CN109685143A - A kind of thyroid gland technetium sweeps the identification model construction method and device of image - Google Patents
A kind of thyroid gland technetium sweeps the identification model construction method and device of image Download PDFInfo
- Publication number
- CN109685143A CN109685143A CN201811599551.9A CN201811599551A CN109685143A CN 109685143 A CN109685143 A CN 109685143A CN 201811599551 A CN201811599551 A CN 201811599551A CN 109685143 A CN109685143 A CN 109685143A
- Authority
- CN
- China
- Prior art keywords
- image
- technetium
- thyroid gland
- alexnet
- sweeps
- 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
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Abstract
The embodiment of the present invention provides the identification model construction method and device that a kind of thyroid gland technetium sweeps image.Method includes: to obtain multiple thyroid gland technetiums to sweep image and sweep image to every technetium and be labeled, and obtains corresponding type label；Thyroid gland technetium after one mark is swept in image input depth convolutional neural networks AlexNet and exports prediction result, penalty values are calculated according to the type label that the thyroid gland technetium after prediction result and mark sweeps image, and the parameter in AlexNet is optimized using penalty values；Thyroid gland technetium after reacquiring a mark sweeps the AlexNet that image is input to after optimization, and is trained again to the AlexNet after optimization, obtains identification model until the condition for reaching deconditioning.Device is for executing the above method.The embodiment of the present invention can be realized by constructing identification model and sweep image to technetium to be identified and identify, so that indirect labor identifies, improve the efficiency and accuracy rate of diagosis.
Description
Technical field
The present invention relates to technical field of image processing, and the identification model of image is swept in particular to a kind of thyroid gland technetium
Construction method and device.
Background technique
The disease incidence of thyroid disease is rising year by year, and can detect can by regional 5% women of rich iodine and 1% male in the world
The thyroid nodule touched, in iodine deficient area, its disease incidence is then higher.Single photon emission computed tomography (Single-
Photon Emission Computed Tomography, SPECT) as Diagnosis of Thyroid Diseases main detection methods it
One, it is that discovery thyroid gland lesion is common and reliable method.Therefore a large amount of doubtful suspicious patients for suffering from thyroid disease need to do
The technetium of SPECT sweep inspection come further auxiliary diagnosis thyroid gland whether dysfunction.
It is to carry out being identified and given out knot to sweep image according to thyroid gland technetium by the mode of the artificial diagosis of doctor in the prior art
Fruit.The thyroid gland technetium of current Nuclear Medicine Dept sweeps that image diagosis amount is larger, the artificial diagosis checked relative to Nuclear Medicine Dept other
For many inspection items, thyroid gland organ morphology is single, and image lesion characteristics are more concentrated, and degree of variation is lower, diagosis difficulty
It is easier to, and the description lexical repetition frequency reported is higher.Nuclear Medicine Dept's diagosis doctor's daily requirement carries out a large amount of each organoid items
The work of purpose diagosis, thyroid gland diagosis need to be found with eyes as high-intensitive repeated labor, doctor during diagosis
Subtle lesion, the error and asthenopia that inevitably human eyesight generates under the amount of image information of magnanimity are held
Easily there is wrong diagnosis and escape phenomenon.
Summary of the invention
In view of this, a kind of identification model building side for being designed to provide thyroid gland technetium and sweeping image of the embodiment of the present invention
Method and device, to solve the above technical problems.
In a first aspect, the embodiment of the invention provides the identification model construction methods that a kind of thyroid gland technetium sweeps image, comprising:
It obtains multiple thyroid gland technetiums and sweeps image, and image is swept to every technetium and is labeled, obtain corresponding type label；
Thyroid gland technetium after one mark is swept in image input depth convolutional neural networks AlexNet, the AlexNet
Prediction result is exported, loss is calculated according to the type label that the thyroid gland technetium after the prediction result and the mark sweeps image
Value, and the parameter in the AlexNet is optimized using the penalty values；
Reacquire one mark after thyroid gland technetium sweep image be input to optimization after AlexNet, and to optimization after
AlexNet is trained again, until reaching the condition of deconditioning, obtains identification model.
Further, the depth convolutional neural networks AlexNet includes five convolutional layers and three full articulamentums；Its
In,
First convolutional layer includes the first convolution core attributes, activation primitive, pond layer and standardization；
Second convolutional layer includes the second convolution core attributes, the activation primitive, the pond layer and the standardization；
Third convolutional layer and Volume Four lamination include third convolution core attributes and the activation primitive；
5th convolutional layer includes the 5th convolution core attributes, the activation primitive and the pond layer；
The activation primitive are as follows: f (x)=max (0, x)；
The standardized function are as follows:
Wherein,Pass through the activation primitive in image at i-th of convolution kernel (x, y) coordinate to sweep in thyroid gland technetium
Output,For thyroid gland technetium sweep in image at i-th of convolution kernel (x, y) coordinate after standardization as a result, k, n, α, β
For hyper parameter.
