CN110390660A - A kind of medical image jeopardizes organ automatic classification method, equipment and storage medium - Google Patents
A kind of medical image jeopardizes organ automatic classification method, equipment and storage medium Download PDFInfo
- Publication number
- CN110390660A CN110390660A CN201810335549.4A CN201810335549A CN110390660A CN 110390660 A CN110390660 A CN 110390660A CN 201810335549 A CN201810335549 A CN 201810335549A CN 110390660 A CN110390660 A CN 110390660A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- depth convolutional
- human body
- medical image
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to medical image and field of computer technology, it is related to a kind of medical image and jeopardizes organ automatic classification method, equipment and storage medium.The classification method includes the following steps: to pre-process medical image to be sorted;Pretreated medical image is input in trained CNN and carries out classification prediction, wherein the CNN training method includes human body being sequentially divided into from top to bottom several regions;Human body tag along sort corresponding with above-mentioned cut zone is made using one-hot coding;It will be used to be input in CNN and be trained after trained image data carries out pretreatment and data enhancing, determine network weight.Whole body provided by the invention based on CNN jeopardizes organ automatic classification method with very high accuracy, and the image layer where only needing very short time whole body can be mainly jeopardized organ predicts and, the working efficiency of doctor can be greatly improved, while providing valuable time again for the timely treatment of patient.
Description
Technical field
The invention belongs to medical images and field of computer technology, are related to a kind of medicine based on depth convolutional neural networks
Image jeopardizes organ automatic classification method, equipment and storage medium.
Background technique
During hospital carries out radiotherapy to patient, delineating to target target area is often related to.Doctor master at present
If being delineated using manual mode to target target area, delineate by hand time-consuming and laborious, influences the working efficiency of doctor, more influence
The timely treatment of patient.
With the development of depth learning technology, the Automatic medical image segmentation technology based on depth learning technology becomes medicine
The popular research in field.But a set of medical image, such as CT images or magnetic resonance imaging etc. usually contain many human bodies and jeopardize device
Official, firstly the need of the layer where filtering out target organ before divide automatically to target organ, when we to human body very
When more jeopardizing organ and have the demand divided automatically, manually go to screen it is each jeopardize organ where layer just seem time-consuming and laborious,
Therefore the whole body based on organization of human body jeopardizes organ and classifies automatically just and is particularly important.
Summary of the invention
It is an object of the invention to provide a kind of based on depth convolutional Neural net to overcome the defect of the above-mentioned prior art
The medical image of network jeopardizes organ automatic classification method, equipment and storage medium.
To achieve the above object, the invention adopts the following technical scheme:
Human medical image (such as CT images, magnetic resonance imaging, PET image etc.) is divided into ten according to organization of human body by the present invention
Class (dividing method is successively to divide nine classes from top to bottom and medical image other than human body is divided into one kind) classifies to this ten kinds
Label is made using one-hot coding, each section of such human body has the label of oneself, then builds using the present invention
Depth convolutional neural networks are trained, and trained model and weight are saved, and then can use the model of preservation
New medical image is predicted with weight, prediction result is exactly classification belonging to the medical image.
A kind of human body based on depth convolutional neural networks jeopardizes organ automatic classification method, suitable for holding in calculating equipment
Row, includes the following steps:
(1) medical image to be sorted is pre-processed;
(2) pretreated medical image is input in trained depth convolutional neural networks and carries out classification prediction;
The wherein training method of the depth convolutional neural networks, comprising the following steps:
(i) human body is sequentially divided into from top to bottom several regions;
(ii) human body tag along sort corresponding with above-mentioned cut zone is made using one-hot coding;
(iii) trained data will be used for and carries out interpolation processing;
(iv) according to the picture position where the organ of target area, training data is cut into fixed dimension;
(v) data enhancing is carried out to the training data cut, to enhance the extensive energy of depth convolutional neural networks model
Power;
(vi) the enhanced training data of data is input in depth convolutional neural networks model and is trained, work as verifying
When the loss value of data set is less than or equal to given threshold, trained depth convolutional neural networks model is obtained.
