CN110503151A - A kind of processing method and system of image - Google Patents
A kind of processing method and system of image Download PDFInfo
- Publication number
- CN110503151A CN110503151A CN201910790787.9A CN201910790787A CN110503151A CN 110503151 A CN110503151 A CN 110503151A CN 201910790787 A CN201910790787 A CN 201910790787A CN 110503151 A CN110503151 A CN 110503151A
- Authority
- CN
- China
- Prior art keywords
- image
- labeled data
- data
- processed
- labeled
- 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.)
- Granted
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
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of processing method and system of image, the method obtains labeled data and the corresponding image to be processed of the labeled data first；The labeled data is the data after operator is labeled the image to be processed；Then the labeled data and the corresponding image to be processed of the labeled data are input in deep learning model basin, are obtained and evaluation result corresponding to the labeled data；Finally export the evaluation result；Since evaluation result is obtained based on multiple preparatory image training models being trained according to image and the corresponding labeled data of image in deep learning model, therefore, available accurate evaluation result, allow operator to know whether to analyse whether accurate enough to image, improves user experience.
Description
Technical field
The invention belongs to depth learning technology field more particularly to a kind of processing method and system of image.
Background technique
Image is by, with the image of researching value, analyzing, can obtain according to image accessed by machine
The analysis result of valuable position in image.
But since operator is by the way that in the analytic process to image, meeting is due to originals such as fatigue or monitor resolutions
Cause is likely to result in erroneous judgement, so that analysis result may be not accurate enough.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of processing method and system of image, to prompt operator
Member may analysis to image it is not accurate enough.Concrete scheme is as follows:
The present invention provides a kind of processing methods of image, comprising:
Obtain labeled data and the corresponding image to be processed of the labeled data；The labeled data is operator couple
The image to be processed be labeled after data；
The labeled data and the corresponding image to be processed of the labeled data are input in deep learning model basin,
It obtains and evaluation result corresponding to the labeled data；
Export the evaluation result；
It wherein, include pre- advanced using image and the corresponding labeled data of the image in the deep learning model basin
Multiple image training models of deep learning are gone, whether the evaluation result is for prompting labeled data described in operator
Accurately.
Preferably, the acquisition labeled data and the corresponding image to be processed of the labeled data include:
Receive labeled data and the corresponding image to be processed of the labeled data transmitted by diagosis machine；Alternatively,
Receive the labeled data as transmitted by diagosis machine；
According to the labeled data, obtained and the mark number from the image database being connected with the diagosis machine
According to the image to be processed to match.
Preferably, the training process of multiple image training models in the deep learning model basin specifically includes:
Obtain multiple target images and the corresponding labeled data of the target image；Wherein, the target image is corresponding
Labeled data be by screening after obtain；
Multiple target images, the corresponding labeled data of the target image are divided into training set and verifying collection；
Each of deep learning model basin image training model is trained by training set；
When accuracy rate when a certain image training model is verified using verifying collection is greater than default accuracy rate, by the shadow
It is added in the reliable model library in the deep learning model basin as training pattern.
Preferably, described that labeled data corresponding to the image to be processed and the image to be processed is input to depth
Spend learning model pond in, obtain include: with evaluation result corresponding to the image to be processed
Using labeled data corresponding to the image to be processed and the image to be processed as the reliable model library
In each depth image training model input parameter, obtain each image training model output the image to be processed
The rate of missed diagnosis of corresponding labeled data, false positive rate and accuracy；
Multiple rates of missed diagnosis, the false positive rate and the accuracy be weighted and averaged respectively and is obtained after processing
The false positive rate of target rate of missed diagnosis, target and aiming accuracy are as evaluation result.
Preferably, it is described by multiple rates of missed diagnosis, the false positive rate and the accuracy be weighted and averaged respectively
Before processing further include:
If it is determined that the rate of missed diagnosis of specific image training model output, false positive rate and accuracy do not meet respective ginseng
The corresponding preset condition of number, then remove from the reliable model library by the specific image training model and delete the specific image
Rate of missed diagnosis, false positive rate and the accuracy that training pattern is exported.
Preferably, the multiple target images of the acquisition and the corresponding labeled data of the target image specifically include:
Obtain multiple target images in image database；
Obtain the target labeled data to match with the multiple target image in reliable set for training；Wherein,
The reliable set be stored with from filter out in received labeled data and meet the target labeled data of preset condition.