Further, before obtaining multiple training images, the method, further includes:
Obtain the initial pictures of one or more first size；
Sliding interception is carried out to every initial pictures according to the second size, obtains multiple the corresponding interceptions of every initial pictures
Image, wherein the second size is less than first size；
Mirror image operation is carried out to every interception image, obtains corresponding mirror image, and all interception images and mirror image
Multiple thyroid gland technetiums described in image construction sweep image.
Further, after obtaining multiple thyroid gland technetiums and sweeping image, the method, further includes:
Image is swept to every thyroid gland technetium and carries out pretreatment operation, wherein the pretreatment operation includes: that picture format turns
It changes, the conversion of data cleansing and pixel value.
Further, the method, further includes:
It obtains multiple verifying technetiums and sweeps image, and image is swept according to the verifying technetium, the identification model is verified.
Second aspect, the embodiment of the invention provides the identification model construction devices that a kind of thyroid gland technetium sweeps image, comprising:
First obtains module, sweeps image for obtaining multiple thyroid gland technetiums, and sweep image to every technetium and be labeled, and obtains
Corresponding type label；
First training module, for the thyroid gland technetium after a mark to be swept image input depth convolutional neural networks
In AlexNet, the AlexNet exports prediction result, sweeps figure according to the thyroid gland technetium after the prediction result and the mark
The type label of picture calculates penalty values, and is optimized using the penalty values to the parameter in the AlexNet；
Second training module, for reacquire one mark after thyroid gland technetium sweep image be input to optimization after
AlexNet, and the AlexNet after optimization is trained again, until reaching the condition of deconditioning, obtain identification mould
Type.
Further, the depth convolutional neural networks AlexNet includes five convolutional layers and three full articulamentums；Its
In,
First convolutional layer includes the first convolution core attributes, activation primitive, pond layer and standardization；
Second convolutional layer includes the second convolution core attributes, the activation primitive, the pond layer and the standardization；
Third convolutional layer and Volume Four lamination include third convolution core attributes and the activation primitive；
5th convolutional layer includes the 5th convolution core attributes, the activation primitive and the pond layer；
The activation primitive are as follows: f (x)=max (0, x)；
The standardized function are as follows:
Wherein,Pass through the activation primitive in image at i-th of convolution kernel (x, y) coordinate to sweep in thyroid gland technetium
Output,For thyroid gland technetium sweep in image at i-th of convolution kernel (x, y) coordinate after standardization as a result, k, n, α, β
For hyper parameter.
Further, described device, further includes:
Second obtains module, for obtaining the initial pictures of one or more first size；
Interception module obtains every initial pictures for carrying out sliding interception to every initial pictures according to the second size
Multiple corresponding interception images, wherein the second size is less than first size；
Mirror image operation module obtains corresponding mirror image, and all for carrying out mirror image operation to every interception image
Interception image and mirror image constitute multiple described thyroid gland technetiums and sweep image.
The third aspect, the embodiment of the present invention provide a kind of electronic equipment, comprising: processor, memory and bus, wherein
The processor and the memory complete mutual communication by the bus；
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to refer to
Enable the method and step for being able to carry out first aspect.
Fourth aspect, the embodiment of the present invention provide a kind of non-transient computer readable storage medium, comprising:
The non-transient computer readable storage medium stores computer instruction, and the computer instruction makes the computer
Execute the method and step of first aspect.
The embodiment of the present invention can be realized by constructing identification model and sweep image to technetium to be identified and identify, to assist
Manual identified improves the efficiency and accuracy rate of diagosis.
Other features and advantages of the present invention will be illustrated in subsequent specification, also, partly be become from specification
It is clear that by implementing understanding of the embodiment of the present invention.The objectives and other advantages of the invention can be by written theory
Specifically noted structure is achieved and obtained in bright book, claims and attached drawing.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the identification model construction method process signal that a kind of thyroid gland technetium provided in an embodiment of the present invention sweeps image
Figure；
Fig. 2 is AlexNet general frame schematic diagram provided in an embodiment of the present invention；
Fig. 3 is the identification device structural schematic diagram that a kind of thyroid gland technetium provided in an embodiment of the present invention sweeps image；
Fig. 4 is the structural block diagram of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Usually exist
The component of the embodiment of the present invention described and illustrated in attached drawing can be arranged and be designed with a variety of different configurations herein.Cause
This, is not intended to limit claimed invention to the detailed description of the embodiment of the present invention provided in the accompanying drawings below
Range, but it is merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.Meanwhile of the invention
In description, term " first ", " second " etc. are only used for distinguishing description, are not understood to indicate or imply relative importance.