It is further preferred that the medical image is CT image, nuclear-magnetism image or PET image etc. in step (1).
In step (1), the pretreatment is interpolation processing and/or cuts out the size in the direction medical image x, y to be sorted
It is cut to consistent with the size in the direction training image x, y, and/or the size in the direction medical image x, y to be sorted is cut into and is instructed
The size for practicing the direction image x, y is consistent.
The depth convolutional neural networks include input layer, convolutional layer, maximum pond layer, merge layer, output layer, wherein
Convolutional layer, maximum pond layer, merging layer are hidden layer.
In the depth convolutional neural networks model, each convolutional layer includes weights initialisation function and activation letter
Number.
It is further preferred that the weights initialisation function is selected from He_normal function, Random_normal function
Or Glorot_normal function;The activation primitive is selected from SeLU function, ReLU function, PReLU function or ELU function.
Wherein He_normal function are as follows:
Wherein i indicates i-th layer of neural network,
W(i)Indicate i-th layer of weight,
n(i)Indicate the quantity of i-th layer of neuron.
Activation primitive SeLU activation primitive are as follows:
Wherein, α=1.6732632423543772848170429916717;
λ=1.0507009873554804934193349852946.
The loss function of the depth convolutional neural networks intersects entropy function using more classification, is defined as:
WhereinakIndicate the output valve of k-th of neuron, zkIndicate the input of k-th of neuron, e
Indicate natural constant, ykIndicate the corresponding true value of k-th of neuron.
In step (iii), the interpolation processing be x in each training data image, y are unified in direction interpolation be it is fixed
Size.
In step (v), the data enhancing includes the rotation for surrounding image center, x, the translation in y-axis direction.Such as
By image x, the shake in y-axis direction can artificially create some new data around rotation of central point etc., to enhance
The generalization ability of model also can be very good to identify when model encounters the data such as some head deflections.
In step (vi), the network parameter threshold value set is less than or equal to 0.01.
In step (vi), training process uses the optimization method of AdaDelta or Adam.
In step (vi), the training includes propagated forward and backpropagation, and a propagated forward and backpropagation are
An iteration, preferably, the number of iterations is more than or equal to 10 times the present invention, it is further preferred that the number of iterations is 10~100 times,
It is highly preferred that the number of iterations is 20~50 times.The accuracy rate of trained depth convolutional neural networks model tends towards stability.Once
Propagated forward and backpropagation in iteration cover all hidden layers.
The present invention also provides a kind of calculating equipment, comprising:
One or more processors;
Memory;And
One or more programs, wherein one or more of programs are stored in the memory and are configured as by one
A or multiple processors execute, and one or more programs include for the above-mentioned medicine based on depth convolutional neural networks
Image jeopardizes the instruction of organ automatic classification method.
The present invention also provides a kind of computer readable storage medium for storing one or more programs, described one or more
A program includes instruction, and described instruction is suitable for being loaded by memory and executing the above-mentioned medicine figure based on depth convolutional neural networks
As jeopardizing organ automatic classification method.
The present invention has following technical effect that
It is to realize the basis for jeopardizing organs automatic segmentation, and doctor screens a danger by hand that whole body jeopardizes organ classifies automatically
And organ needs 1~3 minute, wastes doctor's valuable time.The whole body danger based on depth convolutional neural networks that invention provides
And organ automatic classification method only need 1 second or so whole body can be mainly jeopardized organ where layer predict come, with hand
Work screening is compared, and the time shortens about 98%, this can greatly improve the working efficiency of doctor, while being again the timely of patient
Treatment provides valuable time.In addition, automatic classification method provided by the invention also has very high accuracy.
Detailed description of the invention
Fig. 1 (a) is that the medical image based on depth convolutional neural networks jeopardizes device in a preferred embodiment of the invention
Official's automatic classification method flow chart;
It (b) is depth convolutional neural networks training method flow chart in a preferred embodiment of the invention.
