CN110222741A - Prediction technique, model, device, equipment and the storage medium of medical image - Google Patents
Prediction technique, model, device, equipment and the storage medium of medical image Download PDFInfo
- Publication number
- CN110222741A CN110222741A CN201910433804.3A CN201910433804A CN110222741A CN 110222741 A CN110222741 A CN 110222741A CN 201910433804 A CN201910433804 A CN 201910433804A CN 110222741 A CN110222741 A CN 110222741A
- Authority
- CN
- China
- Prior art keywords
- medical image
- input
- input unit
- characteristic pattern
- network
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
This application involves a kind of prediction technique of medical image, model, device, equipment and storage mediums to obtain the corresponding standard size characteristic pattern of medical image by the way that medical image to be inputted to the input network in preset prediction model；Wherein inputting includes at least two input units in network, and each input unit is used for the size according to medical image, obtains the corresponding standard size characteristic pattern of medical image, and then obtain the corresponding prediction result of medical image according to standard size characteristic pattern.In the application, various sizes of medical image obtains the corresponding prediction result of the medical image by preset prediction model, wherein, preset prediction model is trained model, in the training process to preset prediction model, the training sample type of prediction model is enriched by inputting network, substantially increases preset prediction model accuracy, and then improve the accuracy of the prediction result obtained by preset prediction model.
Description
Technical field
This application involves deep learning field, more particularly to a kind of prediction technique of medical image, model, device,
Equipment and storage medium.
Background technique
Medical image is in imaging process, and for different medical targets, operator can targetedly change imaging and set
Standby parameter, obtain different layers away from and size medical image.When carrying out deep learning to above-mentioned medical image, common depth
Degree learning model requires input data that must dimensionally be consistent, and otherwise will lead to above-mentioned deep learning model can not instruct
Practice.
In view of the above-mentioned problems, common processing method is by medical image according to resolution fractional, each resolution ratio is corresponding
One model of training obtains multiple models, and then carries out deep learning to corresponding medical image respectively by multiple models, obtains
To the corresponding prediction result of each medical image.For example, first by medical image according to layer away from several levels are roughly classified into, according to this
Medical image is grouped by a little levels, is trained to an independent model for each group of medical image, is obtained multiple models.
Then according to the size of medical image to be predicted, its corresponding model is selected, and then medical image to be predicted is inputted and is corresponded to
Model in, obtain the corresponding prediction result of the medical image.
However the above method is used to handle medical image, the accuracy of obtained prediction result is low.
Summary of the invention
Based on this, it is necessary to the low problem of accuracy for prediction result, provide a kind of medical image prediction technique,
Model, device, equipment and storage medium.
In a first aspect, a kind of prediction technique of medical image, this method comprises:
Medical image is inputted into the input network in preset prediction model, it is special to obtain the corresponding standard size of medical image
Sign figure；Inputting includes at least two input units in network；Each input unit is used for the size according to medical image, obtains medicine
The corresponding standard size characteristic pattern of image；
The corresponding prediction result of medical image is obtained according to standard size characteristic pattern.
It is above-mentioned in one of the embodiments, that medical image is inputted into the input network in preset prediction model, it obtains
The corresponding standard size characteristic pattern of medical image, comprising:
According to the size of medical image, the corresponding input unit of medical image is determined；
By the corresponding input unit of medical image input medical image, the corresponding standard size feature of medical image is obtained
Figure.
Each input unit includes convolution block and down-sampling block in one of the embodiments,；Different input units is corresponding
Convolution block and/or down-sampling block quantity it is different；Convolution block is for reducing the X-axis of medical image and the resolution ratio of Y-axis；Under adopt
Sample block is used to reduce X-axis, the resolution ratio of Y-axis and Z axis of medical image.
The size range of medical image handled by each input unit is according to input unit in one of the embodiments,
In convolution block and down-sampling block quantity determine.
This method in one of the embodiments, further include:
Obtain multiple medical images and the corresponding prediction result of multiple medical images；
Using multiple medical images as input, using the corresponding prediction result of multiple medical images as output, training is obtained
Preset prediction model.
It is above-mentioned using multiple medical images as input in one of the embodiments, multiple medical images are corresponding pre-
Result is surveyed as output, training obtains preset prediction model, comprising:
According to the size of medical image, multiple medical images are divided at least two medical image groups；
The medical image at least two medical image groups is randomly selected, medical image to be trained is obtained；
According to medical image to be trained prediction result corresponding with medical image to be trained, medicine to be trained is updated
The corresponding model parameter of the size of image, obtains preset prediction model.
Second aspect, a kind of prediction model of medical image, prediction model include input network and convolutional network；Input net
Network is connect with convolutional network；Network is inputted for the size according to medical image, it is special to obtain the corresponding standard size of medical image
Sign figure；Convolutional network is used to obtain the corresponding prediction result of medical image according to standard size characteristic pattern.
The third aspect, a kind of prediction meanss of medical image, which is characterized in that device includes:
Input module obtains medical image pair for medical image to be inputted to the input network in preset prediction model
The standard size characteristic pattern answered；Inputting includes at least two input units in network；Each input unit is used for according to medical image
Size, obtain the corresponding standard size characteristic pattern of medical image；
Prediction module, for obtaining the corresponding prediction result of medical image according to standard size characteristic pattern.
Fourth aspect, a kind of computer equipment, including memory and processor, the memory are stored with computer journey
The step of sequence, the processor realizes the prediction technique of above-mentioned medical image when executing the computer program.
