CN110070576A - A kind of ultrasound based on deep learning network adopts figure intelligent locating method and system - Google Patents
A kind of ultrasound based on deep learning network adopts figure intelligent locating method and system Download PDFInfo
- Publication number
- CN110070576A CN110070576A CN201910356083.0A CN201910356083A CN110070576A CN 110070576 A CN110070576 A CN 110070576A CN 201910356083 A CN201910356083 A CN 201910356083A CN 110070576 A CN110070576 A CN 110070576A
- Authority
- CN
- China
- Prior art keywords
- image
- dimensional
- deep learning
- learning network
- spatial position
- 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
- 238000013135 deep learning Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000002604 ultrasonography Methods 0.000 title claims abstract description 33
- 239000000523 sample Substances 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 21
- 230000006870 function Effects 0.000 claims description 14
- 230000001575 pathological effect Effects 0.000 claims description 8
- 239000011248 coating agent Substances 0.000 claims description 7
- 238000000576 coating method Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 3
- 238000010191 image analysis Methods 0.000 claims description 3
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 241000208340 Araliaceae Species 0.000 description 4
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 4
- 235000003140 Panax quinquefolius Nutrition 0.000 description 4
- 210000000481 breast Anatomy 0.000 description 4
- 235000008434 ginseng Nutrition 0.000 description 4
- 210000001519 tissue Anatomy 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000009877 rendering Methods 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 210000005075 mammary gland Anatomy 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000010827 pathological analysis Methods 0.000 description 2
- 230000036285 pathological change Effects 0.000 description 2
- 231100000915 pathological change Toxicity 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 210000003734 kidney Anatomy 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 201000004409 schistosomiasis Diseases 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
The invention discloses a kind of ultrasounds based on deep learning network to adopt figure intelligent locating method and system, comprising: scans and obtains two dimensional image and its spatial position coordinate, store interested two dimensional image for establishing reference picture library;The corresponding three-dimensional data of two dimensional image is acquired, and seeks the corresponding cuboid three-dimensional data of three-dimensional data, and establishes the cuboid three-dimensional data space coordinates of specification;Construct deep learning network, random two-dimensional tangent plane picture and its spatial position coordinate are generated using the cuboid three-dimensional data of specification, training deep learning network, with deep learning network satisfaction: when there is the input of two-dimensional scanning image, the automatic spatial position coordinate for exporting the corresponding virtual probe of scan image.The present invention can obtain the spatial position coordinate of two dimensional image section, the position of real-time prompting scanning probe section in human body in real time, and guidance doctor finds interested or marked section, save the time, improve diagnosis efficiency.
Description
Technical field
The present invention relates to ultrasonic imaging technique fields more particularly to a kind of ultrasound based on deep learning network to adopt figure intelligence
Localization method and system.
Background technique
Medical ultrasonic imaging system is suitable for the clinical diagnosis of the organ-tissues such as liver, kidney, mammary gland.How to establish
The mapping relations of ultrasonic 2D slice (or 2D scan image) and 3D volume data.In existing medical ultrasound system, how 2D to be cut
Piece (or 2D scan image) is mapped to corresponding 3D volume data comparative maturity, and there are commonly similar: installing magnetic additional at probe
The geometric position of tracking transducer record 2D slice, using picture frame in probe moving process and frame correlation, use 3D
Volume probe, using the mobile ultrasonic probe of mechanical arm ABUS (Automated Breast Ultrasound System from
Dynamic breast ultrasound system).But in terms of 3D volume data is mapped to 2D image, medical ultrasound system is by 3D volume data at present
It in terms of being mapped to 2D tangent plane picture, depends on during doctor examines ultrasonic 3D rendering closely, the more intuitive 3D figure of selection
A certain section is determined as in and then obtains this section 2D image corresponding in 3D volume data, i.e., the spatial position of section is by curing
Raw (i.e. operator) determines.
To sum up, in the inspection of existing clinical ultrasound, the picture quality of common hand hold transducer and the accuracy of diagnosis are depended on
Doctor's Experiences and Skills are operated, so the best section position that the doctor of different level gets might have deviation.And in ultrasound
When further consultation, doctor can play again figure verifying to pathological characters described in previous case history.Even if described in original case history
Ultrasound image corresponding to pathological characters has been stored in the image library of patient, and doctor still needs to take a significant amount of time, and only may be used
The scanning section that can be found and be previously stored as image, to compare, check and find relevant clinical information.