Fig. 2 is human body classification method schematic diagram from top to bottom in a preferred embodiment of the invention.
Fig. 3 is depth convolutional neural networks structure chart in a preferred embodiment of the invention.
Fig. 4 is the multireel product module block structure in a preferred embodiment of the invention in depth convolutional neural networks structure chart
Schematic diagram.
Fig. 5 is the residual error link block knot in a preferred embodiment of the invention in depth convolutional neural networks structure chart
Structure schematic diagram.
Fig. 6 is training image in a preferred embodiment of the invention.
To be based in a preferred embodiment of the invention, depth convolutional neural networks are of all categories to human body to jeopardize organ to Fig. 7
Prediction result, to predict that accurate quantity explains.
To be based in a preferred embodiment of the invention, depth convolutional neural networks are of all categories to human body to jeopardize organ to Fig. 8
Prediction result is explained with predictablity rate.
Specific embodiment
The present invention is further illustrated below in conjunction with drawings and examples.
The present invention provides a kind of human body based on depth convolutional neural networks and jeopardizes organ automatic classification method.According to human body
Structure is by human medical image (such as CT images) divide into several classes (for example, dividing method is nine successively divided from top to bottom
Class and medical image other than human body is divided into the tenth class), to this ten kinds classification using one-hot coding make label, such human body
Each section have oneself label, be then trained using the depth convolutional neural networks that the present invention is built, instructing
The model and weight perfected are saved, and then can be carried out using the model and weight saved to new medical image pre-
It surveys, prediction result is exactly classification belonging to the medical image.
A kind of human body based on depth convolutional neural networks jeopardizes organ automatic classification method (shown in such as Fig. 1 (a)), is suitable for
It executes, includes the following steps: in calculating equipment
Pretreatment 210 is carried out to medical image to be sorted;
Preferably, the pretreatment be interpolation processing, it is further preferred that pretreatment further include by medical image x to be sorted,
The size in the direction y is cut into consistent with the size in the direction training image x, y.
In the present embodiment preferably, the size in the direction medical image x, y to be sorted after cutting is cut into be schemed with training
As the size in the direction x, y is consistent.
Pretreated medical image is input in trained depth convolutional neural networks and carries out classification prediction 220;
The medical image can be CT image, nuclear-magnetism image or PET image etc..
Fig. 3 is depth convolutional neural networks structure chart in an illustrative embodiment of the invention, the depth convolutional Neural net
It is (as shown in Figure 4) for unit with single multireel volume module in network structure, residual error company is carried out by interval of two multireel volume modules
It connects, constitutes residual error link block (as shown in Figure 5);In figure, arrow indicates the direction of network data conduction, and dotted line indicates that residual error connects
It connects.Wherein, 1*1 convolution kernel, 3*3 convolution kernel and 5*5 convolution kernel have been used simultaneously in multireel product module block structure as shown in Figure 4,
Different convolution kernels can provide different receptive fields, can extract the different feature of image, increase the rich of network.It is residual
Poor link block is attached by the output input and after two multireel volume modules, and usual neural network is with the number of plies
Increase the problem of will appear gradient disperse, and residual error connection can effectively solve this problem.Depth convolution mind shown in Fig. 3
It is included 7 sequentially connected residual error link blocks in network, it will be appreciated by those skilled in the art that as needed, can adjust
The quantity of residual error link block in CNN network;With the increase of residual error module number in network, required calculating memory is bigger.
The wherein training method of depth convolutional neural networks (shown in such as Fig. 1 (b)), comprising the following steps:
Human body is sequentially divided into from top to bottom several regions 301;In an illustrative embodiment of the invention, by human body
Medical image is in turn divided into ten classifications according to organization of human body from top to bottom.Of all categories and its range is as shown in table 1 and Fig. 2:
Wherein, Fig. 2 be human body sagittal plain and coronal bitmap, illustrated using this two width figure Whole Body classify situation.It is more than the crown
Range without organization of human body be classification 1, from the crown to eyes on top layer be classification 2, the top layer under top layer to eyes from eyes
For classification 3, top layer is classification 4 under top layer to cerebellum under eyes, under cerebellum top layer to lower jaw the last layer be classification 5, from
Lower jaw the last layer is classification 6 to lung top layer, is classification 7 from lung top layer to stomach top layer, is classification 8 from stomach top layer to kidney bottom, from
Kidney bottom is classification 9 to bladder top, is classification 10 from bladder top to foot.