Preferably, from from filter out in received labeled data and meet the target labeled data of preset condition and include:
Receive labeled data；
Obtain the account in the labeled data；
The first labeled data that expert level condition is not met in the labeled data is determined according to the account；
First labeled data is added in the mark database；
Determine the second labeled data for meeting expert level in the labeled data；
Identification number according to second labeled data judges to mark in the reliable set with the presence or absence of with described second
The identical third mark data of the identification number of data；
If there is the third mark data, then will save with second labeled data to mark database.
Preferably, further includes:
If there is no the third mark data, then judge in set to be verified whether with there are the thirds to identify number
It is greater than the 4th mark data of preset matching degree according to matching degree；
4th mark data if it exists, delete the 4th mark data and by the third mark data save to
In the reliable set；
The third mark data is then saved in set to be verified by the 4th mark data if it does not exist.
Another aspect of the present invention provides a kind of processing system of image characterized by comprising
Diagosis machine, deep learning server, image data server and detecting instrument；
The diagosis machine is connected with the deep learning server and the image data server respectively, is used for
Obtain the image to be processed and mark behaviour has been carried out to the image to be processed by operator that the image server is stored
The labeled data of work；
The deep learning server is for obtaining labeled data and the corresponding image to be processed of the labeled data；By institute
It states labeled data and the corresponding image to be processed of the labeled data is input in deep learning model basin, obtain and the mark
Evaluation result corresponding to data, and the evaluation result is returned into the diagosis machine；Wherein, the deep learning model basin
In include the multiple image training moulds for having carried out deep learning in advance using image and the corresponding labeled data of the image
Type, whether the evaluation result is for prompting labeled data described in operator accurate；
The image data server is used to receive and store the image of the detecting instrument output.
Preferably, the diagosis machine, the deep learning server, the image data server and the detection
Instrument is in same local area network.
Based on the above-mentioned technical proposal, the present invention provides a kind of processing method and system of image, the method obtains first
Obtain labeled data and the corresponding image to be processed of the labeled data；The labeled data is operator to described to be processed
Image be labeled after data；Then the labeled data and the corresponding image to be processed of the labeled data are input to
In deep learning model basin, obtain and evaluation result corresponding to the labeled data；Finally export the evaluation result；Due to
Evaluation result preparatory is trained according to images and the corresponding labeled data of image based on multiple in deep learning model
Image training model show that therefore, available accurate evaluation result knows whether operator
It analyses whether accurate enough to image, improves user experience.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the present invention
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is a kind of flow diagram of the processing method of image disclosed by the embodiments of the present invention；
Fig. 2 is the exemplary diagram of labeled data in a kind of processing method of image disclosed by the embodiments of the present invention；
Fig. 3 is the stream being trained in a kind of processing method of image to multiple image training models in the embodiment of the present invention
Journey schematic diagram；
Fig. 4 is the received mark of institute in the embodiment of the present invention in a kind of processing method of image disclosed by the embodiments of the present invention
The flow diagram for meeting the target labeled data of preset condition is filtered out in data；
Fig. 5 is a kind of structural schematic diagram of the processing system of image provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
In the embodiment of the present invention, the technical field in image processing can be applied, especially to the phase of medical image processing
It closes in scene.
The present invention provides a kind of processing methods of image, are a kind of shadows disclosed by the embodiments of the present invention referring to Fig. 1, Fig. 1
The flow diagram of the processing method of picture.
A kind of image treatment method of the invention can be applied in deep learning server, also can be applied to it is any can
On processor to execute the process.
A kind of processing method of image provided by the invention, comprising:
S101, labeled data and the corresponding image to be processed of the labeled data are obtained；The labeled data is operation
Personnel the image to be processed is labeled after data；
In the embodiment of the present invention, image to be processed can be medical image, be also possible to other shadows with researching value
Picture.The process for obtaining image and labeled data to be processed describes in detail later.
In the embodiment of the present invention, labeled data can provide goal in research position, shape and class for operator
Alias.Referring to fig. 2, Fig. 2 is the exemplary diagram of labeled data in the embodiment of the present invention.Wherein, Fig. 2 is certain Chest Image, operator
Member provides the position where lung and sketches the contours of shape, and is labeled to it: being herein lung areas.
In the embodiment of the present invention, labeled data can be the position shape and corresponding mark.Since research object is logical
It is often the one or several regions in image, therefore, an image can correspond to several marks.Certainly, acquired mark
It can also include the relevant information such as ID etc. of target of being taken in data.