5th aspect, a kind of computer readable storage medium are stored thereon with computer program, the computer program quilt
The step of processor realizes the prediction technique of above-mentioned medical image when executing.
Prediction technique, model, device, equipment and the storage medium of above-mentioned medical image, it is pre- by inputting medical image
If prediction model in input network, obtain the corresponding standard size characteristic pattern of medical image；It wherein inputs in network and includes
At least two input units, each input unit are used for the size according to medical image, obtain the corresponding standard size of medical image
Characteristic pattern, and then the corresponding prediction result of medical image is obtained according to standard size characteristic pattern.In the application, various sizes of doctor
It learns image and the corresponding prediction result of the medical image is obtained by preset prediction model, wherein preset prediction model is instruction
The model perfected, in the training process to preset prediction model, can by input network in each input unit will be more
The medical image of kind size is converted to standard size characteristic pattern, and then is instructed according to standard-sized characteristic pattern to prediction model
Practice, in other words, the training sample type of prediction model is enriched by inputting network, and a large amount of different type can be used
Training sample trains prediction model, substantially increases preset prediction model accuracy, and then improve through preset prediction
The accuracy for the prediction result that model obtains.
Detailed description of the invention
Fig. 1 is the schematic diagram of the application environment of the prediction technique of one embodiment traditional Chinese medicine image；
Fig. 2 is the flow diagram of the prediction technique of one embodiment traditional Chinese medicine image；
Fig. 2 a is the flow diagram of one embodiment traditional Chinese medicine image preprocessing；
Fig. 3 is the flow diagram of the prediction technique of another embodiment traditional Chinese medicine image；
Fig. 3 a is the structural schematic diagram of preset prediction model in one embodiment；
Fig. 4 is the flow diagram of the prediction technique of another embodiment traditional Chinese medicine image；
Fig. 5 is the flow diagram of the prediction technique of another embodiment traditional Chinese medicine image；
Fig. 6 is the structural schematic diagram of the prediction model of one embodiment traditional Chinese medicine image；
Fig. 7 is the structural schematic diagram of the prediction meanss of the medical image provided in one embodiment；
Fig. 8 is the structural schematic diagram of the prediction meanss of the medical image provided in another embodiment；
Fig. 9 is the structural schematic diagram of the prediction meanss of the medical image provided in another embodiment.
Specific embodiment
Prediction technique, model, device, equipment and the storage medium of medical image provided by the present application, it is intended to be predicted
The problem of the prediction result inaccuracy of medical image.Embodiment will be passed through below and in conjunction with attached drawing specifically to the technology of the application
How the technical solution of scheme and the application, which solve above-mentioned technical problem, is described in detail.These specific implementations below
Example can be combined with each other, and the same or similar concept or process may be repeated no more in certain embodiments.
The prediction technique of medical image provided in this embodiment can be adapted in application environment as shown in Figure 1.Wherein
The prediction terminal of medical image is as shown in Figure 1.The prediction terminal of medical image can be, but not limited to be various personal computers, pen
Remember that this computer, smart phone, tablet computer and portable wearable device, the embodiment of the present application are without limitation.
It should be noted that the prediction technique of medical image provided by the embodiments of the present application, executing subject can be doctor
The prediction meanss of image are learned, which can be implemented as medical image by way of software, hardware or software and hardware combining
Prediction terminal it is some or all of.
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is
Some embodiments of the present application, instead of all the embodiments.
Fig. 2 is the flow diagram of the prediction technique of one embodiment traditional Chinese medicine image.What is involved is bases for the present embodiment
The size of medical image obtains the detailed process of the corresponding prediction result of the medical image.As shown in Fig. 2, this method includes following
Step:
S101, medical image is inputted to the input network in preset prediction model, obtains the corresponding standard of medical image
Size characteristic figure；Inputting includes at least two input units in network；Each input unit is used for the size according to medical image, obtains
Obtain the corresponding standard size characteristic pattern of medical image.
Specifically, medical image may include CT scan image (Computed Tomography,
CT), magnetic resonance image (Magnetic Resonance Imaging, MRI), PET-Positron emission computed tomography image
(Positron Emission Computed Tomography, PET), the embodiment of the present application is without limitation.Meanwhile it curing
Image can be parts of body and be scanned acquisition image.For example, medical image can be scanned the brain of acquisition with brain
Image, since there are a variety of various sizes of positions for brain, when being scanned for different parts, by adjusting scanning device
Resolution ratio, can obtain different layers away from brain image.It, can be to medicine figure when medical image is inputted preset prediction model
As being pre-processed, for example, as shown in Figure 2 a, skull, adaptive cutting, x and y-axis can be carried out to medical image and is adopted again
The operations such as sample, background filling, so that all images are 224 × 224 in the size of x and y-axis, resolution ratio is 1 × 1mm, z-axis point
Resolution remains unchanged.Preset prediction model can be neural network model, which may include input net
Network, input network, which can be, improves acquisition to traditional VGG-16 network, is also possible to VGG-19, ResNet50 etc.
