CN106530280B - The localization method and device of organ in a kind of image - Google Patents
The localization method and device of organ in a kind of image Download PDFInfo
- Publication number
- CN106530280B CN106530280B CN201610902595.9A CN201610902595A CN106530280B CN 106530280 B CN106530280 B CN 106530280B CN 201610902595 A CN201610902595 A CN 201610902595A CN 106530280 B CN106530280 B CN 106530280B
- Authority
- CN
- China
- Prior art keywords
- image
- training
- organ
- unit
- sampled point
- 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.)
- Expired - Fee Related
Links
- 210000000056 organ Anatomy 0.000 title claims abstract description 114
- 238000000034 method Methods 0.000 title claims abstract description 75
- 230000004807 localization Effects 0.000 title claims abstract description 14
- 238000012549 training Methods 0.000 claims description 119
- 238000012360 testing method Methods 0.000 claims description 48
- 238000012545 processing Methods 0.000 claims description 30
- 230000008569 process Effects 0.000 claims description 16
- 238000005070 sampling Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 abstract description 7
- 238000010586 diagram Methods 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 7
- 238000002591 computed tomography Methods 0.000 description 3
- 210000003734 kidney Anatomy 0.000 description 3
- 238000002595 magnetic resonance imaging Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 239000000686 essence Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003709 image segmentation Methods 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
- 239000010410 layer Substances 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
- 238000002600 positron emission tomography Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30084—Kidney; Renal
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The application provides the localization method and device of organ in a kind of image, which comprises obtains image to be processed；Sampled point is chosen in the image to be processed；Offset of each sampled point relative to organ bounding box is predicted using trained recurrence device；According to the corresponding offset of each sampled point, the encirclement box position of organ in the image to be processed is determined.In the embodiment of the present application, when being positioned to the organ in image to be processed, first sampled point is chosen in image to be processed, then, utilize trained recurrence device, it predicts offset of each sampled point relative to organ bounding box, and then the position of organ bounding box in image to be processed can be predicted.The positioning accuracy for not only increasing organ in image in this way reduces calculation amount, also raising operational efficiency.
Description
Technical field
This application involves medical image technology, in particular to the localization method and device of organ in a kind of image.
Background technique
In recent years, medical imaging technology has obtained huge development, X-ray plain film, single layer of the image mode from getting up early
CT scan (CT, Computed Tomography) has developed to current multi-layer spiral CT, magnetic resonance imaging
(MRI, Magnetic Resonance Imaging), Positron Emission Computed Tomography (PET, Positron
Emission tomography)/CT etc., it is provided convenience for doctor.Under normal conditions, doctor usually requires to utilize acquisition
Medical image carries out inspection segmentation to specific histoorgan, and it is automatically fixed that this just needs first to carry out the histoorgan in the image
Position, then marks the spatial position of the histoorgan on the image, finally, using image segmentation algorithm by the organizer of label
Official splits.
Currently, there are many ways to realizing organ automatic positioning, one is: it is based on template matching method (TM, Template
Matching) method, this method create the standard form image an of certain organs first, are then existed using the template image
Scanned in bit image undetermined, find with the most similar region of template image, and using the region as obtained structures locating
Region.Another kind is based on image registration (IR, Image Registration) method, and this method selects a width Prototype drawing first
Then picture carries out position mark to organ therein manually, when needing to carry out structures locating in a width new images, first by mould
To new images, the organic region of the template image after registration is to correspond to the organic region of new images for plate image registration.
But in the method based on template matching, since the organ of different people is there is very big individual difference, it is difficult
A general organ template image is created, therefore, accuracy and robustness are poor in practical applications for this method；And based on figure
In the method for picture registration, in order to improve the accuracy of structures locating, such methods generally use non-rigid registration (NR, Non-
Rigid Registration) template image is registrated on new images by method, and the calculation amount of non-rigid registration is very big, because
This, such methods are usually very time-consuming, to reduce operational efficiency.
Bright content
In view of this, the application provides the localization method and device of organ in a kind of image, with solve in the prior art by
It is not high in positioning accuracy, lead to the problem that operational efficiency is low.
Specifically, the application is achieved by the following technical solution:
First aspect provides a kind of localization method of organ in image, which comprises
Obtain image to be processed；
Sampled point is chosen in the image to be processed；
Offset of each sampled point relative to organ bounding box is predicted using trained recurrence device；
According to the corresponding offset of each sampled point, the bounding box position of organ in the image to be processed is determined
It sets.
Second aspect provides a kind of positioning device of organ in image, comprising:
First acquisition unit, for obtaining image to be processed；
First selection unit, for choosing sampled point in the image to be processed；
Predicting unit, for predicting offset of each sampled point relative to organ bounding box using trained recurrence device
Amount；
First determination unit, for determining the image to be processed according to the corresponding offset of each sampled point
The encirclement box position of middle organ.
In the embodiment of the present application, when the organ to test image positions, first determines a sampled point, utilize training
Good recurrence device, offset of this available relative to organ bounding box, and then the prediction of an available bounding box
Value.The embodiment of the present application uses multiresolution strategy, can select within the scope of whole picture test image on lowest resolution image
Sampled point is taken, to improve positioning accuracy, calculation amount is reduced, improves operational efficiency.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The application can be limited.