In the embodiment of the present invention, image to be processed corresponding to labeled data and the labeled data can be first obtained.I.e. to
Handling image and labeled data has corresponding relationship.The labeled data is given birth to after operator is labeled image to be processed
At data.
It can also include the image to be processed uniquely corresponding ID in the embodiment of the present invention, in image to be processed, mark
It can also include the ID that the labeled data corresponds to image, the relevant information of operator, such as department: analysis section, duty in data
Business: expert etc..
S102, the labeled data and the corresponding image to be processed of the labeled data are input to deep learning model
Chi Zhong is obtained and evaluation result corresponding to the labeled data；Include in the deep learning model basin using image and
The corresponding labeled data of the image has carried out multiple image training models of deep learning in advance；
It include using image and the corresponding mark of the image in the deep learning model basin in the embodiment of the present invention
Data have carried out multiple image training models of deep learning in advance, and the training data of multiple image training models is reliable
Data, authentic data is, for example, the labeled data that professional is labeled image, more after being input to deep learning model basin
A image training model will do it machine learning, after handling the output result of each image training model, the amount of obtaining
The evaluation result of change.
S103, the output evaluation result；Whether the evaluation result is for prompting labeled data described in operator quasi-
Really.
In the embodiment of the present invention, eventually evaluation result is exported, in order to prompt operator's labeled data whether be
Accurately, whether prompt operator needs that image is analyzed and marked again.
As can be seen that the present invention provides a kind of processing method of image, the method obtain first labeled data and
The corresponding image to be processed of the labeled data；The labeled data is after operator is labeled the image to be processed
Data；Then the labeled data and the corresponding image to be processed of the labeled data are input to deep learning model basin
In, it obtains and evaluation result corresponding to the labeled data；Finally export the evaluation result；Since evaluation result is to be based on
Multiple preparatory image training models being trained according to image and the corresponding labeled data of image in deep learning model
Obtain, therefore, available accurate evaluation result, allow operator know whether be to the analysis of image
It is no accurate enough, improve user experience.
In previous embodiment, be referred to the process for obtaining labeled data and the image to be processed corresponding to it, below it is right
This process describes in detail.
In the embodiment of the present invention, image to be processed, the acquisition labeled data and the mark can be got automatically
The corresponding image to be processed of data includes:
Receive labeled data and the corresponding image to be processed of the labeled data transmitted by diagosis machine；Alternatively,
Receive the labeled data as transmitted by diagosis machine；
According to the labeled data, obtained and the mark number from the image database being connected with the diagosis machine
According to the image to be processed to match.
In the embodiment of the present invention, the image to be processed in image database is that treated image, raw video is for example
The image of DICOM format will include the data unrelated with deep learning, and therefore, the influence to be processed in image database is
It is filtered invalid data and carries out compressed image, it can be into reservation ID and picture material.Therefore, it is passed in data
In defeated, bandwidth accounting can reduce 50% compared to DICOM.
In the embodiment of the present invention, diagosis machine can be read from image database according to the operational order of operator to be needed
Image to be processed to be processed, carries out secondary treatment, such as three-dimensionalreconstruction or the processing such as repaints.
It is understood that if from image database obtain image to be processed when, if image database has rejected this
Request, then the labeled data is saved in mark database, when idle between when again to image database issue obtain ask
It asks.
It should be noted that when obtaining image to be processed from image database, can according in labeled data as ID comes
Find the image to be processed to match.
It is understood that image database and mark database can be same storage device, it is also possible to different
Storage device.The storage device and diagosis machine and the processing unit for executing processing unit of the present invention are in same local area network
In.
It should be noted that can also include the related datas storages such as reliable set, set to be verified in storage device
Database.
Wherein, mark database is for storing through labeled data transmitted by diagosis machine.
In the embodiment of the present invention, diagosis machine and the processing unit for executing the application processing method may be at same local
It is off the net, in this way, just without copying data back and forth between diagosis machine.
Since the result of single deep learning model is not accurate enough, and the difference between model is likely to be breached 1% or more, because
This, in the embodiment of the present invention, is handled image using multiple image training models, relative to single deep learning mould
Type, the comprehensive evaluation result for providing quantization of the reasoning results based on a variety of models are more accurate and objective.
The training process of multiple image training models is introduced below.
It is that multiple image training models are carried out in a kind of processing method of image in the embodiment of the present invention referring to Fig. 3, Fig. 3
Trained flow diagram.