Heterogeneous networks structure improves acquisition, and the embodiment of the present application is without limitation.For example, traditional VGG-16 network is by 13
A 2D convolutional layer and 3 full articulamentum compositions.In order to adapt to the brain CT images of 3D, 2D convolutional layer is replaced with into 3D convolutional layer,
And corresponding adjustment has been done to other modules, form VGG3D-16 network.It may include at least two defeated in above-mentioned input network
Enter unit, can be 2 input units, is also possible to 3 input units, can also be more input units, the application
Embodiment is without limitation.Each convolution unit can carry out feature extraction to medical image, obtain according to the size of medical image
To standard size characteristic pattern, wherein standard size characteristic pattern, is the characteristic pattern of same size.It wherein can be a convolution unit
Feature extraction is carried out to the medical image of a size range, obtains the standard size characteristic pattern of the medical image.For example, input
Network includes 3 input units, respectively input unit 1, input unit 2 and input unit 3, wherein input unit 1 can be right
Layer obtains standard size characteristic pattern away from feature extraction is carried out for the brain image of 5mm；Input unit 2 can be to layer away from the brain for 3mm
Image carries out feature extraction, obtains standard size characteristic pattern；Input unit 3 can be to layer away from the brain image progress feature for 1mm
It extracts, obtains standard size characteristic pattern.
S102, the corresponding prediction result of medical image is obtained according to standard size characteristic pattern.
It specifically, can also include that convolutional network obtains on the basis of the above embodiments in above-mentioned preset prediction model
When having arrived standard size characteristic pattern, which can be input in convolutional network, obtain above-mentioned medical image
Corresponding prediction result.The convolutional network can be the convolutional network of parameter sharing, may include down-sampling layer and full articulamentum.
For example, convolutional network may include 2 down-sampling layers and 3 full articulamentums, by 2 down-sampling layers by standard size characteristic pattern
2 times of down-samplings are carried out, obtain the characteristic pattern for reducing resolution ratio, and then obtain by the classifier Softmax after 3 full articulamentums
Final prediction result.
The prediction technique of above-mentioned medical image, by the way that medical image to be inputted to the input network in preset prediction model,
Obtain the corresponding standard size characteristic pattern of medical image；Wherein inputting includes at least two input units in network, and each input is single
Member obtains the corresponding standard size characteristic pattern of medical image for the size according to medical image, and then according to standard size spy
Sign figure obtains the corresponding prediction result of medical image.In the application, various sizes of medical image passes through preset prediction model
Obtain the corresponding prediction result of the medical image, wherein preset prediction model is trained model, to preset prediction
In the training process of model, the medical image of sizes can be converted to by standard by each input unit in input network
Size characteristic figure, and then prediction model is trained according to standard-sized characteristic pattern, in other words, the training of prediction model
Sample type is enriched by inputting network, a large amount of different type training sample training prediction model can be used, significantly
Preset prediction model accuracy is improved, and then improves the accurate of the prediction result obtained by preset prediction model
Degree.
Fig. 3 is the flow diagram of the prediction technique of another embodiment traditional Chinese medicine image.What is involved is such as the present embodiment
What obtains the detailed process of the corresponding standard size characteristic pattern of medical image.As shown in figure 3, method includes the following steps:
S201, the size according to medical image determine the corresponding input unit of medical image.
Specifically, on the basis of the above embodiments, input unit is used for the size according to medical image, obtains medicine figure
As corresponding standard size characteristic pattern.The size for the medical image that each input unit can be handled has range.For example, when defeated
Entering network includes 3 input units, and respectively input unit 1, input unit 2 and input unit 3, input unit 1 can be handled
The size range of medical image be medical image that layer can be handled away from the brain image in 0.5mm-1.5mm, input unit 2
The size range for the medical image that size range, which is layer, can be handled away from the brain image in 2mm-4mm, input unit 3 be layer away from
The brain image of 4.1mm-6mm.According to the size of medical image, the corresponding input unit of medical image is determined, can be selection institute
State the medical image that can be handled size range include the medical image size input unit, be medical image it is corresponding defeated
Enter unit.
Optionally, each input unit includes convolution block and down-sampling block；The corresponding convolution block of different input units and/or
The quantity of down-sampling block is different；Convolution block is for reducing the X-axis of medical image and the resolution ratio of Y-axis；Down-sampling block is for reducing
X-axis, the resolution ratio of Y-axis and Z axis of medical image.
Specifically, each input unit includes convolution block Conv Block and down-sampling block Down Block, different inputs
The quantity of the corresponding convolution block of unit and/or down-sampling block is different.Convolution block can be by 2 to 3 layer 3 × 3 × 3 of convolutional layer and 1
The maximum pond layer composition of layer, maximum pond layer can carry out 2 times of down-samplings to x and y-axis, and z-axis remains unchanged, i.e., convolution block is contracting
The X-axis of small medical image and the resolution ratio of Y-axis；Down-sampling block and convolution block can have identical structure, but in down-sampling block
Maximum pond layer can carry out 2 times of down-samplings to 3 all axis, i.e., down-sampling block is for reducing the X-axis, Y-axis and Z of medical image
The resolution ratio of axis.
Optionally, the size range of medical image handled by each input unit be according in input unit convolution block and
What the quantity of down-sampling block determined.