Relative to Nuclear Medicine Dept's others SPECT check image, the characteristics of thyroid gland technetium sweeps image are as follows: first is that due to right
Thyroid gland technetium is swept when image is identified without complicated data type.Second is that because thyroid gland technetium sweeps the feature extraction phase of image
To simple, the symmetry of Thyroid Imaging shows in image shadow, brightness and thyroid gland double leaf is relied primarily on, figure can be carried out
As processing and identification.
Thyroid position, form, size, the parameters such as whether distribution uniform of imaging agent intake degree and imaging agent,
It is all that thyroid gland technetium sweeps the emphasis for needing to identify when image features extract.
Deep learning is further developed by artificial neural network, and motivation is that foundation, simulation human brain are analyzed
The neural network of study, it imitates the mechanism of human brain to explain data, is learnt by constructing the neural network model of more hidden layers
More useful feature, to finally promote the accuracy of classification and prediction.Accordingly, with respect to individually by the biography of the artificial diagosis of doctor
System mode can significantly be promoted the efficiency of diagosis, be alleviated the diagosis of doctor using deep learning network assistance doctor's diagosis method
Pressure, the working efficiency for promoting doctor, the omission for reducing thyroid gland minimal disease, the accuracy rate for improving diagnosis.
Fig. 1 is the recognition methods flow diagram that a kind of thyroid gland technetium provided in an embodiment of the present invention sweeps image, such as Fig. 1 institute
Show, comprising:
Step 101: obtaining multiple instruction thyroid gland technetiums and sweep image, and image is swept to every technetium and is labeled, obtain corresponding
Type label；
Step 102: the thyroid gland technetium after a mark being swept in image input depth convolutional neural networks AlexNet, institute
AlexNet output prediction result is stated, the type label of image is swept according to the thyroid gland technetium after the prediction result and the mark
Penalty values are calculated, and the parameter in the AlexNet is optimized using the penalty values；
Step 103: the thyroid gland technetium after reacquiring a mark sweeps the AlexNet that image is input to after optimization, and right
AlexNet after optimization is trained again, until reaching the condition of deconditioning, obtains identification model.
In the specific implementation process, before sweeping image to technetium to be identified and identifying, need to pre-establish identification mould
Type, and identification model is trained.Image is swept firstly the need of multiple thyroid gland technetiums are obtained, and manually multiple thyroid gland technetiums are swept
Image carries out type mark, determines that its thyroid gland technetium sweeps the location information in image, shape information developer distributed intelligence
Deng, many of thyroid gland technetium sweep include in image normal type, hyperthyroidism type, first subtract type, thyroid nodule type and from
The quantity of main property hyperactive adenoma type is suitable, these thyroid gland technetiums sweep image as training sample.
When constructing model, depth convolutional neural networks AlexNet can use, and be in advance AlexNet setting loss
Function.Therefore, a training sample can be input in AlexNet, AlexNet can be carried out for the training sample of input
Prediction obtains prediction result, then substitutes into annotation results and prediction result in loss function, calculates and obtains penalty values, and benefit
The parameter inside AlexNet is optimized with penalty values, the AlexNet after being optimized.
Then it reselects in the AlexNet after a training sample is input to optimization, and as aforesaid operations, again
AlexNet after optimization is trained, until reaching the condition of deconditioning.Wherein, the condition of deconditioning can be with
For the amplitude of variation of loss function is less than preset value or frequency of training reaches preset threshold.It is obtained after the completion of training final
Identification model.It should be noted that other types of neural network model can also be utilized, the embodiment of the present invention does not make this
It is specific to limit.It in training, can be trained based on TensorFlow or Caffe, TensorFlow is by complicated nerve
The expression conversion of the program of network query function and data structure is expressed to the program with legibility and carries out the information of modeling and reasoning
The system of processing is widely used for speech recognition or image for the second generation artificial intelligence learning system of Google's research and development
Multinomial machine learning and the deep learning fields such as identification.
The embodiment of the present invention can be realized by constructing identification model and sweep image to technetium to be identified and identify, to assist
Manual identified improves the efficiency and accuracy rate of diagosis.
On the basis of the above embodiments, the depth convolutional neural networks AlexNet includes five convolutional layers and three
Full articulamentum；Wherein,
First convolutional layer includes the first convolution core attributes, activation primitive, pond layer and standardization；
Second convolutional layer includes the second convolution core attributes, the activation primitive, the pond layer and the standardization；
Third convolutional layer and Volume Four lamination include third convolution core attributes and the activation primitive；
5th convolutional layer includes the 5th convolution core attributes, the activation primitive and the pond layer；
The activation primitive are as follows: f (x)=max (0, x)；
The standardized function are as follows:
Wherein,Pass through the activation primitive in image at i-th of convolution kernel (x, y) coordinate to sweep in thyroid gland technetium
Output,For thyroid gland technetium sweep in image at i-th of convolution kernel (x, y) coordinate after standardization as a result, k, n, α, β
For hyper parameter, the meaning of standardized function is exactly that a value and n value before and after it standardize, wherein k=2, n=5, α
=10-4, β=0.75.