Table 1
The one-to-one label 302 of human body classification corresponding with above-mentioned cut zone is made using one-hot coding;In this hair
It is as shown in table 2 human body corresponding one-hot coding label of all categories in a bright exemplary embodiment.
Table 2
Classification | Label |
1 | 1000000000 |
2 | 0100000000 |
3 | 0010000000 |
4 | 0001000000 |
5 | 0000100000 |
6 | 0000010000 |
7 | 0000001000 |
8 | 0000000100 |
9 | 0000000010 |
10 | 0000000001 |
It will be used for trained data and carry out interpolation processing 303;The interpolation processing is x, the direction y in each training data image
Unified interpolation is fixed size x0Mm, y0Mm, the direction x and the direction y are as shown in Figure 6 in training data image;Preferably, x0Mm,
y0Mm is a statistical result according to multiple hospitals data format, most values of Qu Ge hospital.
According to the picture position where the organ of target area, training data is cut into fixed dimension 304;
Data enhancing is carried out to the training data cut, to enhance the generalization ability of depth convolutional neural networks model
305;Preferably, data enhancing includes rotation or the x around image center, the one or more such as the translation in y-axis direction.
Such as by image x, the shake in y-axis direction can artificially create some new data around rotation of central point etc., thus
The generalization ability for enhancing model, also can be very good to identify when model encounters the data such as some head deflections.
The enhanced training data of data is input in depth convolutional neural networks model and is trained, verify data is worked as
When the loss value of collection is less than or equal to given threshold, trained depth convolutional neural networks model 306 is obtained.
Further, the threshold value of the validation data set loss value of setting is less than or equal to 0.0.
Further, the optimization method of CNN neural network training process can use the optimization side of AdaDelta or Adam
Method.
Training includes propagated forward and backpropagation, and a propagated forward and backpropagation are an iteration, the present embodiment
In preferably, the number of iterations be more than or equal to 10 times, it is further preferred that the number of iterations be 10~100 times, it is highly preferred that iteration
Number is 20~50 times.The accuracy rate of trained depth convolutional neural networks model tends towards stability.Forward direction in an iteration
It propagates and backpropagation covers all hidden layers.
In an illustrative embodiment of the invention, depth convolutional neural networks include input layer, convolutional layer, maximum pond
Layer merges layer, and output layer, wherein convolutional layer, maximum pond layer, merging layer are hidden layer.The depth convolutional neural networks model
In, each convolutional layer includes weights initialisation function and activation primitive.
It is further preferred that it will be understood by those skilled in the art that wherein weights initialisation function can be selected from He_
Normal function, Random_normal function or Glorot_normal function;Activation primitive can be selected from SeLU function, ReLU
Function, PReLU function or ELU function.
Wherein He_normal function are as follows:
Wherein i indicates i-th layer of neural network,
W(i)Indicate i-th layer of weight,
n(i)Indicate the quantity of i-th layer of neuron.
Activation primitive SeLU activation primitive are as follows:
Wherein, α=1.6732632423543772848170429916717;
λ=1.0507009873554804934193349852946.
The loss function of the depth convolutional neural networks intersects entropy function using more classification, is defined as:
WhereinakIndicate the output valve of k-th of neuron, zkIndicate the input of k-th of neuron, e is indicated
Natural constant, ykIndicate the corresponding true value of k-th of neuron.