Specifically, on the basis of the above embodiments, it is used to reduce the X-axis of medical image and point of Y-axis due to convolution block
Resolution；Down-sampling block is used to reduce X-axis, the resolution ratio of Y-axis and Z axis of medical image, then can pass through setting different number
Convolution block and down-sampling block determine the size range of medical image handled by input unit.The size area master of medical image
It concentrates on Z axis, different convolution block and down-sampling block is set, the X-axis and Y-axis of medical image can be reduced by convolution block
Resolution ratio and down-sampling block be used to reduce the X-axis of medical image, Y-axis and Z axis resolution ratio characteristic, will be various sizes of
Medical Image Processing is standard-sized characteristic pattern.For example, as shown in Figure 3a, input network includes 3 input units, wherein most
The input unit in left side is input unit 1, and intermediate input unit is input unit 2, and the input unit on right side is input unit
3, input unit 1 includes 3 convolution block Conv Block, and input unit 2 includes adopting under 2 convolution block Conv Block and 1
Sample block Down Block, input unit 3 includes 1 convolution block Conv Block and 2 down-sampling block Down Block, wherein most
The latter down-sampling block and input unit 2 are multiplexed.The medicine figure that the size of the medical image of the processing of input unit 1 is thickness 5mm
Picture, corresponding resolution ratio are 224 × 224 × 28；The medicine figure that the size of the medical image of the processing of input unit 2 is thickness 3mm
Picture, corresponding resolution ratio are 224 × 224 × 56；The medicine figure that the size of the medical image of the processing of input unit 3 is thickness 1mm
Picture, corresponding resolution ratio are 224 × 224 × 112.As described in Fig. 3 a, the medical image that resolution ratio is 224 × 224 × 28 passes through
3 convolution block Conv Bloc in input unit 1 reduce X-axis, the resolution ratio of Y-axis, obtain the mark that resolution ratio is 28 × 28 × 28
Object staff cun characteristic pattern；The medical image that resolution ratio is 224 × 224 × 56 passes through 2 convolution block Conv in input unit 2
Block reduces X-axis, the resolution ratio of Y-axis, obtains 56 × 56 × 56 characteristic pattern, and then pass through 1 down-sampling block Down Block
X-axis, the resolution ratio of Y-axis and Z axis are reduced, the standard size characteristic pattern that resolution ratio is 28 × 28 × 28 is obtained；Resolution ratio be 224 ×
224 × 112 medical image input unit 3 reduces X-axis, the resolution ratio of Y-axis by 1 convolution block Conv Block, obtains 112
× 112 × 112 characteristic pattern, and then X-axis, the resolution ratio of Y-axis and Z axis are reduced by 2 down-sampling block Down Block, it obtains
The standard size characteristic pattern that resolution ratio is 28 × 28 × 28.In other words, 3 convolution block Conv are set by input unit
Block, then the input unit can handle the medical image that resolution ratio is 224 × 224 × 28, i.e. input unit handles medicine figure
The size range of picture is thickness 5mm；2 convolution block Conv Block and 1 down-sampling block Down are set by input unit
Block, then the input unit can handle the medical image that resolution ratio is 224 × 224 × 56, the i.e. medical image of thickness 3mm；
1 convolution block Conv Block and 2 down-sampling block Down Block are set by input unit, then the input unit can be located
Manage the medical image that resolution ratio is 224 × 224 × 112, the i.e. medical image of thickness 1mm.
S202, medical image is inputted to the corresponding input unit of medical image, obtains the corresponding standard size of medical image
Characteristic pattern.
It specifically, on the basis of the above embodiments, can basis when medical image being inputted its corresponding input unit
The setting of input unit obtains the corresponding standard size feature of the medical image.For example, medical image is the doctor of thickness 5mm
Image, corresponding resolution ratio 224 × 224 × 28 are learned, corresponding input unit is the input unit of 3 convolution blocks, passes through 3
Convolution unit reduces the resolution ratio of X-axis and Y-axis, obtains 28 × 28 × 28 standard size characteristic pattern.
The prediction technique of above-mentioned medical image determines the corresponding input unit of medical image according to the size of medical image,
And by the corresponding input unit of medical image input medical image, the corresponding standard size characteristic pattern of medical image is obtained, so that
Various sizes of medical image can input its corresponding input unit, and it is special to obtain standard size by its corresponding input unit
Sign figure, allows to train the same prediction model by the medical image of sizes, so that a kind of ruler wherein
In the case that very little medical image sample size is less, stable prediction model can be also obtained, is obtained to reduce sample image
The manpower and material resources spent when taking thereby reduce the training difficulty of preset prediction model.
On the basis of the above embodiments, terminal can obtain above-mentioned preset prediction model by training.Below by
Embodiment illustrated in fig. 4 is described in detail.
Fig. 4 is the flow diagram of the prediction technique of another embodiment traditional Chinese medicine image.What is involved is such as the present embodiment
What obtains preset prediction model detailed process by training.As shown in figure 4, this method is further comprising the steps of:
S301, multiple medical images and the corresponding prediction result of multiple medical images are obtained.
Specifically, when obtaining medical image, available various types of medical images.For example, when medical image is
When the CT brain phantom that CT scan equipment scans, by obtaining two veteran radiologists to cerebral haemorrhage feelings
Condition carries out two classification markers, obtains the prediction result to above-mentioned CT brain phantom.The prediction result is for indicating CT brain phantom
Inside whether there is hemorrhagic focus.It, can be by fine tuning resolution ratio and by medicine figure when medical image is other kinds of medical image
The size adjusting of picture is at the correspondingly-sized for inputting network.
S302, it is trained using multiple medical images as input using the corresponding prediction result of multiple medical images as output
Obtain preset prediction model.