Summary of the invention
An object of the present invention at least that, for how to overcome the above-mentioned problems of the prior art, provide one kind
Ultrasound based on deep learning network adopts figure intelligent locating method and system, can obtain the space bit of two dimensional image section in real time
Coordinate, the position of real-time prompting scanning probe section in human body are set, guidance doctor finds interested or marked cut
Face.
To achieve the goals above, the technical solution adopted by the present invention includes following aspects.
A kind of ultrasound based on deep learning network adopts figure intelligent locating method, comprising:
Scan and obtain two dimensional image and its spatial position coordinate, store interested two dimensional image for establish with reference to figure
As library;The corresponding three-dimensional data of the two dimensional image is acquired, and seeks the corresponding cuboid said three-dimensional body of the three-dimensional data
Data, and establish the cuboid three-dimensional data space coordinates of specification;
Construct deep learning network, using specification cuboid three-dimensional data generate random two-dimensional tangent plane picture and its
The spatial position coordinate training deep learning network, with deep learning network satisfaction: when there is the input of two-dimensional scanning image
When, the spatial position coordinate of the corresponding virtual probe of the scan image is exported automatically.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method,
It further include interested two dimensional image section being intercepted from three-dimensional data based on image analysis algorithm, and pass through institute
The spatial position coordinate that deep learning network seeks the interested two dimensional image section is stated, by the interested X-Y scheme
As section and its spatial position coordinate label are in the reference picture library;
Or the three-dimensional data is generated to the 3-D image of different angle based on 3-D image method for drafting, from described
Interested two dimensional image section is chosen in 3-D image, and the interested two dimension is sought by the deep learning network
The spatial position coordinate of image slices, by the interested two dimensional image section and its spatial position coordinate label in the ginseng
It examines in image library.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method, further includes:
When user demand enters " training " or " further consultation " mode, the ginseng that stores in the reference picture library based on user's selection
It examines two dimensional image and its is adjusted by the resulting spatial position coordinate pair current probe scanning area of deep learning network, directly
The spatial position coordinate obtained to gained two-dimensional scanning image based on the deep learning network is deposited with described with reference to two dimensional image
The spatial position coordinate difference of storage stops the adjustment of probe positions when being less than preset value, user is prompted to select " further consultation " or " training " mould
Formula.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method, further includes:
When getting user's selection entrance " further consultation " mode, by the area-of-interest and ginseng in the two-dimensional scanning image
It examines area-of-interest marked in two dimensional image and carries out analyzing image texture, the region of interest in the two-dimensional scanning image
When the image texture difference of area-of-interest in the reference picture of domain and label is greater than preset value, then pathological characters are judged
Variation, then enter " system update " mode, system based on reference picture library described in currently available two-dimensional scanning image update, and
Spatial position coordinate data based on currently available two-dimensional scanning image updates the deep learning network.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method, further includes when the scanning
The image texture difference of area-of-interest in image and the area-of-interest in the reference picture of storage is less than or equal to preset value
When, then judge that pathological characters are unchanged and stop operation.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method, the deep learning network
It include: an image input layer, multiple full articulamentums and active coating and a recurrence layer active coating using line rectification
Function, the loss function for returning layer use half mean square error function.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method, using installation magnetic field tracking
Device method, method for registering images install three-dimensional volume probe additional, one of are scanned method using mechanical arm and acquire the two dimension
The corresponding three-dimensional data of image.
Preferably, a kind of ultrasound based on deep learning network is adopted in figure intelligent locating method, using picture frame and frame it
Between interpolation method generate the cuboid three-dimensional data of the three-dimensional data.
A kind of ultrasound based on deep learning network adopts figure intelligent positioning system, including at least one processor, Yi Jiyu
The memory of at least one processor communication connection;The memory, which is stored with, to be executed by least one described processor
Instruction, described instruction is executed by least one described processor, so that at least one described processor is able to carry out above-mentioned side
Method.
In conclusion by adopting the above-described technical solution, the present invention at least has the advantages that
By constructing deep learning network, reference picture library, it can calculate in real time and currently cut according to the image of two dimension slicing
Position of the piece in volume data provides phase while providing current probe section present position in the tissue for doctor in real time
The reference slice (doctor is helped to improve further consultation efficiency, save the plenty of time when needing further consultation) answered, and gradually can guide and mention
Show that the insufficient doctor of experience finds the interested region or the best section (training as doctor with pathological analysis meaning
Tool), it efficiently and accurately completes to check, plays the role of auxiliary diagnosis.
Detailed description of the invention
Fig. 1 is that the ultrasound according to an exemplary embodiment of the present invention based on deep learning network adopts figure intelligent locating method stream
Cheng Tu.
Fig. 2 is intelligent positioning network structure topological diagram according to an exemplary embodiment of the present invention.