The training process of multiple image training models in the deep learning model basin specifically includes:
S301, multiple target images and the corresponding labeled data of the target image are obtained；Wherein, the target image
Corresponding labeled data is obtained after screening；
In the embodiment of the present invention, there are multiple image training models in deep learning model basin, wherein deep learning model basin
In include reliable model library and basic model pond.Multiple untrained image training models are stored in basic model pond.
The image training model that accuracy rate reaches default accuracy rate is stored in reliable model library.In the embodiment of the present invention, evaluation result
It is to be obtained using the image training model in reliable model library, ensure that accuracy.
In the embodiment of the present invention, labeled data is the reliable mark obtained after screening, and the process specifically screened is rear
Face is introduced.The evaluation result accuracy rate that trained Image model can be made to export using the labeled data after screening is more
It is high.
S302, multiple target images, the corresponding labeled data of the target image are divided into training set and verifying
Collection；
S303, each of deep learning model basin image training model is trained by training set；
When S304, the accuracy rate when a certain image training model is verified using verifying collection are greater than default accuracy rate,
The image training model is added in the reliable model library in the deep learning model basin.
In the embodiment of the present invention, the target image of acquisition and corresponding labeled data are divided into two groups, one group is instruction
Practice collection another group for verifying collection, using in training set target image and labeled data to the image training in basic model pond
Model is trained, and is verified and is learnt using verifying collection, when a certain image training model is in verifying, accuracy rate for example with
The registration of labeled data reaches 95%, then it is assumed that the image training model, which can be added in reliable model library, to be realized to shadow
The processing of picture.
In the embodiment of the present invention, evaluation result is specifically as follows rate of missed diagnosis, false positive rate and accuracy, wherein rate of missed diagnosis
Refer to that operator does not mark, but the ratio of target that image training model obtains.False sun rate refers to that operator marks out
Carry out the ratio for the target that still Image model does not have reasoning to obtain.Accuracy refers to that inputted labeled data and Image model export
The identical ratio of labeled data.
It is understood that, in order to guarantee accuracy, before evaluation result is calculated, can also in the embodiment of the present invention
It carries out data and screens out process.
Specifically, if it is determined that the rate of missed diagnosis of specific image training model output, false positive rate and accuracy are not inconsistent
The corresponding preset condition of each autoregressive parameter is closed, then remove the specific image training model from the reliable model library and deletes this
The rate of missed diagnosis of specific image training model output, false positive rate and accuracy.
In the embodiment of the present invention, if it is determined that the rate of missed diagnosis of a certain image training model output, false positive rate and accurate
Degree does not meet the corresponding preset condition of the parameter, then removes the image training model from the reliable model library.
Wherein, preset condition may include the evaluation result and other image training moulds that a certain image training model exports
It is compared in the evaluation result that type is exported, each single item parameter is maximum value or minimum value in all evaluation results, is being obtained
Before evaluation result, screened out.To guarantee the accuracy of evaluation result.
Wherein, removal, which can be, directly deletes the image training model, is also possible to be placed back into basic model pond.
In the embodiment of the present invention, extra Image model can be filtered out, guarantees the fairness and accuracy of evaluation result.
It is described by labeled data corresponding to the image to be processed and the image to be processed in the embodiment of the present invention
Be input in deep learning model basin, obtain include: with evaluation result corresponding to the image to be processed
Using labeled data corresponding to the image to be processed and the image to be processed as the reliable model library
In each depth image training model input parameter, obtain each image training model output the image to be processed
The rate of missed diagnosis of corresponding labeled data, false positive rate and accuracy；
Multiple rates of missed diagnosis, the false positive rate and the accuracy be weighted and averaged respectively and is obtained after processing
The false positive rate of target rate of missed diagnosis, target and aiming accuracy are as evaluation result.
In the embodiment of the present invention, due to can export respective quantized result with each image training model, because
This, evaluation result, which can be, carries out the result after operating of averaging to each quantized result.In this way, can guarantee to greatest extent
The accuracy of evaluation result.
The foregoing description multiple target images of the acquisition and the corresponding labeled data of the target image, below to this
Process describes in detail.
The multiple target images of the acquisition and the corresponding labeled data of the target image specifically include:
Obtain multiple target images in image database；
Obtain the target labeled data to match with the multiple target image in reliable set for training；Wherein,
The reliable set be stored with from filter out in received labeled data and meet the target labeled data of preset condition；
Referring to fig. 4, Fig. 4 is connect in the embodiment of the present invention in a kind of processing method of image disclosed by the embodiments of the present invention
The flow diagram for meeting the target labeled data of preset condition is filtered out in the labeled data of receipts.