Fig. 2 is AlexNet general frame schematic diagram provided in an embodiment of the present invention, as shown in Figure 2:
The structure of convolutional neural networks is not each layer of simple combination, it is by " module " organic composition one by one
, in inside modules, each layer of arrangement is worthy of careful study.Such as the structure chart of AlexNet, it is made of eight modules
's.
It since the bottom of Fig. 2, counts from bottom to top, first layer to layer 5 is the first module, and layer 6 is to the 9th layer
For the second module, one computing module of the first module and the second module composition is exactly one layer of convolution from the point of view of macroscopic perspective,
One layer down-sampled to recycle in this way, and centre is appropriately interposed some functions to control the range of numerical value, so that subsequent circulation is counted
It calculates.
Tenth layer to the 13rd layer is third module and the 4th module and two convolution process, and difference is to have lacked drop to adopt
Sample, reason is just related with the size of input, and the data volume of feature is smaller, so not down-sampled.
14th layer to the 16th layer is all to do convolution as the first module and the second module for the 5th module
Operation.
17th layer to the 19th layer be the 6th module, the 20th layer to 24 layers be the 7th module and the 8th module,
Wherein, the 6th module and the 7th module are full articulamentum, and the 8th module is output as a result, making classification in conjunction with softmax, are had several
Class just exports several nodes, and what each node saved is the probability value for belonging to the category.
It should be noted that AlexNet has used ReLU to solve as the activation primitive of each layer of convolutional neural networks
Gradient disperse problem of the sigmoid function when network is deeper.In addition, using maximum pond in convolutional neural networks, so that
The feature of extraction has translation invariance, avoids the blurring effect in mean value pond.And using the size for being less than Chi Huahe
Step-length so that have overlapping and covering between the output of pond layer, lifting feature it is rich.Have again to local neuron
Activity creation competition mechanism, adds LRN layers, to expand characteristic effect, enhances the generalization ability of model.
And Dropout technology is used in training, ignores a part of neuron so that 0.5 drop rate is temporary at random, from
And solve the overfitting problem of model.
It should be noted that obtain multiple verifying technetiums in training identification model and sweep image, verifying technetium is swept image and is used to pair
Trained identification model is verified, and is optimized according to verification result to the parameter in identification model.
The embodiment of the present invention is by carrying out model training, the identification mould of acquisition using depth convolutional neural networks AlexNet
Type can be improved the recognition accuracy that image is swept to technetium to be identified.
On the basis of the above embodiments, before obtaining multiple training images, the method, further includes:
Obtain the initial pictures of one or more first size；
Sliding interception is carried out to every initial pictures according to the second size, obtains multiple the corresponding interceptions of every initial pictures
Image, wherein the second size is less than first size；
Mirror image operation is carried out to every interception image, obtains corresponding mirror image, and all interception images and mirror image
Multiple thyroid gland technetiums described in image construction sweep image.
In the specific implementation process, due to the limited amount in the training sample being collected into, in order to further mention
The accuracy of high identification model needs to expand acquired training sample, and the specific method is as follows for amplification:
Obtain the initial pictures of a first size, wherein first size can be 64*64, press to this initial pictures
Sliding interception is carried out according to the second size, multiple screenshot images after obtaining sliding interception, wherein the second size can be 56*56.
Mirror image operation is carried out to each interception image, corresponding mirror image is obtained, as a result, by 1 original graph
After carrying out incremental processing, available thyroid gland technetium sweeps the quantity of image as 2* (64-56) ^2=128.
The embodiment of the present invention can mitigate over-fitting significantly, promote extensive energy by carrying out incremental processing to original image
Power.
Fig. 3 is the identification model construction device structural representation that a kind of thyroid gland technetium provided in an embodiment of the present invention sweeps image
Figure, as shown in figure 3, the device includes: the first acquisition module 301, the first training module 302 and the second training module 303,
In:
First acquisition module 301 sweeps image for obtaining multiple thyroid gland technetiums, and sweeps image to every technetium and be labeled, and obtains
Obtain corresponding type label；First training module 302 is used to sweeping the thyroid gland technetium after a mark into image input depth convolution
In neural network AlexNet, the AlexNet exports prediction result, according to the first shape after the prediction result and the mark
The type label that gland technetium sweeps image calculates penalty values, and is optimized using the penalty values to the parameter in the AlexNet；
Second training module 303 sweeps the AlexNet after image is input to optimization for the thyroid gland technetium after reacquiring a mark, and
AlexNet after optimization is trained again, until reaching the condition of deconditioning, obtains identification model.