Fig. 3 is according to an exemplary embodiment of the present invention based on deep learning network " further consultation " or " training " model process
Figure.
Fig. 4 is simulation slice position guidance schematic diagram according to an exemplary embodiment of the present invention.
Fig. 5 is that prediction according to an exemplary embodiment of the present invention and sectioning image (value of the non-outlier) comparison of actual position are shown
It is intended to figure.
Fig. 6 is prediction according to an exemplary embodiment of the present invention and sectioning image (outlier) the comparison signal of actual position
Figure.
Fig. 7 is that the ultrasound according to an exemplary embodiment of the present invention based on deep learning network adopts figure intelligent positioning system knot
Structure schematic diagram.
Specific embodiment
With reference to the accompanying drawings and embodiments, the present invention will be described in further detail, so that the purpose of the present invention, technology
Scheme and advantage are more clearly understood.It should be appreciated that described herein, specific examples are only used to explain the present invention, and does not have to
It is of the invention in limiting.
Fig. 1 shows the ultrasound according to an exemplary embodiment of the present invention based on deep learning network and adopts figure intelligent positioning side
Method.The method of the embodiment specifically includes that
Scan and obtain two dimensional image and its spatial position coordinate, store interested two dimensional image for establish with reference to figure
As library;The corresponding three-dimensional data of the two dimensional image is acquired, and seeks the corresponding cuboid said three-dimensional body of the three-dimensional data
Data, and establish the cuboid three-dimensional data space coordinates of specification.
Construct deep learning network, using specification cuboid three-dimensional data generate random two-dimensional tangent plane picture and its
The spatial position coordinate training deep learning network, with deep learning network satisfaction: when there is the input of two-dimensional scanning image
When, the spatial position coordinate of the corresponding virtual probe of the scan image is exported automatically.
Specifically, mobile ultrasonic probe, scans human body area-of-interest (i.e. the pathological regions of human body) and obtains corresponding
2D image uses and installs magnetic field tracking device, method for registering images additional, installs 3D volume probe additional or using mechanical arm to generate described one
Serial 2D image and its 3D volume data.Storage scans resulting interested two dimensional image and its corresponds in deep learning network
Spatial position coordinate (the spatial position coordinate acquired based on deep learning network) for establishing reference picture library.
Further, the Method And Principle for installing magnetic field tracking device additional is, a magnetic field tracking sensor is fixed on probe,
When mobile ultrasonic probe obtains 2D slice, this 2D slice is recorded in the geometric position (i.e. geometric coordinate) of 3d space;From a system
Oriented 2D slice is arranged, obtains 3D volume data using image interpolation, at this time the spatial position of 2D slice and magnetic field tracking sensing
The coordinate system of device is related, and unrelated with the image information of 2D slice.Ultrasound image is utilized to visit using method for registering images
In head moving process, the image correlation (correlation) between frame and frame acquires the spatial position of 2D slice to obtain 3D
Volume data, the accuracy of this correlation depend on image in the resolution ratio and image noise of probe side surface direction (elevation)
Than.It installs 3D volume probe additional, i.e., is rotated using mechanical device and record the probe of the conventional ultrasound built in one and scanned in formation 2D
Rotational angle when image recycles image interpolation to obtain 3D volume data to form a 3D scanning space.Although this 3D is scanned
The limited size in space in the angle of the mechanical rotation of probe, but due to do not need magnetic field tracking sensor (hospital's use can be by
To the interference of other electronic instruments).The use of 3D volume probe is the most widely used method of current 3D ultrasound.However, 3D holds
Product its scanning space of probe is limited by probe width built in it and mechanical rotation angle.And be using mechanical arm method,
Using the mobile ultrasonic probe of mechanical arm to obtain 3D volume data.Most widely used at present is automatic breast ultrasound (Automated
Breast Ultrasound System, ABUS) device, its working principle is that a ultrasonic wire array probe up to 15cm is embedding
It can be covered at one on the plate that supramammary size is 15cm x 17cm, mechanical arm then translates this probe with constant speed
17cm and formed an area be 15cm x 17cm, depth be ultrasonic scanning depth 3D volume data.The principle of ABUS can also push away
It is wide to pop one's head in using the mobile specific customization of mechanical arm to obtain the 3D volume data at other positions other than mammary gland.In addition to pole superficial
Layer because by covering caused by plate outside Tissue approximation, such device be at present do not depend on ultrasonic doctor gimmick and it is available most
Accurate 3D volume data method.