From filter out in received labeled data and meet the target labeled data of preset condition and include:
S401, labeled data is received；
Obtain the account in the labeled data；
S403, the first mark number that expert level condition is not met in the labeled data is determined according to the account
According to；
First labeled data is added in the mark database；
Determine the second labeled data for meeting expert level in the labeled data；
S405, according to second labeled data identification number judge in the reliable set whether there is and described second
The identical third mark data of the identification number of labeled data；
If there is the third mark data, then will save with second labeled data to mark database.
Further include:
S407, if there is no the third mark data, then judge in set to be verified whether with there are the thirds
Identification data matches degree is greater than the 4th mark data of preset matching degree；
4th mark data if it exists, delete the 4th mark data and by the third mark data save to
In the reliable set；
The third mark data is then saved in set to be verified by the 4th mark data if it does not exist.
Described in storage device in previous embodiment may include having mark database, reliable set and collection to be verified
The datum number storages such as conjunction are according to library, wherein mark database stores the labeled data that all diagosis machines transmit, reliable set storage
It can be used for trained labeled data, the unknown labeled data of set memory reliability to be verified.
In actual use, received new labeled data not only has the information of research object in image, also there is mark people's
Account information, the contents such as name, reference numbers, rank, department or laboratory for example including doctor or researcher.Root first
The labeled data of non-expert's rank is screened out according to account information.Secondly it has been searched whether from reliable mark set according to ID identical
Labeled data, if there is being then sent to mark database.If comparing set to be verified without if, see there are other whether before
Expert has done similar mark, if it is not, set to be verified be added this labeled data, if there is set of metadata of similar data, then plus
Enter reliable connection, and all similar marks are deleted from set to be verified.
It is understood that whether similar can be of two or more labeled data judges whether reference name similar and picture
Plain registration is more than certain threshold value such as 90%.
In actual use, image to be processed has been carried out labeling operation on diagosis machine by operator, produces mark
After data, using the processing method of the embodiment of the present invention, the labeled data can be automatically obtained, is then executed subsequent processed
Journey obtains the evaluation result to the labeled data, therefore, for operator, the extra time-consuming operation such as copies without carrying out,
Without complex operations such as uploading pictures download models, so that it may directly obtain return evaluation result.
Another aspect of the present invention provides a kind of processing system of image.
It is a kind of structural schematic diagram of the processing system of image provided in an embodiment of the present invention referring to Fig. 5, Fig. 5.
A kind of processing system of image provided by the invention characterized by comprising
Diagosis machine 1, deep learning server 2, image data server 3 and detecting instrument 4；
The diagosis machine is connected with the deep learning server and the image data server respectively, is used for
Obtain the image to be processed and mark behaviour has been carried out to the image to be processed by operator that the image server is stored
The labeled data of work；
The deep learning server is for obtaining labeled data and the corresponding image to be processed of the labeled data；By institute
It states labeled data and the corresponding image to be processed of the labeled data is input in deep learning model basin, obtain and the mark
Evaluation result corresponding to data, and the evaluation result is returned into the diagosis machine；Wherein, in the deep learning model
Multiple image training models including having carried out deep learning in advance using image and the corresponding labeled data of the image,
Whether the evaluation result is for prompting labeled data described in operator accurate；
The image data server is used to receive and store the image of the detecting instrument output.
Preferably, the diagosis machine, the deep learning server, the image data server and the detection
Instrument is in same local area network.
In the embodiment of the present invention, image to be processed corresponding to labeled data and the labeled data can be first obtained.I.e. to
Handling image and labeled data has corresponding relationship.The labeled data is given birth to after operator is labeled image to be processed
At data.
It include using image and the corresponding mark of the image in the deep learning model basin in the embodiment of the present invention
Data have carried out multiple image training models of deep learning in advance, and the training data of multiple image training models is reliable
Data, authentic data is, for example, the labeled data that professional is labeled image, more after being input to deep learning model basin
A image training model will do it machine learning, after handling the output result of each image training model, the amount of obtaining
The evaluation result of change.
In the embodiment of the present invention, eventually evaluation result is exported, in order to prompt operator's labeled data whether be
Accurately, whether prompt operator needs that image is analyzed and marked again.
In the embodiment of the present invention, detecting instrument can acquire image and generate the image of reference format, such as DICOM format
Image.It imported into the image database in image data server.