On the basis of the above embodiments, the depth convolutional neural networks AlexNet includes five convolutional layers and three
Full articulamentum；Wherein,
First convolutional layer includes the first convolution core attributes, activation primitive, pond layer and standardization；
Second convolutional layer includes the second convolution core attributes, the activation primitive, the pond layer and the standardization；
Third convolutional layer and Volume Four lamination include third convolution core attributes and the activation primitive；
5th convolutional layer includes the 5th convolution core attributes, the activation primitive and the pond layer；
The activation primitive are as follows: f (x)=max (0, x)；
The standardized function are as follows:
Wherein,Pass through the activation primitive in image at i-th of convolution kernel (x, y) coordinate to sweep in thyroid gland technetium
Output, k, n, α, β are hyper parameter.
On the basis of the above embodiments, described device, further includes:
Incremental modular, for obtaining the initial pictures of one or more first size；It is initial to every according to the second size
Image carries out sliding interception, obtains multiple corresponding interception images of every initial pictures, wherein the second size is less than the first ruler
It is very little；Mirror image operation is carried out to every interception image, obtains corresponding mirror image, and all interception images and mirror image structure
Image is swept at multiple described thyroid gland technetiums.
On the basis of the above embodiments, described device, further includes:
Preprocessing module carries out pretreatment operation for sweeping image to every thyroid gland technetium, wherein the pretreatment operation
It include: image format conversion, data cleansing and pixel value conversion.
On the basis of the above embodiments, described device, further includes:
Authentication module sweeps image for obtaining multiple verifying technetiums, and sweeps image to the identification mould according to the verifying technetium
Type is verified.
It is apparent to those skilled in the art that for convenience and simplicity of description, the device of foregoing description
Specific work process, no longer can excessively be repeated herein with reference to the corresponding process in preceding method.
Image is swept to technetium to be identified know in conclusion the embodiment of the present invention can be realized by constructing identification model
Not, so that indirect labor identifies, the efficiency and accuracy rate of diagosis are improved.
The present embodiment discloses a kind of computer program product, and the computer program product includes being stored in non-transient calculating
Computer program on machine readable storage medium storing program for executing, the computer program include program instruction, when described program instruction is calculated
When machine executes, computer is able to carry out method provided by above-mentioned each method embodiment, for example, obtains multiple thyroid gland technetiums
Image is swept, and image is swept to every technetium and is labeled, obtains corresponding type label；Thyroid gland technetium after one mark is swept into figure
As in input depth convolutional neural networks AlexNet, the AlexNet exports prediction result, according to the prediction result and institute
The type label that thyroid gland technetium after stating mark sweeps image calculates penalty values, and using the penalty values in the AlexNet
Parameter optimize；Thyroid gland technetium after reacquiring a mark sweeps the AlexNet that image is input to after optimization, and to excellent
AlexNet after change is trained again, until reaching the condition of deconditioning, obtains identification model.The present embodiment mentions
For a kind of non-transient computer readable storage medium, the non-transient computer readable storage medium stores computer instruction, institute
Stating computer instruction makes the computer execute method provided by above-mentioned each method embodiment, for example, obtains multiple first
Shape gland technetium sweeps image, and sweeps image to every technetium and be labeled, and obtains corresponding type label；By the thyroid gland after a mark
Technetium is swept in image input depth convolutional neural networks AlexNet, and the AlexNet exports prediction result, is tied according to the prediction
The type label that thyroid gland technetium after fruit and the mark sweeps image calculates penalty values, and using the penalty values to described
Parameter in AlexNet optimizes；Reacquire one mark after thyroid gland technetium sweep image be input to optimization after
AlexNet, and the AlexNet after optimization is trained again, until reaching the condition of deconditioning, obtain identification mould
Type.
Fig. 4 is the structural block diagram of electronic equipment provided in an embodiment of the present invention.Electronic equipment may include identification device
401, memory 402, storage control 403, processor 404, Peripheral Interface 405, input-output unit 406, audio unit
407, display unit 408.
The memory 402, storage control 403, processor 404, Peripheral Interface 405, input-output unit 406, sound
Frequency unit 407, each element of display unit 408 are directly or indirectly electrically connected between each other, to realize the transmission or friendship of data
Mutually.It is electrically connected for example, these elements can be realized between each other by one or more communication bus or signal wire.The identification
Device 401 includes that at least one can be stored in the memory 402 or solidify in the form of software or firmware (firmware)
Software function module in the operating system (operating system, OS) of identification device 401.The processor 404 is used
In executing the executable module that stores in memory 402, such as the software function module that includes of identification device 401 or computer journey
Sequence.
Wherein, memory 402 may be, but not limited to, random access memory (Random Access Memory,
RAM), read-only memory (Read Only Memory, ROM), programmable read only memory (Programmable Read-Only
Memory, PROM), erasable read-only memory (Erasable Programmable Read-Only Memory, EPROM),
Electricallyerasable ROM (EEROM) (Electric Erasable Programmable Read-Only Memory, EEPROM) etc..