After obtaining the 3-D volume data of 2D image, specification is produced using the interpolation between traditional images frame and frame
(canonical) cuboid 3-D volume data.At this point, the coordinate system of 3-D volume data is defined by the cuboid standardized, and originally by
The coordinate that distinct methods generate 3-D volume data after 2-D image scanning is unrelated.Therefore, figure is done using deep learning subsequent
When as with the mapping of probe spatial position, all using the space coordinates of the 3-D volume data uniquely standardized.
And the method also includes: based on interested two dimension is intercepted described in image analysis algorithm from three-dimensional data
Image slices, and by the interested two dimensional image section label in the reference picture library;
Or the three-dimensional data is generated to the 3-D image of different angle based on 3-D image method for drafting, from described
Interested two dimensional image section is chosen in 3-D image, and by the interested two dimensional image section label in the reference
In image library, reference picture library is improved with this.
Specifically, doctor or user can check the 2D image that virtual probe is scanned by the 3D volume data by specification,
And the image of interest with clinical meaning is stored in above-mentioned reference picture library.This review process can also be extended to and transfer to figure
As parser, under the conditions of following the characteristics of image with clinical meaning, 2D image slices are intercepted simultaneously from 3D volume data automatically
It is stored in reference picture library.Or doctor or user also can be via traditional 3D object plotting methods by raw in 3-D volume data
At the 3D rendering of different angle, by (similar to the stereoscopic three-dimensional surface of tissue/organ), determining that sense is emerging in more intuitive 3D rendering
The inside 2D image slices of interest.Virtual probe coordinate (i.e. its spatial position coordinate) corresponding to these 2D image slices exists simultaneously
It has been acquired when generating 2D image, by interested 2D image slices label, has been stored in reference picture library, ginseng is improved with this
Examine image library.
Then, deep learning network (being also named as intelligent positioning network) is constructed, as shown in Fig. 2, the deep learning network
It include: an image input layer, multiple full articulamentums and active coating, a recurrence layer.Wherein, that input layer input is 64*64*
1 gray level image.One layer of active coating is all closely followed after every layer of full articulamentum, active coating uses line rectification function (Rectified
Linear Unit,ReLU).The last layer that layer is network structure is returned, the preceding layer for returning layer needs to increase comprising four minds
Full articulamentum through member, has corresponded to four location parameters.The loss function for returning layer uses half mean square error function.Training is entire
Deep learning network model framework is selected during training network to minimize loss function and update Optimal Parameters
Stochastic gradient descent (stochastic gradient descent with momentum, SGDM) with momentum is as excellent
Change device control convergence, network is made finally to export it with four parameters predicting according to input picture.During network training,
Schistosomiasis control rate is taken biggish learning rate in initial training stage, is finely adjusted later with lesser learning rate, and use is square
Root error (RMSE) indicates the spatial position of neural network forecast and the difference of physical location.Wherein the structure of fully-connected network can also
It is replaced with convolutional neural networks or other structures.It is " intelligent fixed in this deep learning network, that is, this system when network convergence
Position network ".
Specifically, one group of two dimensional image and its 3D volume data (spatial position coordinate) are generated using two dimensional image scanning method,
The space coordinates of 3-D volume data based on specification adjust the 3D volume data of the two dimensional image.To utilize this group of X-Y scheme
3D volume data after picture and its specification is trained deep learning network, optimizes network architecture parameters, so that network meets: when
When having the input of two-dimensional scanning image, spatial position coordinate (the i.e. network of the corresponding virtual probe of the scan image is exported automatically
Convergence) when deconditioning, the intelligent positioning network has been built up completion at this time.
Thus, it is possible to find out, the ultrasonic intelligent localization method provided by the invention based on deep learning network can basis
The image of two dimension slicing, calculating position of the current slice in volume data in real time, (input picture, network can automatically generate its sky
Between position coordinates).Sectioning image is not limited only to test set and training set, can come from any position in volume data.In this hair
In bright further embodiment, we carry out performance survey using deep learning network of 10,000 both two-dimensional image slices to us
Examination.As shown in Table 1, the location parameter precision in the deep learning network query function result can achieve pixel scale, rotation parameter
Mean error less than 1 °, predict every slice probably spend 0.1 millisecond, meet real-time demand, have to clinical application
Very big value and significance.