Specifically, on the basis of the above embodiments, can be using multiple medical images as input, it can also be to more
A medical image is grouped, and obtaining one or more medical image set cooperations is input, and the embodiment of the present application does not limit this
System.During being specifically trained to prediction model, it can be and preset one group of training parameter, input multiple medicine figures
Picture obtains corresponding prediction result by the prediction model of pre-set one group of training parameter, by the prediction result with it is corresponding
The corresponding actual prediction result of medical image compare, according to the comparing result, adjusting training parameter, until by pre-
The prediction result that model obtains is surveyed, actual prediction result corresponding with medical image meets preset requirement, and as target is instructed
Practice parameter and preset prediction model is determined according to the target training parameter.
The prediction technique of above-mentioned medical image, by obtaining multiple medical images and the corresponding prediction of multiple medical images
As a result, and using multiple medical images as input, using the corresponding prediction result of multiple medical images as exporting, it is trained obtain it is pre-
If prediction model.So that preset prediction model is by multiple medical images and the corresponding prediction knot of multiple medical images
What fruit obtained, the accuracy of preset prediction model is improved, and then improve the prediction obtained according to preset prediction model
As a result accuracy.
Fig. 5 is the flow diagram of the prediction technique of another embodiment traditional Chinese medicine image.What is involved is ends for the present embodiment
How end is using multiple medical images as input, and using the corresponding prediction result of multiple medical images as output, training obtains pre-
If prediction model detailed process.As shown in figure 5, S302 is " using multiple medical images as input, by multiple medical images pair
The prediction result answered obtains preset prediction model as output, training " a kind of possible implementation method the following steps are included:
Multiple medical images are divided at least two medical image groups by S401, the size according to medical image.
Specifically, on the basis of the above embodiments, there are different sizes for medical image, then can be according to different rulers
It is very little, medical image is divided at least two medical image groups, wherein the size of the medical image in each medical image group is at this
In the corresponding size range of medical image group.For example, 5 medical images include medical image 1, medical image 2, medical image
3, medical image 4 and medical image 5, wherein medical image 1 is the medical image of thickness 5mm, and medical image 2 is thickness 4.5mm
Medical image, medical image 3 be thickness 2.7mm medical image, medical image 4 be thickness 3mm medical image, medicine figure
Picture 5 is the medical image of thickness 3.4mm, above-mentioned 5 medical images can be divided into 2 medical image groups, i.e. medical image group 1
With medical image group 2, its corresponding size range of medical image group 1 is thickness 4mm-6mm, which includes medicine
Image 1 and medical image 2, the corresponding size range of medical image group 2 are thickness 2.5mm-3.5mm, which includes
Medical image 3, medical image 4 and medical image 5.
S402, medical image at least two medical image groups is randomly selected, obtains medical image to be trained.
It specifically, can be by randomly selecting the medical image at least two medical image groups, as doctor to be trained
Learn image.For example, can be by first randomly selecting the medical image group at least two medical image groups, then from the medical image
Medical image is randomly selected in group, obtains medical image to be trained.
S403, basis medical image to be trained prediction result corresponding with medical image to be trained, update wait train
Medical image the corresponding model parameter of size, obtain preset prediction model.
Specifically, on the basis of the above embodiments, the convolutional network in preset prediction model can be parameter sharing
Convolutional network, may include multiple groups model parameter in the convolutional network, wherein a group model parameter can correspond to a kind of size
Medical image, then the medical image to be trained that can be obtained through the foregoing embodiment, and medical image to be trained is corresponding
Prediction result, update the corresponding model parameter of size of medical image to be trained, so by multiple groups it is various sizes of to
Trained medical image is trained in turn, updates corresponding model parameter, obtains preset prediction model.
Multiple medical images are divided at least two according to the size of medical image by the prediction technique of above-mentioned medical image
Medical image group, and the medical image at least two medical image groups is randomly selected, medical image to be trained is obtained, in turn
According to medical image to be trained prediction result corresponding with medical image to be trained, the ruler of medical image to be trained is updated
Very little corresponding model parameter, obtains preset prediction model, so that the model parameter in default prediction model is according to different rulers
What very little medical image was updated, it is targetedly to update, improves the accuracy of updated model parameter, Jin Erti
The high accuracy of preset prediction model.
Although should be understood that each step in the flow chart of Fig. 2-5 according to the instruction of arrow, is successively shown,
It is these steps is not that the inevitable sequence according to arrow instruction successively executes.Unless expressly stating otherwise herein, these steps
There is no stringent sequences to limit for rapid execution, these steps can execute in other order.Moreover, in Fig. 2-5 at least
A part of step may include that perhaps these sub-steps of multiple stages or stage are not necessarily in same a period of time to multiple sub-steps
Quarter executes completion, but can execute at different times, the execution in these sub-steps or stage be sequentially also not necessarily according to
Secondary progress, but in turn or can replace at least part of the sub-step or stage of other steps or other steps
Ground executes.
Fig. 6 is the prediction model of one embodiment traditional Chinese medicine image, and prediction model includes input network and convolutional network；It is defeated
Enter network to connect with convolutional network；Network is inputted for the size according to medical image, obtains the corresponding gauge of medical image
Very little characteristic pattern；Convolutional network is used to obtain the corresponding prediction result of medical image according to standard size characteristic pattern.