It is understood that scanning each time, unique sequence number ID can be all generated, certainly, the image of reference format is also
It may include having the data such as information and the image information of the people that is taken.
Diagosis machine can read image to be processed to be treated from image database, carry out secondary treatment, such as
Three-dimensionalreconstruction repaints.
Deep learning server can receive the mark marked by operator from image data database or diagosis machine
Data, to export corresponding evaluation result.Also it is used to realize the training to multiple Image models.
It is understood that the image data server disclosed in the embodiment of the present invention can also realize a kind of image
Each step of processing method, detailed process is referring to previous embodiment, herein without repeating.
In the embodiment of the present invention, the diagosis machine, the deep learning server, the image data server and
The detecting instrument is in same local area network, and deep learning server can receive image data and the image to be processed
Labeled data dynamically trains a variety of deep learning models i.e. image training model, according to a variety of trained image training moulds
Type provides the evaluation result of quantization, allows researcher to carry out valuable reference, avoids being likely to occur when analysis
Such as the problems such as inaccuracy.
Through the above technical solution as can be seen that due to evaluation result be based in deep learning model it is multiple in advance according to
It is obtained according to the image training model that image and the corresponding labeled data of image are trained, it is therefore, available compared with subject to
True evaluation result allows operator to know whether to analyse whether accurate enough to image, improves user experience.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For system disclosed in embodiment
For, since the function realization of its each device corresponds to the methods disclosed in the examples, so be described relatively simple, it is related
Place is referring to method part illustration.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments in the case where not departing from core of the invention thought or scope.Therefore, originally
Invention is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein
Consistent widest scope.
Claims (10)
1. a kind of processing method of image characterized by comprising
Obtain labeled data and the corresponding image to be processed of the labeled data；The labeled data is operator to described
Image to be processed be labeled after data；
The labeled data and the corresponding image to be processed of the labeled data are input in deep learning model basin, obtained
With evaluation result corresponding to the labeled data；
Export the evaluation result；
It wherein, include being carried out in advance using image and the corresponding labeled data of the image in the deep learning model basin
Multiple image training models of deep learning, whether the evaluation result is for prompting labeled data described in operator quasi-
Really.
2. processing method according to claim 1, which is characterized in that the acquisition labeled data and the labeled data
Corresponding image to be processed includes:
Receive labeled data and the corresponding image to be processed of the labeled data transmitted by diagosis machine；Alternatively,
Receive the labeled data as transmitted by diagosis machine；
According to the labeled data, obtained and the labeled data phase from the image database being connected with the diagosis machine
Matched image to be processed.
3. processing method according to claim 1, which is characterized in that multiple images instruction in the deep learning model basin
The training process for practicing model specifically includes:
Obtain multiple target images and the corresponding labeled data of the target image；Wherein, the corresponding mark of the target image
Note data are obtained after screening；
Multiple target images, the corresponding labeled data of the target image are divided into training set and verifying collection；
Each of deep learning model basin image training model is trained by training set；
When accuracy rate when a certain image training model is verified using verifying collection is greater than default accuracy rate, which is instructed
Practice model to be added in the reliable model library in the deep learning model basin.
4. processing method according to claim 3, which is characterized in that described by the image to be processed and described wait locate
Labeled data corresponding to reason image is input in deep learning model basin, is obtained and evaluation corresponding to the image to be processed
Result includes:
Using labeled data corresponding to the image to be processed and the image to be processed as every in the reliable model library
The input parameter of one depth image training model, the image institute to be processed for obtaining the output of each image training model are right
The rate of missed diagnosis for the labeled data answered, false positive rate and accuracy；
Multiple rates of missed diagnosis, the false positive rate and the accuracy are weighted and averaged the target obtained after processing respectively
The false positive rate of rate of missed diagnosis, target and aiming accuracy are as evaluation result.
5. processing method according to claim 4, which is characterized in that it is described by multiple rates of missed diagnosis, the false positive rate
And before the accuracy is weighted and averaged processing respectively further include:
If it is determined that the rate of missed diagnosis of specific image training model output, false positive rate and accuracy do not meet each autoregressive parameter pair
The specific image training model is then removed from the reliable model library and deletes the specific image training by the preset condition answered
The rate of missed diagnosis of model output, false positive rate and accuracy.