1. 1 ten thousand prediction results being sliced at random of table
Further, the outlier in table one refers to that discrete error differs farther away predicted value with other data, is specifically defined
For the numerical value for differing by more than three times conversion MAD with median, remaining predicted value is value of the non-outlier.Convert MAD is defined as: c*
median(abs(A-median(A))).Abs () is to seek absolute value, and median () is to seek median, the calculation formula of coefficient c
For c=-1/ (sqrt (2) * erfcinv (3/2)), erfcinv () is the inverse function of complementary error function erfc ().RMSE is
Root-mean-square error, for indicating the difference between true value and predicted value.Formula is as follows:
Further, as shown in figure 3, being chosen when user demand enters " training " or " further consultation " mode based on user
The reference two dimensional image that is stored in reference picture library and its it is corresponding with reference to spatial position coordinate pair current probe scanning area into
Row adjustment, until the spatial position coordinate that gained two-dimensional scanning image is obtained based on the deep learning network refers to two with described
Tie up image storage spatial position coordinate difference be less than preset value when stop probe positions adjustment, prompt user select " further consultation " or
" training " mode.
Specifically, when the intelligent positioning network of patient has been established and needs doctor's further consultation or the shallow doctor of money to need to fight each other figure hand
When method gives training, ultrasound 2-D image is scanned in specific region (pathological regions of patient) by doctor first, obtains width ultrasound
Image IC.By Current Scan image ICInto intelligent positioning network, intelligent positioning network then exports the spatial position popped one's head at present
PC.Then interested 2-D tangent plane picture I is chosen in the reference picture library of patientREFAnd its corresponding probe spatial position
PREF.As the probe spatial position P exported at present by intelligent positioning networkCWith probe corresponding to conceivable reference picture
Spatial position PREFWhen difference is larger, into " the 3-D interactive interface of guidance doctor's mobile probe ", (i.e. guidance doctor pops one's head in
Position adjustment, Fig. 4 show illustrative simulation slice position (probe positions adjustment schematic diagram) guidance schematic diagram of the invention).
Work as PCSufficiently close to PREFWhen, i.e., the spatial position coordinate that is obtained based on the deep learning network of gained two-dimensional scanning image with
When the spatial position coordinate difference stored with reference to two dimensional image is less than preset value (a good minimum of default), stop
The adjustment of probe positions, system prompt user select to enter " reinspection or training " process.
Specifically, when user selects to enter " further consultation " mode, by the area-of-interest (disease in the two-dimensional scanning image
Manage region) analyzing image texture is carried out with reference to area-of-interest marked in two dimensional image, when the two-dimensional scanning image
In area-of-interest and label reference picture in area-of-interest image texture difference be greater than preset value when, then judge pathology
Feature changes (for example, current scanline image ICIn original tumor locus ROICBoundary change, i.e., and reference picture
IREFIn tumor locus ROIREFMake comparisons, the obvious pathological change for being greater than preset value occur), then enter " system update " mould
Formula, system is based on reference picture library described in currently available two-dimensional scanning image update, and based on currently available two-dimensional scanning
The spatial position coordinate data of image updates the deep learning network.Conversely, if twice in image area-of-interest image
Texture difference is less than preset value, then judges that the state of an illness of patient without obvious pathological change, then can terminate further consultation.
Or when selecting to enter " training " mode according to user, the reference that stores in the reference picture library based on user's selection
Two dimensional image and its corresponding spatial position coordinate pair current probe scanning area that refers to are adjusted, so that gained two-dimensional scanning
The spatial position coordinate and the spatial position seat stored with reference to two dimensional image that image is obtained based on the deep learning network
Mark is consistent, repeats above-mentioned two dimensional image scanning process.
In further embodiment of the present invention, as shown in table 2, Fig. 5, when predicting that error belongs to value of the non-outlier, prediction
It is compared with the sectioning image of actual position, it can be seen that the image of the two is essentially identical;Fig. 6 is when there is outlier in error
(overstriking is outlier in prediction space position parameter), the sectioning image of prediction and actual position compares, although peeling off
Value can still see that the image and true picture difference generated according to predicted position is little from image, there is much like line
Feature is managed, within the acceptable range.
2. 5 groups of table is sliced true and prediction space position parameter (value of the non-outlier) at random
It can thus be seen that (being marked on reference picture when the slice position of an area-of-interest is artificially marked
In library), which provides phase while providing current probe section present position in the tissue, for doctor in real time
The reference slice (doctor is helped to improve further consultation efficiency, save the plenty of time when needing further consultation) answered, and gradually can guide and mention
Show that the insufficient doctor of experience finds the interested region or the best section (training as doctor with pathological analysis meaning
Tool), it efficiently and accurately completes to check, plays the role of auxiliary diagnosis.