Fig. 7 and Fig. 8 is to utilize depth convolutional neural networks shown in Fig. 3 in the present embodiment (Adam network optimized approach, setting
The threshold value of validation data set loss jeopardizes organ prediction result to be 0.01) of all categories to human body.Wherein, ordinate is really to mark
Label, abscissa are prediction labels, and Fig. 6 diagonal line is the correct quantity of prediction, by taking classification 2 as an example, altogether to 1289 (9+1269+
11) the CT image of example classification 2 is classified automatically, and it is 1269 that wherein CNN neural network forecast, which is the quantity of classification 2, in the present embodiment
Example, being predicted as classification 1 is 9, and being predicted as classification 3 is 11, and predictablity rate isAccordingly,
Diagonal line display predicts that correct ratio is 0.98 in Fig. 8.It can be seen that prediction result most from the prediction result in Fig. 7 and Fig. 8
The accuracy rate of low classification 7 is 0.92, and preferably up to reaching 0.99, total test set quantity is for classification 1 and 6 result of classification
10483, prediction correct number is 10061, and consensus forecast accuracy rate is 0.96.
In addition, only needing 1s or so that can realize that the automatic of medical image is divided using the CNN network in the embodiment of the present invention
Class (wherein hardware configuration are as follows: GPU:GEFORCE GTX 1080;CPU:Intel Xeon E3-1230 v5;Memory: 16G).
Embodiment 2
A kind of calculating equipment, comprising:
One or more processors;
Memory;And
One or more programs, the storage of wherein one or more programs in the memory and be configured as by one or
Multiple processors execute, and said one or multiple programs include jeopardizing device for the medical image based on depth convolutional neural networks
The instruction of official's automatic classification method, this method comprises the following steps:
(1) medical image to be sorted is pre-processed;
(2) pretreated medical image is input in trained depth convolutional neural networks and carries out classification prediction;
The wherein training method of above-mentioned depth convolutional neural networks, comprising the following steps:
(i) human body is sequentially divided into from top to bottom several regions;
(ii) human body tag along sort corresponding with above-mentioned cut zone is made using one-hot coding;
(iii) trained data will be used for and carries out interpolation processing;
(iv) according to the picture position where the organ of target area, training data is cut into fixed dimension;
(v) data enhancing is carried out to the training data cut, to enhance the extensive energy of depth convolutional neural networks model
Power;
(vi) the enhanced training data of data is input in depth convolutional neural networks model and is trained, instructed
The depth convolutional neural networks model perfected.
Embodiment 3
A kind of computer readable storage medium storing one or more programs, wherein one or more programs include referring to
It enables, which is suitable for being loaded by memory and being executed the medical image based on depth convolutional neural networks and jeopardize organ and classify automatically
Method, this method comprises the following steps:
(1) medical image to be sorted is pre-processed;
(2) pretreated medical image is input in trained depth convolutional neural networks and carries out classification prediction;
The wherein training method of above-mentioned depth convolutional neural networks, comprising the following steps:
(i) human body is sequentially divided into from top to bottom several regions;
(ii) human body tag along sort corresponding with above-mentioned cut zone is made using one-hot coding;
(iii) trained data will be used for and carries out interpolation processing;
(iv) according to the picture position where the organ of target area, training data is cut into fixed dimension;
(v) data enhancing is carried out to the training data cut, to enhance the extensive energy of depth convolutional neural networks model
Power;
(vi) the enhanced training data of data is input in depth convolutional neural networks model and is trained, instructed
The depth convolutional neural networks model perfected.
Those skilled in the art will understand that can be carried out adaptively to the module in the equipment in embodiment
Change and they are arranged in one or more devices different from this embodiment.It can be the module or list in embodiment
Member or component are combined into a module or unit or component, and furthermore they can be divided into multiple submodule or subelement or
Sub-component.Other than such feature and/or at least some of process or unit exclude each other, it can use any
Combination is to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosed
All process or units of what method or apparatus are combined.Unless expressly stated otherwise, this specification is (including adjoint power
Benefit require, abstract and attached drawing) disclosed in each feature can carry out generation with an alternative feature that provides the same, equivalent, or similar purpose
It replaces.