Detailed description of the invention
Fig. 1 is the flow chart of the localization method of organ in a kind of image provided by the embodiments of the present application；
Fig. 2 is another flow chart of the localization method of organ in a kind of image provided by the embodiments of the present application；
Fig. 3 is another flow chart of the localization method of organ in a kind of image provided by the embodiments of the present application；
Fig. 4 is a kind of trained flow chart for returning device provided by the embodiments of the present application；
Fig. 5 is a kind of trained another flow chart for returning device provided by the embodiments of the present application；
Fig. 6 is a kind of trained another flow chart for returning device provided by the embodiments of the present application；
Fig. 7 is a kind of hardware structure diagram of equipment where the positioning device of organ in image provided by the present application；
Fig. 8 is the structural schematic diagram of the positioning device of organ in a kind of image provided by the embodiments of the present application；
Fig. 9 is another structural schematic diagram of the positioning device of organ in a kind of image provided by the embodiments of the present application；
Figure 10 is another structural schematic diagram of the positioning device of organ in a kind of image provided by the embodiments of the present application；
Figure 11 is another structural schematic diagram of the positioning device of organ in a kind of image provided by the embodiments of the present application；
Figure 12 provides a kind of schematic diagram of application example for the embodiment of the present application.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the application.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the application.
It is only to be not intended to be limiting the application merely for for the purpose of describing particular embodiments in term used in this application.
It is also intended in the application and the "an" of singular used in the attached claims, " described " and "the" including majority
Form, unless the context clearly indicates other meaning.It is also understood that term "and/or" used herein refers to and wraps
It may be combined containing one or more associated any or all of project listed.
It will be appreciated that though various information, but this may be described using term first, second, third, etc. in the application
A little information should not necessarily be limited by these terms.These terms are only used to for same type of information being distinguished from each other out.For example, not departing from
In the case where the application range, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as
One information.Depending on context, word as used in this " if " can be construed to " ... when " or " when ...
When " or " in response to determination ".
Referring to Fig. 1, Fig. 1 is the flow chart of the localization method of organ in a kind of image provided by the embodiments of the present application；Institute
The method of stating includes:
Step 101: obtaining image to be processed；
Wherein, image to be processed is the image for needing to carry out structures locating in the picture.
Step 102: choosing sampled point in the image to be processed；
It, can be in the uniform selection sampled point in image to be processed, for example, when image to be processed is minimum point in the step
When resolution image, sampled point can be uniformly chosen in whole picture image range to be processed.And the mistake of sampled point is chosen in image
Journey has been known technology to those skilled in the art, and details are not described herein.
Step 103: predicting offset of each sampled point relative to organ bounding box using trained recurrence device；
In the embodiment, if image to be processed is two-dimensional medical images, offset refers to sampled point to bounding box
4 sides offset；If image to be processed is 3 d medical images, offset refers to sampled point to the 6 of bounding box
The offset in a face.
Its process predicted are as follows: each sampled point is input in trained recurrence device (Regressor), output is just
It is offset of each sampled point relative to organ bounding box.That is, return device receive input sampled point (such as
The pixel etc. of sampled point) when can select a kind of machine learning method, for example hold vector regression (SVR, Support Vector
) etc. Regression it can be obtained by offset of each sampled point relative to organ bounding box, and then can determine fallout predictor
The encirclement box position of official.Wherein, offset pair of each sampled point relative to organ bounding box is determined using machine learning method
It has been known technology, details are not described herein for those skilled in the art.
Step 104: according to the corresponding offset of each sampled point, determining that organ surrounds in the image to be processed
The position of box.
In the step, all sampled points and its corresponding offset can use, so that it may determine organ of interest packet
Enclose the position of box.Since each sampled point can obtain the position of a bounding box, the position of final bounding box can be used
The mode of (Voting) of voting obtains.That is, for each sampled point, using the space coordinate of the sampled point with
And corresponding bounding box offset predicted value, that is, it can determine that a predicted position of bounding box (is 4 for two-dimensional medical images
The X on a side, Y-axis value are the X, Y, Z axis value in 6 faces for 3 d medical images), since each sampled point can provide one
Bounding box predicted position, therefore the bounding box predicted position at most voted can be obtained i.e. as final by the way of ballot
Encirclement box position.
In the embodiment of the present application, when being positioned to the organ in image to be processed, first chosen in image to be processed
Then sampled point using trained recurrence device, predicts offset of each sampled point relative to organ bounding box, Jin Erke
To predict the position of organ bounding box in image to be processed.The positioning accurate of organ in image is not only increased in this way
Degree, reduces calculation amount, also raising operational efficiency.
Also referring to Fig. 2, Fig. 2 is another process of the localization method of organ in a kind of image provided by the embodiments of the present application
Figure, which comprises
Step 201: obtaining image to be processed；
Step 201 is detailed in step 101, and details are not described herein.