Fig. 7 shows ultrasonic intelligent positioning system according to an exemplary embodiment of the present invention, i.e., electronic equipment 310 (such as
Have program execute function computer server) comprising at least one processor 311, power supply 314, and with it is described extremely
The memory 312 and input/output interface 313 of a few processor 311 communication connection;The memory 312 is stored with can be by institute
State the instruction of at least one processor 311 execution, described instruction executed by least one described processor 311 so that it is described extremely
A few processor 311 is able to carry out method disclosed in aforementioned any embodiment;The input/output interface 313 may include
Display, keyboard, mouse and USB interface are used for inputoutput data;Power supply 314 is used to provide electricity for electronic equipment 310
Energy.
It will be appreciated by those skilled in the art that: realize that all or part of the steps of above method embodiment can pass through program
Relevant hardware is instructed to complete, program above-mentioned can store in computer-readable storage medium, which is executing
When, execute step including the steps of the foregoing method embodiments;And storage medium above-mentioned includes: movable storage device, read-only memory
The various media that can store program code such as (Read Only Memory, ROM), magnetic or disk.
When the above-mentioned integrated unit of the present invention be realized in the form of SFU software functional unit and as the sale of independent product or
In use, also can store in a computer readable storage medium.Based on this understanding, the skill of the embodiment of the present invention
Substantially the part that contributes to existing technology can be embodied in the form of software products art scheme in other words, the calculating
Machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be individual
Computer, server or network equipment etc.) execute all or part of each embodiment the method for the present invention.And it is aforementioned
Storage medium include: the various media that can store program code such as movable storage device, ROM, magnetic or disk.
The above, the only detailed description of the specific embodiment of the invention, rather than limitation of the present invention.The relevant technologies
The technical staff in field is not in the case where departing from principle and range of the invention, various replacements, modification and the improvement made
It should all be included in the protection scope of the present invention.
Claims (9)
1. a kind of ultrasound based on deep learning network adopts figure intelligent locating method, which is characterized in that described to include:
Two dimensional image and its spatial position coordinate are scanned and obtained, stores interested two dimensional image for establishing reference picture
Library;The corresponding three-dimensional data of the two dimensional image is acquired, and seeks the corresponding cuboid said three-dimensional body number of the three-dimensional data
According to, and establish the cuboid three-dimensional data space coordinates of specification;
Deep learning network is constructed, the random two-dimensional tangent plane picture generated using the cuboid three-dimensional data of specification and its space
The position coordinates training deep learning network, with deep learning network satisfaction: when there is the input of two-dimensional scanning image, from
The dynamic spatial position coordinate for exporting the corresponding virtual probe of the scan image.
2. the method according to claim 1, wherein further including being based on image analysis algorithm from three-dimensional data
Interested two dimensional image section is intercepted, and the interested two dimensional image section is sought by the deep learning network
Spatial position coordinate, by the interested two dimensional image section and its spatial position coordinate label in the reference picture library
In;
Or the three-dimensional data is generated to the 3-D image of different angle based on 3-D image method for drafting, from the three-dimensional
Interested two dimensional image section is chosen in image, and the interested two dimensional image is sought by the deep learning network
The spatial position coordinate of section, by the interested two dimensional image section and its spatial position coordinate label described with reference to figure
As in library.
3. according to the method described in claim 2, it is characterized by further comprising:
When user demand enters " training " or " further consultation " mode, the reference two that stores in the reference picture library based on user's selection
It ties up image and its is adjusted by the resulting spatial position coordinate pair current probe scanning area of deep learning network, Zhi Daosuo
The two-dimensional scanning image spatial position coordinate that is obtained based on the deep learning network and described stored with reference to two dimensional image
Spatial position coordinate difference stops the adjustment of probe positions when being less than preset value, user is prompted to select " further consultation " or " training " mode.
4. according to the method described in claim 3, it is characterized by further comprising:
When getting user's selection entrance " further consultation " mode, by the area-of-interest in the two-dimensional scanning image and with reference to two
Tie up marked area-of-interest in image and carry out analyzing image texture, when in the two-dimensional scanning image area-of-interest with
When the image texture difference of area-of-interest in the reference picture of label is greater than preset value, then judge that pathological characters become
Change, then enter " system update " mode, system is based on reference picture library described in currently available two-dimensional scanning image update, and base
The deep learning network is updated in the spatial position coordinate data of currently available two-dimensional scanning image.
5. according to the method described in claim 4, it is characterized in that, further include area-of-interest in the scan image with
When the image texture difference of area-of-interest in the reference picture of storage is less than or equal to preset value, then judge pathological characters without change
Change and stops operation.
6. the method according to claim 1, wherein the deep learning network include: an image input layer,
Multiple full articulamentums and active coating and a recurrence layer active coating use line rectification function, the loss for returning layer
Function uses half mean square error function.