As used in this, unless specifically stated, come using ordinal number " first ", " second ", " third " etc.
Description plain objects, which are merely representative of, is related to the different instances of similar object, and is not intended to imply that the object being described in this way must
Must have the time it is upper, spatially, sequence aspect or given sequence in any other manner.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments mean it is of the invention
Within the scope of and form different embodiments.For example, in the following claims, embodiment claimed is appointed
Meaning one of can in any combination mode come using.
It should be appreciated that various technologies described herein are realized together in combination with hardware or software or their combination.From
And some aspects or part of the process and apparatus of the present invention or the process and apparatus of the present invention can take the tangible matchmaker of insertion
It is situated between, such as the program code in floppy disk, CD-ROM, hard disk drive or other any machine readable storage mediums (refers to
Enable) form, wherein when program is loaded into the machine of such as computer etc, and when being executed by the machine, which becomes real
Trample equipment of the invention.
By way of example and not limitation, computer-readable medium includes computer storage media and communication media.It calculates
Machine storage medium stores the information such as computer readable instructions, data structure, program module or other data.Communication media one
As with the modulated message signals such as carrier wave or other transmission mechanisms embody computer readable instructions, data structure, program
Module or other data, and including any information transmitting medium.Above any combination is also included within computer-readable
Within the scope of medium.
This hair can be understood and applied the above description of the embodiments is intended to facilitate those skilled in the art
It is bright.Person skilled in the art obviously easily can make various modifications to these embodiments, and described herein
General Principle is applied in other embodiments without having to go through creative labor.Therefore, the present invention is not limited to implementations here
Example, those skilled in the art's announcement according to the present invention, improvement and modification made without departing from the scope of the present invention all should be
Within protection scope of the present invention.
Claims (10)
1. a kind of human body based on depth convolutional neural networks jeopardizes organ automatic classification method, suitable for being held in calculating equipment
Row, characterized by the following steps:
(1) medical image to be sorted is pre-processed;
(2) pretreated medical image is input in trained depth convolutional neural networks and carries out classification prediction;
The wherein training method of the depth convolutional neural networks, comprising the following steps:
(i) human body is sequentially divided into from top to bottom several regions;
(ii) human body tag along sort corresponding with above-mentioned cut zone is made using one-hot coding;
(iii) trained data will be used for and carries out interpolation processing;
(iv) according to the picture position where the organ of target area, training data is cut into fixed dimension;
(v) data enhancing is carried out to the training data cut, to enhance the generalization ability of depth convolutional neural networks model;
(vi) the enhanced training data of data is input in depth convolutional neural networks model and is trained, work as verify data
When the loss value of collection is less than or equal to given threshold, trained depth convolutional neural networks model is obtained.
2. the human body according to claim 1 based on depth convolutional neural networks jeopardizes organ automatic classification method, special
Sign is: in step (1), the medical image is CT image, nuclear-magnetism image or PET image;
Or the pretreatment is interpolation processing, and/or the size in the direction medical image x, y to be sorted is cut into and is schemed with training
As the size in the direction x, y is consistent.
3. the human body according to claim 1 based on depth convolutional neural networks jeopardizes organ automatic classification method, special
Sign is: the depth convolutional neural networks include input layer, convolutional layer, maximum pond layer, merge layer, output layer, wherein
Convolutional layer, maximum pond layer, merging layer are hidden layer.
4. the human body according to claim 1 or 3 based on depth convolutional neural networks jeopardizes organ automatic classification method,
Be characterized in that: in the depth convolutional neural networks, each convolutional layer includes weights initialisation function and activation primitive.
5. the human body according to claim 4 based on depth convolutional neural networks jeopardizes organ automatic classification method, special
Sign is: the weights initialisation function is selected from He_normal function, Random_normal function or Glorot_normal
Function;
The activation primitive is selected from SeLU function, ReLU function, PReLU function or ELU function.