6. according to processing method described in claim 3,4 or 5, which is characterized in that described to obtain multiple target images and institute
The corresponding labeled data of target image is stated to specifically include:
Obtain multiple target images in image database；
Obtain the target labeled data to match with the multiple target image in reliable set for training；Wherein, described
Reliable set be stored with from filter out in received labeled data and meet the target labeled data of preset condition.
7. processing method according to claim 6, which is characterized in that filter out and meet from from the received labeled data of institute
The target labeled data of preset condition includes:
Receive labeled data；
Obtain the account in the labeled data；
The first labeled data that expert level condition is not met in the labeled data is determined according to the account；
First labeled data is added in the mark database；
Determine the second labeled data for meeting expert level in the labeled data；
Identification number according to second labeled data judges to whether there is and second labeled data in the reliable set
The identical third mark data of identification number；
If there is the third mark data, then will save with second labeled data to mark database.
8. processing method according to claim 7, which is characterized in that further include:
If there is no the third mark data, then judge in set to be verified whether with there are the third mark datas
It is greater than the 4th mark data of preset matching degree with degree；
4th mark data if it exists deletes the 4th mark data and saves the third mark data to described
In reliable set；
The third mark data is then saved in set to be verified by the 4th mark data if it does not exist.
9. a kind of processing system of image characterized by comprising
Diagosis machine, deep learning server, image data server and detecting instrument；
The diagosis machine is connected with the deep learning server and the image data server respectively, for obtaining
The image to be processed and labeling operation has been carried out to the image to be processed by operator that the image server is stored
Labeled data；
The deep learning server is for obtaining labeled data and the corresponding image to be processed of the labeled data；By the mark
Note data and the corresponding image to be processed of the labeled data are input in deep learning model basin, are obtained and the labeled data
Corresponding evaluation result, and the evaluation result is returned into the diagosis machine；Wherein, it is wrapped in the deep learning model basin
Include the multiple image training models for having carried out deep learning in advance using image and the corresponding labeled data of the image, institute
Evaluation result is stated for prompting labeled data described in operator whether accurate；
The image data server is used to receive and store the image of the detecting instrument output.
10. processing system according to claim 9, which is characterized in that the diagosis machine, the deep learning service
Device, the image data server and the detecting instrument are in same local area network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910790787.9A CN110503151B (en) | 2019-08-26 | 2019-08-26 | Image processing method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910790787.9A CN110503151B (en) | 2019-08-26 | 2019-08-26 | Image processing method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110503151A true CN110503151A (en) | 2019-11-26 |
| CN110503151B CN110503151B (en) | 2020-11-03 |
Family
ID=68589538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910790787.9A Active CN110503151B (en) | 2019-08-26 | 2019-08-26 | Image processing method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110503151B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862105A (en) * | 2019-11-28 | 2021-05-28 | 杭州海康威视数字技术股份有限公司 | Data publishing system, method and device |
| WO2021169101A1 (en) * | 2020-02-28 | 2021-09-02 | 平安国际智慧城市科技股份有限公司 | Method and apparatus for generating medical image recognition model, computer device and medium |
| CN113450381A (en) * | 2021-06-16 | 2021-09-28 | 上海深至信息科技有限公司 | System and method for evaluating accuracy of image segmentation model |
| CN114171166A (en) * | 2021-01-20 | 2022-03-11 | 赛维森(广州)医疗科技服务有限公司 | Management system of model of visual digital pathological artificial intelligence |
| US20220309281A1 (en) * | 2021-03-26 | 2022-09-29 | Here Global B.V. | Method, apparatus, and system for removing outliers from road lane marking data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709917A (en) * | 2017-01-03 | 2017-05-24 | 青岛海信医疗设备股份有限公司 | Neural network model training method, device and system |
| CN107729901A (en) * | 2016-08-10 | 2018-02-23 | 阿里巴巴集团控股有限公司 | Method for building up, device and the image processing method and system of image processing model |
| WO2018142765A1 (en) * | 2017-02-03 | 2018-08-09 | パナソニックＩｐマネジメント株式会社 | Learned model provision method, and learned model provision device |