Wherein, memory 402 is for storing program, and the processor 404 executes described program after receiving and executing instruction, aforementioned
Method performed by the server that the stream process that any embodiment of the embodiment of the present invention discloses defines can be applied to processor 404
In, or realized by processor 404.
Processor 404 can be a kind of IC chip, the processing capacity with signal.Above-mentioned processor 404 can
To be that general processor is aided with acceleration equipment, including central processing unit (Central Processing Unit, abbreviation CPU) knot
It closes graphics processing unit (Graphic Processing Unit, abbreviation GPU)；It can also be specific integrated circuit (ASIC), show
At programmable gate array (FPGA) or other programmable logic device, discrete hardware components.It may be implemented or execute this hair
Disclosed each method, step and logic diagram in bright embodiment.General processor can be microprocessor or the processor
404 are also possible to any conventional processor etc..
Various input/output devices are couple processor 404 and memory 402 by the Peripheral Interface 405.Some
In embodiment, Peripheral Interface 405, processor 404 and storage control 403 can be realized in one single chip.Other one
In a little examples, they can be realized by independent chip respectively.
Input-output unit 406 realizes user and the server (or local terminal) for being supplied to user input data
Interaction.The input-output unit 406 may be, but not limited to, mouse and keyboard etc..
Audio unit 407 provides a user audio interface, may include one or more microphones, one or more raises
Sound device and voicefrequency circuit.
Display unit 408 provides an interactive interface (such as user interface) between the electronic equipment and user
Or it is referred to for display image data to user.In the present embodiment, the display unit 408 can be liquid crystal display or touching
Control display.It can be the touching of the capacitance type touch control screen or resistance-type of support single-point and multi-point touch operation if touch control display
Control screen etc..Single-point and multi-point touch operation is supported to refer to that touch control display can sense on the touch control display one or more
The touch control operation generated simultaneously at a position, and the touch control operation that this is sensed transfers to processor 404 to be calculated and handled.
Various input/output devices are couple processor 404 and memory 402 by the Peripheral Interface 405.Some
In embodiment, Peripheral Interface 405, processor 404 and storage control 403 can be realized in one single chip.Other one
In a little examples, they can be realized by independent chip respectively.
Input-output unit 406 is used to be supplied to the interaction that user input data realizes user and processing terminal.It is described defeated
Entering output unit 406 may be, but not limited to, mouse and keyboard etc..
It is appreciated that structure shown in Fig. 4 is only to illustrate, the electronic equipment may also include it is more than shown in Fig. 4 or
The less component of person, or with the configuration different from shown in Fig. 4.Each component shown in Fig. 4 can using hardware, software or
A combination thereof is realized.
In several embodiments provided herein, it should be understood that disclosed device and method can also pass through
Other modes are realized.The apparatus embodiments described above are merely exemplary, for example, flow chart and block diagram in attached drawing
Show the device of multiple embodiments according to the present invention, the architectural framework in the cards of method and computer program product,
Function and operation.In this regard, each box in flowchart or block diagram can represent the one of a module, section or code
Part, a part of the module, section or code, which includes that one or more is for implementing the specified logical function, to be held
Row instruction.It should also be noted that function marked in the box can also be to be different from some implementations as replacement
The sequence marked in attached drawing occurs.For example, two continuous boxes can actually be basically executed in parallel, they are sometimes
It can execute in the opposite order, this depends on the function involved.It is also noted that every in block diagram and or flow chart
The combination of box in a box and block diagram and or flow chart can use the dedicated base for executing defined function or movement
It realizes, or can realize using a combination of dedicated hardware and computer instructions in the system of hardware.
In addition, each functional module in each embodiment of the present invention can integrate one independent portion of formation together
Point, it is also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
It, can be with if the function is realized and when sold or used as an independent product in the form of software function module
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.It should also be noted that similar label and letter exist
Similar terms are indicated in following attached drawing, therefore, once being defined in a certain Xiang Yi attached drawing, are then not required in subsequent attached drawing
It is further defined and explained.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Claims (10)
1. the identification model construction method that a kind of thyroid gland technetium sweeps image characterized by comprising
It obtains multiple thyroid gland technetiums and sweeps image, and image is swept to every technetium and is labeled, obtain corresponding type label；
Thyroid gland technetium after one mark is swept in image input depth convolutional neural networks AlexNet, the AlexNet output
Prediction result calculates penalty values according to the type label that the thyroid gland technetium after the prediction result and the mark sweeps image, and
The parameter in the AlexNet is optimized using the penalty values；
Reacquire one mark after thyroid gland technetium sweep image be input to optimization after AlexNet, and to optimization after
AlexNet is trained again, until reaching the condition of deconditioning, obtains identification model.