6. the human body according to claim 1 or 3 based on depth convolutional neural networks jeopardizes organ automatic classification method,
Be characterized in that: the loss function of the depth convolutional neural networks intersects entropy function using more classification, is defined as:
WhereinakIndicate the output valve of k-th of neuron, zkIndicate the input of k-th of neuron, e indicates nature
Constant, ykIndicate the corresponding true value of k-th of neuron.
7. the human body according to claim 1 based on depth convolutional neural networks jeopardizes organ automatic classification method, special
Sign is: in step (iii), the interpolation processing be x in each training data image, y are unified in direction interpolation be it is fixed
Size;
Or in step (v), the data enhancing includes the rotation for surrounding image center, x, the translation in y-axis direction.
8. the human body according to claim 1 based on depth convolutional neural networks jeopardizes organ automatic classification method, special
Sign is: in step (vi), loss threshold value is less than or equal to 0.01;
Or training process uses the optimization method of AdaDelta or Adam;
Or the training includes propagated forward and backpropagation, a propagated forward and backpropagation are an iteration, once
Propagated forward and backpropagation in iteration cover all hidden layers.
9. a kind of calculating equipment, comprising:
One or more processors;
Memory;And
One or more programs, wherein the storage of one or more of programs in the memory and be configured as by one or
Multiple processors execute, and one or more programs include for any described based on depth in the claims 1-8
The medical image of degree convolutional neural networks jeopardizes the instruction of organ automatic classification method.
10. a kind of computer readable storage medium for storing one or more programs, one or more programs include referring to
Enable, described instruction be suitable for load by memory and execute in the claims 1-8 it is any described in based on depth convolutional Neural
The medical image of network jeopardizes organ automatic classification method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810335549.4A CN110390660A (en) | 2018-04-16 | 2018-04-16 | A kind of medical image jeopardizes organ automatic classification method, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810335549.4A CN110390660A (en) | 2018-04-16 | 2018-04-16 | A kind of medical image jeopardizes organ automatic classification method, equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110390660A true CN110390660A (en) | 2019-10-29 |
Family
ID=68283735
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810335549.4A Pending CN110390660A (en) | 2018-04-16 | 2018-04-16 | A kind of medical image jeopardizes organ automatic classification method, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110390660A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112183718A (en) * | 2020-08-31 | 2021-01-05 | 华为技术有限公司 | Deep learning training method and device for computing equipment |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102609680A (en) * | 2011-12-22 | 2012-07-25 | 中国科学院自动化研究所 | Method for detecting human body parts by performing parallel statistical learning based on three-dimensional depth image information |
CN105740892A (en) * | 2016-01-27 | 2016-07-06 | 北京工业大学 | High-accuracy human body multi-position identification method based on convolutional neural network |
CN106204587A (en) * | 2016-05-27 | 2016-12-07 | 孔德兴 | Multiple organ dividing method based on degree of depth convolutional neural networks and region-competitive model |
CN107194338A (en) * | 2017-05-14 | 2017-09-22 | 北京工业大学 | Traffic environment pedestrian detection method based on human body tree graph model |
CN107403201A (en) * | 2017-08-11 | 2017-11-28 | 强深智能医疗科技(昆山)有限公司 | Tumour radiotherapy target area and jeopardize that organ is intelligent, automation delineation method |
US20180060652A1 (en) * | 2016-08-31 | 2018-03-01 | Siemens Healthcare Gmbh | Unsupervised Deep Representation Learning for Fine-grained Body Part Recognition |
-
2018
- 2018-04-16 CN CN201810335549.4A patent/CN110390660A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102609680A (en) * | 2011-12-22 | 2012-07-25 | 中国科学院自动化研究所 | Method for detecting human body parts by performing parallel statistical learning based on three-dimensional depth image information |
