object updating method, device and processing system |
CN113506302B (en) * | 2021-07-27 | 2023-12-12 | 四川九洲电器集团有限责任公司 | Interactive object updating method, device and processing system |
CN114937019B (en) * | 2022-05-30 | 2022-12-23 | 杭州健培科技有限公司 | Key point detection method and device based on self-adaptive local gray scale balance and application |
CN114937019A (en) * | 2022-05-30 | 2022-08-23 | 杭州健培科技有限公司 | Key point detection method and device based on self-adaptive local gray scale balance and application |
CN116071375B (en) * | 2023-03-10 | 2023-09-26 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Image segmentation method and device, storage medium and electronic equipment |
CN116071375A (en) * | 2023-03-10 | 2023-05-05 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Image segmentation method and device, storage medium and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109671076A (en) | Blood vessel segmentation method, apparatus, electronic equipment and storage medium | |
CN110570492B (en) | CT artifact suppression method, device and medium based on neural network | |
CN110074813B (en) | Ultrasonic image reconstruction method and system | |
US10062207B2 (en) | Method and system for reconstructing a three-dimensional model of point clouds | |
CN108829639B (en) | Magnetic resonance imaging method and equipment | |
CN110648337A (en) | Hip joint segmentation method, hip joint segmentation device, electronic apparatus, and storage medium | |
CN110458817B (en) | Medical image quality prediction method, device, equipment and storage medium | |
CN110310287A (en) | It is neural network based to jeopardize the automatic delineation method of organ, equipment and storage medium | |
CN111462264B (en) | Medical image reconstruction method, medical image reconstruction network training method and device | |
CN112288731A (en) | Blood vessel stenosis analysis method and device | |
CN110136147A (en) | A kind of method, apparatus and storage medium of the Medical Image Segmentation based on U-Net model | |
CN109712128A (en) | Feature point detecting method, device, computer equipment and storage medium | |
KR102166835B1 (en) | Method for Training Neural Network and Device Thereof | |
CN109087708A (en) | Model training method, device, equipment and storage medium for patch segmentation | |
CN110781976A (en) | Extension method of training image, training method and related device | |
CN113450328A (en) | Medical image key point detection method and system based on improved neural network | |
EP3973508A1 (en) | Sampling latent variables to generate multiple segmentations of an image | |
CN106548213A (en) | Blood vessel recognition methodss and device | |
CN112598790A (en) | Brain structure three-dimensional reconstruction method and device and terminal equipment | |
CN111210484A (en) | Medical image generation method, model training method, device and medium | |
CN112101438B (en) | Left-right eye classification method, device, server and storage medium | |
CN111063000B (en) | Magnetic resonance rapid imaging method and device based on neural network structure search | |
CN112017161A (en) | Pulmonary nodule detection method and device based on central point regression | |
CN113706546B (en) | Medical image segmentation method and device based on lightweight twin network | |
CN111968112B (en) | CT three-dimensional positioning image acquisition method and device and computer equipment |
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: 20190423 |