2. the method according to claim 1, wherein the depth convolutional neural networks AlexNet includes five
Convolutional layer and three full articulamentums；Wherein,
First convolutional layer includes the first convolution core attributes, activation primitive, pond layer and standardization；
Second convolutional layer includes the second convolution core attributes, the activation primitive, the pond layer and the standardization；
Third convolutional layer and Volume Four lamination include third convolution core attributes and the activation primitive；
5th convolutional layer includes the 5th convolution core attributes, the activation primitive and the pond layer；
The activation primitive are as follows: f (x)=max (0, x)；
The standardized function are as follows:
Wherein,To sweep the output for passing through the activation primitive in image at i-th of convolution kernel (x, y) coordinate in thyroid gland technetium,To sweep in image at i-th of convolution kernel (x, y) coordinate after standardization as a result, k, n, α, β are super ginseng in thyroid gland technetium
Number.
3. the method according to claim 1, wherein the method is also wrapped before obtaining multiple training images
It includes:
Obtain the initial pictures of one or more first size；
Sliding interception is carried out to every initial pictures according to the second size, obtains multiple corresponding interception figures of every initial pictures
Picture, wherein the second size is less than first size；
Mirror image operation is carried out to every interception image, obtains corresponding mirror image, and all interception images and mirror image
It constitutes multiple described thyroid gland technetiums and sweeps image.
4. the method according to claim 1, wherein obtaining after multiple thyroid gland technetiums sweep image, the side
Method, further includes:
To every thyroid gland technetium sweep image carry out pretreatment operation, wherein the pretreatment operation include: image format conversion,
Data cleansing and pixel value conversion.
5. the method according to claim 1, wherein the method, further includes:
It obtains multiple verifying technetiums and sweeps image, and image is swept according to the verifying technetium, the identification model is verified.
6. the identification model construction device that a kind of thyroid gland technetium sweeps image characterized by comprising
First obtains module, sweeps image for obtaining multiple thyroid gland technetiums, and sweep image to every technetium and be labeled, is corresponded to
Type label；
First training module, for the thyroid gland technetium after a mark to be swept image input depth convolutional neural networks AlexNet
In, the AlexNet exports prediction result, and the type of image is swept according to the thyroid gland technetium after the prediction result and the mark
Label calculates penalty values, and is optimized using the penalty values to the parameter in the AlexNet；
Second training module, for reacquire one mark after thyroid gland technetium sweep image be input to optimization after AlexNet,
And the AlexNet after optimization is trained again, until reaching the condition of deconditioning, obtain identification model.
7. device according to claim 6, which is characterized in that the depth convolutional neural networks AlexNet includes five
Convolutional layer and three full articulamentums；Wherein,
First convolutional layer includes the first convolution core attributes, activation primitive, pond layer and standardization；
Second convolutional layer includes the second convolution core attributes, the activation primitive, the pond layer and the standardization；
Third convolutional layer and Volume Four lamination include third convolution core attributes and the activation primitive；
5th convolutional layer includes the 5th convolution core attributes, the activation primitive and the pond layer；
The activation primitive are as follows: f (x)=max (0, x)；
The standardized function are as follows:
Wherein,To sweep the output for passing through the activation primitive in image at i-th of convolution kernel (x, y) coordinate in thyroid gland technetium,To sweep in image at i-th of convolution kernel (x, y) coordinate after standardization as a result, k, n, α, β are super ginseng in thyroid gland technetium
Number.
8. device according to claim 6, which is characterized in that described device, further includes:
Second obtains module, for obtaining the initial pictures of one or more first size；
It is corresponding to obtain every initial pictures for carrying out sliding interception to every initial pictures according to the second size for interception module
Multiple interception images, wherein the second size be less than first size；
Mirror image operation module obtains corresponding mirror image, and all sections for carrying out mirror image operation to every interception image
It takes image and mirror image to constitute multiple described thyroid gland technetiums and sweeps image.