CN105740892A (en) * | 2016-01-27 | 2016-07-06 | 北京工业大学 | High-accuracy human body multi-position identification method based on convolutional neural network |
CN106204587A (en) * | 2016-05-27 | 2016-12-07 | 孔德兴 | Multiple organ dividing method based on degree of depth convolutional neural networks and region-competitive model |
US20180060652A1 (en) * | 2016-08-31 | 2018-03-01 | Siemens Healthcare Gmbh | Unsupervised Deep Representation Learning for Fine-grained Body Part Recognition |
CN107194338A (en) * | 2017-05-14 | 2017-09-22 | 北京工业大学 | Traffic environment pedestrian detection method based on human body tree graph model |
CN107403201A (en) * | 2017-08-11 | 2017-11-28 | 强深智能医疗科技(昆山)有限公司 | Tumour radiotherapy target area and jeopardize that organ is intelligent, automation delineation method |
Non-Patent Citations (2)
Title |
---|
ICTCXQ: "TensorFlow 多分类标签转换成One-hot", 《CSDN博客-HTTPS://BLOG.CSDN.NET/ICTCXQ/ARTICLE/DETAILS/78545282》 * |
YUNCHAO WEI 等: "HCP: A Flexible CNN Framework for Multi-Label Image Classification", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112183718A (en) * | 2020-08-31 | 2021-01-05 | 华为技术有限公司 | Deep learning training method and device for computing equipment |
WO2022042713A1 (en) * | 2020-08-31 | 2022-03-03 | 华为技术有限公司 | Deep learning training method and apparatus for use in computing device |
CN112183718B (en) * | 2020-08-31 | 2023-10-10 | 华为技术有限公司 | Deep learning training method and device for computing equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Salehi et al. | A CNN model: earlier diagnosis and classification of Alzheimer disease using MRI | |
CN109522973A (en) | Medical big data classification method and system based on production confrontation network and semi-supervised learning | |
CN107506797A (en) | One kind is based on deep neural network and multi-modal image alzheimer disease sorting technique | |
CN107715314B (en) | Deep learning-based radiotherapy system and method | |
CN106204587A (en) | Multiple organ dividing method based on degree of depth convolutional neural networks and region-competitive model | |
CN110310287A (en) | It is neural network based to jeopardize the automatic delineation method of organ, equipment and storage medium | |
CN108171711A (en) | A kind of infant's brain Magnetic Resonance Image Segmentation method based on complete convolutional network | |
CN106682616A (en) | Newborn-painful-expression recognition method based on dual-channel-characteristic deep learning | |
CN107977969A (en) | A kind of dividing method, device and the storage medium of endoscope fluorescence image | |
CN106815481A (en) | A kind of life cycle Forecasting Methodology and device based on image group | |
CN107358600A (en) | Automatic hook Target process, device and electronic equipment in radiotherapy planning | |
CN110070540A (en) | Image generating method, device, computer equipment and storage medium | |
CN109589092A (en) | Method and system are determined based on the Alzheimer's disease of integrated study | |
CN109472263A (en) | A kind of brain magnetic resonance image dividing method of the global and local information of combination | |
CN109308477A (en) | A kind of medical image automatic division method, equipment and storage medium based on rough sort | |
CN112614133B (en) | Three-dimensional pulmonary nodule detection model training method and device without anchor point frame | |
CN110363760A (en) | The computer system of medical image for identification | |
CN114782350A (en) | Multi-modal feature fusion MRI brain tumor image segmentation method based on attention mechanism | |
CN109215040A (en) | A kind of tumor of breast dividing method based on multiple dimensioned weighting study | |
CN109461161A (en) | A method of human organ in medical image is split based on neural network | |
CN110059656A (en) | The leucocyte classification method and system for generating neural network are fought based on convolution | |
CN110223275A (en) | A kind of cerebral white matter fiber depth clustering method of task-fMRI guidance | |
CN110097128A (en) | Medical Images Classification apparatus and system | |
CN110232721A (en) | A kind of crisis organ delineates the training method and device of model automatically | |
CN107958472A (en) | PET imaging methods, device, equipment and storage medium based on sparse projection data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191029 |
|
RJ01 | Rejection of invention patent application after publication |