7. the method according to claim 1, wherein using installing magnetic field tracking device method additional, method for registering images, adding
Dress three-dimensional volume probe one of is scanned method using mechanical arm and acquires the corresponding said three-dimensional body number of the two dimensional image
According to.
8. the method according to claim 1, wherein using described in the interpolation method generation between picture frame and frame
The cuboid three-dimensional data of three-dimensional data.
9. a kind of ultrasound based on deep learning network adopts figure intelligent positioning system, which is characterized in that including at least one processing
Device, and the memory being connect at least one described processor communication;The memory be stored with can by it is described at least one
The instruction that processor executes, described instruction is executed by least one described processor, so that at least one described processor can
Method described in any one of perform claim requirement 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910356083.0A CN110070576A (en) | 2019-04-29 | 2019-04-29 | A kind of ultrasound based on deep learning network adopts figure intelligent locating method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910356083.0A CN110070576A (en) | 2019-04-29 | 2019-04-29 | A kind of ultrasound based on deep learning network adopts figure intelligent locating method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110070576A true CN110070576A (en) | 2019-07-30 |
Family
ID=67369536
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910356083.0A Pending CN110070576A (en) | 2019-04-29 | 2019-04-29 | A kind of ultrasound based on deep learning network adopts figure intelligent locating method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110070576A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110974305A (en) * | 2019-12-13 | 2020-04-10 | 山东大学齐鲁医院 | Remote heart ultrasonic three-dimensional imaging system and method based on deep learning |
CN111178429A (en) * | 2019-11-25 | 2020-05-19 | 上海联影智能医疗科技有限公司 | System and method for providing medical guidance using patient depth images |
CN112089438A (en) * | 2020-08-31 | 2020-12-18 | 北京理工大学 | Four-dimensional reconstruction method and device based on two-dimensional ultrasonic image |
CN112201335A (en) * | 2020-07-23 | 2021-01-08 | 中国人民解放军总医院 | System and method for identifying structure in abdominal cavity under linear array ultrasonic endoscope |
CN112215843A (en) * | 2019-12-31 | 2021-01-12 | 无锡祥生医疗科技股份有限公司 | Ultrasonic intelligent imaging navigation method and device, ultrasonic equipment and storage medium |
CN112568994A (en) * | 2019-09-27 | 2021-03-30 | 首都医科大学附属北京安贞医院 | CTA and X-ray image registration method and device in thoracic aorta intracavity repair |
CN113171118A (en) * | 2021-04-06 | 2021-07-27 | 上海深至信息科技有限公司 | Ultrasonic inspection operation guiding method based on generating type countermeasure network |
CN113658168A (en) * | 2021-08-25 | 2021-11-16 | 上海联影医疗科技股份有限公司 | Method, system, terminal and storage medium for acquiring designated dose area |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102551807A (en) * | 2012-02-23 | 2012-07-11 | 东南大学 | Intelligent ultrasonic diagnostic navigation device based on three-axis gyroscope |
CN103565470A (en) * | 2012-08-07 | 2014-02-12 | 香港理工大学 | Ultrasonic image automatic annotating method and system based on three-dimensional virtual image |
CN106447707A (en) * | 2016-09-08 | 2017-02-22 | 华中科技大学 | Image real-time registration method and system |
CN107374674A (en) * | 2017-08-28 | 2017-11-24 | 深圳开立生物医疗科技股份有限公司 | A kind of ultrasonic probe scanning control method and device |
US20180153505A1 (en) * | 2016-12-07 | 2018-06-07 | Bay Labs, Inc. | Guided navigation of an ultrasound probe |
-
2019
- 2019-04-29 CN CN201910356083.0A patent/CN110070576A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102551807A (en) * | 2012-02-23 | 2012-07-11 | 东南大学 | Intelligent ultrasonic diagnostic navigation device based on three-axis gyroscope |
CN103565470A (en) * | 2012-08-07 | 2014-02-12 | 香港理工大学 | Ultrasonic image automatic annotating method and system based on three-dimensional virtual image |
CN106447707A (en) * | 2016-09-08 | 2017-02-22 | 华中科技大学 | Image real-time registration method and system |
US20180153505A1 (en) * | 2016-12-07 | 2018-06-07 | Bay Labs, Inc. | Guided navigation of an ultrasound probe |
CN107374674A (en) * | 2017-08-28 | 2017-11-24 | 深圳开立生物医疗科技股份有限公司 | A kind of ultrasonic probe scanning control method and device |
Non-Patent Citations (1)
Title |
---|