| CN109003679A (en) * | 2018-06-28 | 2018-12-14 | 众安信息技术服务有限公司 | A kind of cerebrovascular hemorrhage and ischemic prediction technique and device |
-
2019
- 2019-08-26 CN CN201910790787.9A patent/CN110503151B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729901A (en) * | 2016-08-10 | 2018-02-23 | 阿里巴巴集团控股有限公司 | Method for building up, device and the image processing method and system of image processing model |
| CN106709917A (en) * | 2017-01-03 | 2017-05-24 | 青岛海信医疗设备股份有限公司 | Neural network model training method, device and system |
| WO2018142765A1 (en) * | 2017-02-03 | 2018-08-09 | パナソニックＩｐマネジメント株式会社 | Learned model provision method, and learned model provision device |
| CN109003679A (en) * | 2018-06-28 | 2018-12-14 | 众安信息技术服务有限公司 | A kind of cerebrovascular hemorrhage and ischemic prediction technique and device |
Non-Patent Citations (2)
| Title |
|---|
| 公安部第三研究所: "《多摄像机协同关注目标检测跟踪技术》", 30 June 2017, 东南大学出版社 * |
| 孙雪梅: "《数学教学设计》", 30 October 2014, 哈尔滨工业大学出版社 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862105A (en) * | 2019-11-28 | 2021-05-28 | 杭州海康威视数字技术股份有限公司 | Data publishing system, method and device |
| CN112862105B (en) * | 2019-11-28 | 2023-08-08 | 杭州海康威视数字技术股份有限公司 | Data release system, method and device |
| WO2021169101A1 (en) * | 2020-02-28 | 2021-09-02 | 平安国际智慧城市科技股份有限公司 | Method and apparatus for generating medical image recognition model, computer device and medium |
| CN114171166A (en) * | 2021-01-20 | 2022-03-11 | 赛维森(广州)医疗科技服务有限公司 | Management system of model of visual digital pathological artificial intelligence |
| US20220309281A1 (en) * | 2021-03-26 | 2022-09-29 | Here Global B.V. | Method, apparatus, and system for removing outliers from road lane marking data |
| CN113450381A (en) * | 2021-06-16 | 2021-09-28 | 上海深至信息科技有限公司 | System and method for evaluating accuracy of image segmentation model |
| CN113450381B (en) * | 2021-06-16 | 2022-10-18 | 上海深至信息科技有限公司 | System and method for evaluating accuracy of image segmentation model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110503151B (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110503151A (en) | A kind of processing method and system of image | |
| US8155406B2 (en) | Image processing system, particularly for use with diagnostic images | |
| CN105069308B (en) | Medical Devices intelligent monitor system | |
| CN107133682B (en) | Equipment log-based equipment operation time estimation system and method, equipment log-based dynamic reservation system and implementation method thereof | |
| JP2021535484A (en) | System for automatic tumor detection and classification | |
| CN102667786A (en) | Protocol guided imaging procedure | |
| JP2007172604A (en) | Method and apparatus for selecting computer-assisted algorithm based on protocol and/or parameter of acquisition system | |
| CN108009786A (en) | Work attendance statistics method, apparatus, computer equipment and readable storage medium storing program for executing | |
| CN108665963A (en) | A kind of image data analysis method and relevant device | |
| CN109785977A (en) | Automated information input method, system, device and storage medium | |
| US20200129114A1 (en) | Automatic computerized joint segmentation and inflammation quantification in mri | |
| CN109390040A (en) | A kind of intelligent electronic medical records system | |
| CN114746953A (en) | AI system for predictive review of reading time and reading complexity of 2D/3D breast images | |
| CN108492885A (en) | Check that workflow recommends method, apparatus and terminal | |
| CN111226287B (en) | Method, system, program product and medium for analyzing medical imaging data sets | |
| CN111684534A (en) | Apparatus, system and method for optimizing pathology workflow | |
| CN110957014B (en) | Early lung cancer screening information management system | |
| CN101836211A (en) | System and method for combining serial ECG analysis and ECG ordering | |
| CN116631626A (en) | Patient clinical risk assessment method, device, equipment and medium | |
| CN110147791A (en) | Character recognition method, device, equipment and storage medium | |
| JP2019054896A (en) | Medical image processing device, medical image processing method, and processing program for use in the same | |
| TWI774982B (en) | Medical resource integration system, computer device and medical resource integration method | |
| CN114037540A (en) | Medical insurance fund supervision system and monitoring method thereof | |
| CN113724820A (en) | Fuzzy positioning method, system, equipment and storage medium for focus of medical image | |
| JP2022035396A (en) | Image conversion program, method, and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: Room B401, floor 4, building 1, Shangdi Information 12, Haidian District, Beijing 100085 Patentee after: Tuxiang Medical Technology Co., Ltd Address before: Room B401, floor 4, building 1, Shangdi Information 12, Haidian District, Beijing 100085 Patentee before: Beijing Tuoxiang Technology Co.,Ltd. |