The prediction model of medical image provided in an embodiment of the present invention can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
Fig. 7 is the structural schematic diagram of the prediction meanss of the medical image provided in one embodiment, as shown in fig. 7, the doctor
The prediction meanss for learning image include: input module 10 and prediction module 20, in which:
Input module 10 obtains medical image for medical image to be inputted to the input network in preset prediction model
Corresponding standard size characteristic pattern；Inputting includes at least two input units in network；Each input unit is used for according to medicine figure
The size of picture obtains the corresponding standard size characteristic pattern of medical image；
Prediction module 20, for obtaining the corresponding prediction result of medical image according to standard size characteristic pattern.
The prediction meanss of medical image provided in an embodiment of the present invention can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
Fig. 8 is the structural schematic diagram of the prediction meanss of the medical image provided in another embodiment, implementation shown in Fig. 7
On the basis of example, as shown in figure 8, input module 10 comprises determining that unit 101 and input unit 102, in which:
Determination unit 101 determines the corresponding input unit of medical image for the size according to medical image；
Input unit 102, for it is corresponding to obtain medical image by the corresponding input unit of medical image input medical image
Standard size characteristic pattern.
In one embodiment, each input unit includes convolution block and down-sampling block；The corresponding volume of different input units
Block and/or the quantity of down-sampling block are different；Convolution block is for reducing the X-axis of medical image and the resolution ratio of Y-axis；Down-sampling block
For reducing the X-axis of medical image, the resolution ratio of Y-axis and Z axis.
In one embodiment, the size range of medical image handled by each input unit is according in input unit
What the quantity of convolution block and down-sampling block determined.
The prediction meanss of medical image provided in an embodiment of the present invention can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
Fig. 9 is the structural schematic diagram of the prediction meanss of the medical image provided in another embodiment, in Fig. 7 or Fig. 8 institute
On the basis of showing embodiment, as shown in figure 9, the prediction meanss of medical image further include training pattern 30, training module 30 includes
Acquiring unit 301 and training unit 302, in which:
Acquiring unit 301, for obtaining multiple medical images and the corresponding prediction result of multiple medical images；
Training unit 302, for using multiple medical images as input, the corresponding prediction result of multiple medical images to be made
For output, training obtains preset prediction model.
In one embodiment, training unit 302 is specifically used for the size according to medical image, by multiple medical images point
For at least two medical image groups；The medical image at least two medical image groups is randomly selected, medicine to be trained is obtained
Image；According to medical image to be trained prediction result corresponding with medical image to be trained, medicine figure to be trained is updated
The corresponding model parameter of the size of picture, obtains preset prediction model.
It should be noted that Fig. 9 is based on being shown on the basis of Fig. 8, certain Fig. 9 can also be based on the knot of Fig. 7
Structure is shown, and is only a kind of example here.
The prediction meanss of medical image provided in an embodiment of the present invention can execute above method embodiment, realize former
Reason is similar with technical effect, and details are not described herein.
A kind of specific restriction of prediction meanss about medical image may refer to above to the prediction side of medical image
The restriction of method, details are not described herein.Modules in the prediction meanss of above-mentioned medical image can fully or partially through software,
Hardware and combinations thereof is realized.Above-mentioned each module can be embedded in the form of hardware or independently of the processor in computer equipment
In, it can also be stored in a software form in the memory in computer equipment, in order to which processor calls execution above each
The corresponding operation of module.
In one embodiment, a kind of computer equipment is provided, which can be terminal device, inside
Structure chart can be as shown in Figure 1.The computer equipment include by system bus connect processor, memory, network interface,
Display screen and input unit.Wherein, the processor of the computer equipment is for providing calculating and control ability.The computer equipment
Memory include non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system and calculating
Machine program.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.It should
The network interface of computer equipment is used to communicate with external terminal by network connection.The computer program is executed by processor
When to realize the prediction technique of medical image a kind of.The display screen of the computer equipment can be liquid crystal display or electronic ink
Water display screen, the input unit of the computer equipment can be the touch layer covered on display screen, be also possible to computer equipment
Key, trace ball or the Trackpad being arranged on shell can also be external keyboard, Trackpad or mouse etc..
It will be understood by those skilled in the art that structure shown in Fig. 1, only part relevant to application scheme is tied
The block diagram of structure does not constitute the restriction for the computer equipment being applied thereon to application scheme, specific computer equipment
It may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.
In one embodiment, a kind of terminal device, including memory and processor are provided, the memory is stored with
Computer program, the processor perform the steps of when executing the computer program
Medical image is inputted into the input network in preset prediction model, it is special to obtain the corresponding standard size of medical image
Sign figure；Inputting includes at least two input units in network；Each input unit is used for the size according to medical image, obtains medicine
The corresponding standard size characteristic pattern of image；
The corresponding prediction result of medical image is obtained according to standard size characteristic pattern.
In one embodiment, the ruler according to medical image is also performed the steps of when processor executes computer program
It is very little, determine the corresponding input unit of medical image；By the corresponding input unit of medical image input medical image, medicine figure is obtained
As corresponding standard size characteristic pattern.
In one embodiment, each input unit includes convolution block and down-sampling block；The corresponding volume of different input units
Block and/or the quantity of down-sampling block are different；Convolution block is for reducing the X-axis of medical image and the resolution ratio of Y-axis；Down-sampling block
For reducing the X-axis of medical image, the resolution ratio of Y-axis and Z axis.
In one embodiment, the size range of medical image handled by each input unit is according in input unit
What the quantity of convolution block and down-sampling block determined.