9. a kind of electronic equipment characterized by comprising processor, memory and bus, wherein
The processor and the memory complete mutual communication by the bus；
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to instruct energy
Enough execute the method according to claim 1 to 7.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method according to claim 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811599551.9A CN109685143A (en) | 2018-12-26 | 2018-12-26 | A kind of thyroid gland technetium sweeps the identification model construction method and device of image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811599551.9A CN109685143A (en) | 2018-12-26 | 2018-12-26 | A kind of thyroid gland technetium sweeps the identification model construction method and device of image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109685143A true CN109685143A (en) | 2019-04-26 |
Family
ID=66189460
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811599551.9A Pending CN109685143A (en) | 2018-12-26 | 2018-12-26 | A kind of thyroid gland technetium sweeps the identification model construction method and device of image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685143A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110797101A (en) * | 2019-10-28 | 2020-02-14 | 腾讯医疗健康(深圳)有限公司 | Medical data processing method, device, readable storage medium and computer equipment |
| CN111832639A (en) * | 2020-06-30 | 2020-10-27 | 山西大学 | Drawing emotion prediction method based on transfer learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056595A (en) * | 2015-11-30 | 2016-10-26 | 浙江德尚韵兴图像科技有限公司 | Method for automatically identifying whether thyroid nodule is benign or malignant based on deep convolutional neural network |
| CN106951928A (en) * | 2017-04-05 | 2017-07-14 | 广东工业大学 | The Ultrasound Image Recognition Method and device of a kind of thyroid papillary carcinoma |
| CN107680678A (en) * | 2017-10-18 | 2018-02-09 | 北京航空航天大学 | Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system |
-
2018
- 2018-12-26 CN CN201811599551.9A patent/CN109685143A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056595A (en) * | 2015-11-30 | 2016-10-26 | 浙江德尚韵兴图像科技有限公司 | Method for automatically identifying whether thyroid nodule is benign or malignant based on deep convolutional neural network |
| CN106951928A (en) * | 2017-04-05 | 2017-07-14 | 广东工业大学 | The Ultrasound Image Recognition Method and device of a kind of thyroid papillary carcinoma |
| CN107680678A (en) * | 2017-10-18 | 2018-02-09 | 北京航空航天大学 | Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system |
Non-Patent Citations (3)
| Title |
|---|
| CHENYUPIING666: "深度学习之经典网络架构（二）：AlexNet", 《CSDN博客HTTPS://BLOG.CSDN.NET/CHENYUPING333/ARTICLE/DETAILS/82178335》 * |
| 左东奇等: "基于卷积神经网络提取超声图像甲状腺", 《生物医学工程学杂志》 * |
| 追梦者_ALER: "[深度神经网络]一、AlexNet网络架构详解", 《CSDN博客HTTPS://BLOG.CSDN.NET/QQ_30091945/ARTICLE/DETAILS/83930056?DEPTH_1-UTM_SOURCE=DISTRIBUTE.PC_RELEVANT.NONE-TASK-BLOG-OPENSEARCH-4&UTM_SOURCE=DISTRIBUTE.PC_RELEVANT.NONE-TASK-BLOG-OPENSEARCH-4》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110797101A (en) * | 2019-10-28 | 2020-02-14 | 腾讯医疗健康(深圳)有限公司 | Medical data processing method, device, readable storage medium and computer equipment |
| CN110797101B (en) * | 2019-10-28 | 2023-11-03 | 腾讯医疗健康(深圳)有限公司 | Medical data processing method, medical data processing device, readable storage medium and computer equipment |
| CN111832639A (en) * | 2020-06-30 | 2020-10-27 | 山西大学 | Drawing emotion prediction method based on transfer learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6522161B2 (en) | Medical data analysis method based on deep learning and intelligent analyzer thereof | |
| CN110503654A (en) | A kind of medical image cutting method, system and electronic equipment based on generation confrontation network | |
| CN109886986A (en) | A kind of skin lens image dividing method based on multiple-limb convolutional neural networks | |
| CN110197492A (en) | A kind of cardiac MRI left ventricle dividing method and system | |
| Pawar et al. | Implementation of blockchain technology using extended CNN for lung cancer prediction | |
| CN107492099A (en) | Medical image analysis method, medical image analysis system and storage medium | |
| CN109166130A (en) | A kind of image processing method and image processing apparatus | |
| CN109559300A (en) | Image processing method, electronic equipment and computer readable storage medium | |
| CN107622485A (en) | A kind of medical image data analysis method and system for merging depth tensor neutral net | |
| Kim et al. | Comparison techniques utilized in spatial 3D and 4D data visualizations: A survey and future directions | |
| Ji et al. | Graph model-based salient object detection using objectness and multiple saliency cues | |
| CN107609563A (en) | Picture semantic describes method and device | |
| Gao et al. | A novel face feature descriptor using adaptively weighted extended LBP pyramid | |
| CN109886153A (en) | A kind of real-time face detection method based on depth convolutional neural networks | |
| CN110334719A (en) | The method and system of object image are built in a kind of extraction remote sensing image | |
| CN109325513A (en) | A kind of image classification network training method based on magnanimity list class single image | |
| CN107480720A (en) | Human body attitude model training method and device | |
| CN108596256A (en) | One kind being based on RGB-D object identification grader building methods | |
| CN109685143A (en) | A kind of thyroid gland technetium sweeps the identification model construction method and device of image | |
| Pedro et al. | Using grammars for pattern recognition in images: a systematic review | |
| Shao et al. | Application of U-Net and Optimized Clustering in Medical Image Segmentation: A Review. | |
| Dai et al. | DFN-PSAN: Multi-level deep information feature fusion extraction network for interpretable plant disease classification | |
| Ning et al. | DRAN: Deep recurrent adversarial network for automated pancreassegmentation | |
| Liao et al. | Convolutional herbal prescription building method from multi-scale facial features | |
| Zhang et al. | Deep hyperspherical clustering for skin lesion medical image segmentation |
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: 20190426 |
|
| RJ01 | Rejection of invention patent application after publication |