石明国等: "《医学超声设备原理、设计与应用》", 陕西科学技术出版社, pages: 481 - 484 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112568994A (en) * | 2019-09-27 | 2021-03-30 | 首都医科大学附属北京安贞医院 | CTA and X-ray image registration method and device in thoracic aorta intracavity repair |
CN111178429A (en) * | 2019-11-25 | 2020-05-19 | 上海联影智能医疗科技有限公司 | System and method for providing medical guidance using patient depth images |
CN110974305A (en) * | 2019-12-13 | 2020-04-10 | 山东大学齐鲁医院 | Remote heart ultrasonic three-dimensional imaging system and method based on deep learning |
CN110974305B (en) * | 2019-12-13 | 2021-04-27 | 山东大学齐鲁医院 | Remote heart ultrasonic three-dimensional imaging system and method based on deep learning |
CN112215843A (en) * | 2019-12-31 | 2021-01-12 | 无锡祥生医疗科技股份有限公司 | Ultrasonic intelligent imaging navigation method and device, ultrasonic equipment and storage medium |
CN112201335A (en) * | 2020-07-23 | 2021-01-08 | 中国人民解放军总医院 | System and method for identifying structure in abdominal cavity under linear array ultrasonic endoscope |
CN112201335B (en) * | 2020-07-23 | 2023-05-26 | 中国人民解放军总医院 | System and method for identifying intra-abdominal structure under linear array ultrasonic endoscope |
CN112089438A (en) * | 2020-08-31 | 2020-12-18 | 北京理工大学 | Four-dimensional reconstruction method and device based on two-dimensional ultrasonic image |
CN112089438B (en) * | 2020-08-31 | 2021-07-30 | 北京理工大学 | Four-dimensional reconstruction method and device based on two-dimensional ultrasonic image |
CN113171118A (en) * | 2021-04-06 | 2021-07-27 | 上海深至信息科技有限公司 | Ultrasonic inspection operation guiding method based on generating type countermeasure network |
CN113658168A (en) * | 2021-08-25 | 2021-11-16 | 上海联影医疗科技股份有限公司 | Method, system, terminal and storage medium for acquiring designated dose area |
CN113658168B (en) * | 2021-08-25 | 2024-04-16 | 上海联影医疗科技股份有限公司 | Method, system, terminal and storage medium for acquiring designated dose zone |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110070576A (en) | A kind of ultrasound based on deep learning network adopts figure intelligent locating method and system | |
JP6312898B2 (en) | Information processing apparatus, information processing method, and program | |
CN106102585B (en) | The display processing method and 3-D supersonic imaging method and system of three-dimensional imaging data | |
CN102920477B (en) | Device and method for determining target region boundary of medical image | |
CN102421373B (en) | Medical image diagnosis device and region-of-interest setting method therefor | |
CN110163897B (en) | Multi-modal image registration method based on synthetic ultrasound image | |
KR101805619B1 (en) | Apparatus and method for creating optimal 2-dimensional medical image automatically from 3-dimensional medical image | |
US8265363B2 (en) | Method and apparatus for automatically identifying image views in a 3D dataset | |
CN102016957B (en) | Medical training method and apparatus | |
CN110087550B (en) | Ultrasonic image display method, equipment and storage medium | |
CN109584350A (en) | Measurement point in diagnosis imaging determines | |
CN102763135A (en) | Methods for automatic segmentation and temporal tracking | |
RU2540829C2 (en) | Interactive iterative closest point algorithm for organ segmentation | |
WO2015188445A1 (en) | Point cloud three-dimensional model reconstruction method and system | |
CN106137249A (en) | Carry out registrating in the case of narrow visual field merging for multi-modal medical imaging | |
CN110313941B (en) | Data processing method, device, equipment and storage medium | |
WO2012136669A2 (en) | Measurement system for medical images | |
CN110051387A (en) | A kind of ultrasound computed tomography image rebuilding method and system based on ray theory | |
CN107209924A (en) | Utilize the survey tool of the plane projection in rendered volume imagery | |
CN110458872A (en) | The system and method for the image registration of biomethanics driving are executed using ultrasonic elastograph imaging | |
CN107106128A (en) | Supersonic imaging device and method for splitting anatomical object | |
CN106133789A (en) | For splitting image processing apparatus and the method for area-of-interest | |
CN108428245A (en) | Sliding method for registering images based on self-adapting regular item | |
CN109919916A (en) | A kind of wall shear stress optimization method and device, storage medium | |
JP5647398B2 (en) | Method and apparatus for combined four-dimensional presentation of quantitative local measurement results and morphological structures |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190730 |