In one embodiment, it is also performed the steps of when processor executes computer program and obtains multiple medical images,
And the corresponding prediction result of multiple medical images；Using multiple medical images as input, by the corresponding prediction of multiple medical images
As a result as output, training obtains preset prediction model.
In one embodiment, the ruler according to medical image is also performed the steps of when processor executes computer program
It is very little, multiple medical images are divided at least two medical image groups；Randomly select the medicine figure at least two medical image groups
Picture obtains medical image to be trained；According to medical image to be trained prediction result corresponding with medical image to be trained,
The corresponding model parameter of size for updating medical image to be trained, obtains preset prediction model.
Terminal device provided in this embodiment, implementing principle and technical effect are similar with above method embodiment, herein
It repeats no more.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program performs the steps of when being executed by processor
Medical image is inputted into the input network in preset prediction model, it is special to obtain the corresponding standard size of medical image
Sign figure；Inputting includes at least two input units in network；Each input unit is used for the size according to medical image, obtains medicine
The corresponding standard size characteristic pattern of image；
The corresponding prediction result of medical image is obtained according to standard size characteristic pattern.
In one embodiment, the ruler according to medical image is performed the steps of when computer program is executed by processor
It is very little, determine the corresponding input unit of medical image；By the corresponding input unit of medical image input medical image, medicine figure is obtained
As corresponding standard size characteristic pattern.
In one embodiment, each input unit includes convolution block and down-sampling block；The corresponding volume of different input units
Block and/or the quantity of down-sampling block are different；Convolution block is for reducing the X-axis of medical image and the resolution ratio of Y-axis；Down-sampling block
For reducing the X-axis of medical image, the resolution ratio of Y-axis and Z axis.
In one embodiment, the size range of medical image handled by each input unit is according in input unit
What the quantity of convolution block and down-sampling block determined.
In one embodiment, it is performed the steps of when computer program is executed by processor and obtains multiple medical images,
And the corresponding prediction result of multiple medical images；Using multiple medical images as input, by the corresponding prediction of multiple medical images
As a result as output, training obtains preset prediction model.
In one embodiment, the ruler according to medical image is performed the steps of when computer program is executed by processor
It is very little, multiple medical images are divided at least two medical image groups；Randomly select the medicine figure at least two medical image groups
Picture obtains medical image to be trained；According to medical image to be trained prediction result corresponding with medical image to be trained,
The corresponding model parameter of size for updating medical image to be trained, obtains preset prediction model.
Computer readable storage medium provided in this embodiment, implementing principle and technical effect and above method embodiment
Similar, details are not described herein.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of prediction technique of medical image, which is characterized in that the described method includes:
Medical image is inputted into the input network in preset prediction model, it is special to obtain the corresponding standard size of the medical image
Sign figure；It include at least two input units in the input network；Each input unit is used for according to the medical image
Size obtains the corresponding standard size characteristic pattern of the medical image；
The corresponding prediction result of the medical image is obtained according to the standard size characteristic pattern.
2. method according to claim 1, which is characterized in that it is described medical image is inputted it is defeated in preset prediction model
Enter network, obtain the corresponding standard size characteristic pattern of the medical image, comprising:
According to the size of the medical image, the corresponding input unit of the medical image is determined；
The medical image is inputted into the corresponding input unit of the medical image, obtains the corresponding gauge of the medical image
Very little characteristic pattern.
3. method according to claim 2, which is characterized in that each input unit includes convolution block and down-sampling block；No
The quantity of the same corresponding convolution block of input unit and/or the down-sampling block is different；The convolution block is for reducing institute
State the X-axis of medical image and the resolution ratio of Y-axis；The down-sampling block is used to reduce the X-axis of the medical image, Y-axis and Z axis
Resolution ratio.
4. method according to claim 3, which is characterized in that the ruler of the medical image handled by each input unit
Very little range is determined according to the quantity of the convolution block and the down-sampling block in the input unit.
5. any one of -4 the method according to claim 1, which is characterized in that the method also includes:
Obtain multiple medical images and the corresponding prediction result of the multiple medical image；
Using the multiple medical image as input, using the corresponding prediction result of the multiple medical image as output, training
Obtain the preset prediction model.
6. method according to claim 5, which is characterized in that it is described using the multiple medical image as input, it will be described
The corresponding prediction result of multiple medical images obtains the preset prediction model as output, training, comprising:
According to the size of the medical image, the multiple medical image is divided at least two medical image groups；
A medical image in at least two medical images group is randomly selected, medical image to be trained is obtained；
According to the medical image to be trained and the corresponding prediction result of the medical image to be trained, update described wait instruct
The corresponding model parameter of the size of experienced medical image obtains the preset prediction model.
7. a kind of prediction model of medical image, which is characterized in that the model includes input network and convolutional network；It is described defeated
Entering network includes at least two input units；The input network is connect with the convolutional network；Each input unit is used for
According to the size of medical image, the corresponding standard size characteristic pattern of the medical image is obtained；The convolutional network is used for basis
The standard size characteristic pattern obtains the corresponding prediction result of the medical image.
8. a kind of prediction meanss of medical image, which is characterized in that described device includes:
Input module obtains the medical image pair for medical image to be inputted to the input network in preset prediction model
The standard size characteristic pattern answered；It include at least two input units in the input network；Each input unit is used for basis
The size of the medical image obtains the corresponding standard size characteristic pattern of the medical image；
Prediction module, for obtaining the corresponding prediction result of the medical image according to the standard size characteristic pattern.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In when the processor executes the computer program the step of any one of realization claim 1-6 the method.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method of any of claims 1-6 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910433804.3A CN110222741A (en) | 2019-05-23 | 2019-05-23 | Prediction technique, model, device, equipment and the storage medium of medical image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910433804.3A CN110222741A (en) | 2019-05-23 | 2019-05-23 | Prediction technique, model, device, equipment and the storage medium of medical image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110222741A true CN110222741A (en) | 2019-09-10 |
Family
ID=67818306
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910433804.3A Pending CN110222741A (en) | 2019-05-23 | 2019-05-23 | Prediction technique, model, device, equipment and the storage medium of medical image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110222741A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192255A (en) * | 2019-12-30 | 2020-05-22 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
| CN111243026A (en) * | 2020-01-19 | 2020-06-05 | 武汉联影智融医疗科技有限公司 | Anatomical mark point positioning method and device, computer equipment and storage medium |
| CN113627492A (en) * | 2021-07-20 | 2021-11-09 | 东软医疗系统股份有限公司 | Method for determining size of scanning object, scanning method, scanning device and electronic equipment |
| CN111192255B (en) * | 2019-12-30 | 2024-04-26 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780448A (en) * | 2016-12-05 | 2017-05-31 | 清华大学 | A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features |
| CN106991666A (en) * | 2017-02-24 | 2017-07-28 | 中国科学院合肥物质科学研究院 | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations |
| CN107967484A (en) * | 2017-11-14 | 2018-04-27 | 中国计量大学 | A kind of image classification method based on multiresolution |
| CN109754447A (en) * | 2018-12-28 | 2019-05-14 | 上海联影智能医疗科技有限公司 | Image generating method, device, equipment and storage medium |
-
2019
- 2019-05-23 CN CN201910433804.3A patent/CN110222741A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780448A (en) * | 2016-12-05 | 2017-05-31 | 清华大学 | A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features |
| CN106991666A (en) * | 2017-02-24 | 2017-07-28 | 中国科学院合肥物质科学研究院 | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations |
| CN107967484A (en) * | 2017-11-14 | 2018-04-27 | 中国计量大学 | A kind of image classification method based on multiresolution |
| CN109754447A (en) * | 2018-12-28 | 2019-05-14 | 上海联影智能医疗科技有限公司 | Image generating method, device, equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| QI DOU ET AL: "Multilevel Contextual 3-D CNNs for False Positive Reduction in Pulmonary Nodule Detection", 《IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111192255A (en) * | 2019-12-30 | 2020-05-22 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
| CN111192255B (en) * | 2019-12-30 | 2024-04-26 | 上海联影智能医疗科技有限公司 | Index detection method, computer device, and storage medium |
| CN111243026A (en) * | 2020-01-19 | 2020-06-05 | 武汉联影智融医疗科技有限公司 | Anatomical mark point positioning method and device, computer equipment and storage medium |
| CN111243026B (en) * | 2020-01-19 | 2023-09-22 | 武汉联影智融医疗科技有限公司 | Anatomical landmark point positioning method, apparatus, computer device, and storage medium |
| CN113627492A (en) * | 2021-07-20 | 2021-11-09 | 东软医疗系统股份有限公司 | Method for determining size of scanning object, scanning method, scanning device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102016211642B4 (en) | PATCH MEMORY SYSTEM | |
| CN111192356B (en) | Method, device, equipment and storage medium for displaying region of interest | |
| RU2739713C1 (en) | Training annotation of objects in an image | |
| CN108960062A (en) | Correct method, apparatus, computer equipment and the storage medium of invoice image | |
| CN109754447A (en) | Image generating method, device, equipment and storage medium | |
| CN111161269B (en) | Image segmentation method, computer device, and readable storage medium | |
| CN109493417A (en) | Three-dimension object method for reconstructing, device, equipment and storage medium | |
| DE102014014944A1 (en) | Fill a target area using transformations | |
| CN110197491A (en) | Image partition method, device, equipment and storage medium | |
| CN110222741A (en) | Prediction technique, model, device, equipment and the storage medium of medical image | |
| WO2007141995A1 (en) | Display processing device | |
| CN110210519B (en) | Classification method, computer device, and storage medium | |
| CN110363774A (en) | Image partition method, device, computer equipment and storage medium | |
| US11683438B2 (en) | Systems and methods to semi-automatically segment a 3D medical image using a real-time edge-aware brush | |
| CN110211130A (en) | Image partition method, computer equipment and storage medium | |
| DE10234135A1 (en) | Systems and methods for linking a graphic display and an n-dimensional data structure in a graphic user interface | |
| CN107392893A (en) | Tissue density's analysis method and system | |
| DE112022001343T5 (en) | Proposal for an edge using a neural network | |
| CN111127487A (en) | Real-time multi-tissue medical image segmentation method | |
| CN110309855A (en) | Training method, computer equipment and the storage medium of image segmentation | |
| CN113744843A (en) | Medical image data processing method and device, computer equipment and storage medium | |
| CN112967331B (en) | Image processing method, electronic equipment and storage medium | |
| CN110390680A (en) | Image partition method, computer equipment and storage medium | |
| CN114663449A (en) | Road defect area calculation method and device, computer equipment and storage medium | |
| CN110766653B (en) | Image segmentation method and device, computer equipment and storage medium |
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 |