Step 202: down-sampled processing at least once being carried out to the image to be processed, obtains at least one down-sampled figure
Picture；
In the step, down-sampled processing is exactly the resolution ratio reduction of image, such as: assuming that the resolution ratio of former image to be processed
(or pixel) be 512 × 512, by 1 time two times it is down-sampled after image resolution ratio be 256 × 256, adopted by 2 two times of drops
Picture size after sample is 128 × 128, and so on.
Step 203: the down-sampled image and former image to be processed are arranged according to the sequence of resolution ratio from low to high
Sequence, and be handled as follows since the minimum image of resolution ratio；
Step 204: judging whether current image to be tested is lowest resolution image, if so, executing step 205；Such as
Fruit is not to execute step 208；
A kind of its judgment mode can judge according to picture size, if the size of current image to be tested with minimum point
The size of resolution image is consistent, it is determined that current image to be tested is exactly lowest resolution image；Otherwise, it determines current to be tested
Image is not lowest resolution image.
Step 205: choosing sampled point in the current image to be tested；
The process that sampled point is chosen in the step is specifically detailed in above-mentioned steps 102, and details are not described herein.
Step 206: predicting offset of each sampled point relative to organ bounding box using trained recurrence device；
Wherein, step 206 is detailed in step 103, and details are not described herein.
Step 207: according to the corresponding offset of each sampled point, determining organ in the current test image
Surround box position；
The step 207 is detailed in step 104, and details are not described herein.
Step 208: the bounding box that prediction obtains organ on the low level-one image in different resolution of the current test image is reflected
It is mapped on the current resolution image, the predicted value as bounding box in the current image to be tested；
Such as: on low level-one image in different resolution 4 apex coordinates of bounding box be (50,50), (50,100), (100,
50), (100,100), then on current resolution image bounding box 4 apex coordinates be (100,100), (100,200),
(200,100), (200,200), i.e., by the coordinate on bounding box vertex on low level-one image in different resolution all multiplied by 2.
Step 209: choosing sampled point in preset range around the bounding box；Step 206 and step are executed later
207。
The selection process, can be and uniformly choose sampled point near bounding box, be also possible to non-homogeneous selection sampling
Point.Its process for choosing sampled point has been known technology, details are not described herein to those skilled in the art.
In the embodiment of the present application, when the organ to image to be processed positions, first determination is adopted in image to be processed
Sampling point P1, using trained recurrence device, offset of the available sampled point relative to organ bounding box, and then can obtain
To the predicted value B of an organ bounding box1.When sampled point apart from organ of interest farther out when, the organ bounding box precision of prediction
It is relatively poor, therefore there is no directly by predicted value B by the present invention1As final prediction result, but it is pre- in organ bounding box
Measured value B1Near one point P of stochastical sampling again2, and then a new predicted value B for obtaining organ bounding box2, due to P2Distance
Bounding box is closer, therefore the predicted value B for the organ bounding box predicted2It is more accurate.Certainly, the above process can be with iteration n times, most
The predicted value B of the higher organ bounding box of a precision is obtained eventuallyN.That is, since the application uses iterated revision plan
Slightly, compared with traditional method based on structures locating, positioning when farther out is not only solved due to sampled point apart from organ of interest
The not high problem of precision, and also improve the accuracy and robustness of structures locating.
Further, multiresolution strategy is additionally used in the embodiment of the present application, it can be in whole picture on lowest resolution image
It is sampled in image range, and is only sampled near organ bounding box on higher resolution image, thus greatly
The calculation amount of this method is reduced, operational efficiency is improved.
Also referring to Fig. 3, Fig. 3 is another process of the localization method of organ in a kind of image provided by the embodiments of the present application
Figure, the embodiment is on the basis of above-mentioned Fig. 2 embodiment, can also include judging institute after determining the encirclement box position
State whether current test image is highest resolution image, implement process are as follows:
Step 301 is detailed in step 201 to step 209 to step 309；
Step 310: judging whether the current test image is highest resolution image, if so, executing step 311；
If it is not, executing step 312；
In the step, highest resolution image is exactly former test image.
Step 311: determining that the organ in the current test image surrounds box position is that final organ surrounds box position, defeated
The encirclement box position out；
Step 312: the high level-one image in different resolution of the creation current test image；Return step 304.
In the embodiment, judge whether current test image is highest resolution image, i.e. original test image, if
It is the bounding box for then exporting prediction and obtaining；Otherwise, high level-one resolution chart of the creation test image relative to current resolution
Picture, creation method are to generally use to carry out a liter method for sampling to current test image.For example, if current test image
Resolution ratio be 128 × 128, then high level-one image in different resolution be 256 × 256.
In another embodiment, which can also include that training returns device on the basis of above-mentioned all embodiments,
The recurrence device is used to predict a sampled point to the offset of organ bounding box.Wherein, the training process that training returns device can
To be trained in advance to before carrying out structures locating in a width new images, can also be carried out when to structures locating in new images
Training, the present embodiment is with no restriction.
Wherein, a kind of trained flow chart for returning device is as shown in Figure 4, comprising:
Step 401: obtaining one group of training image that device is returned for training；
Step 402: the bounding box of organ is marked in every width training image；
In the embodiment, first choose one group for training recurrence device training of medical image, and in every width medical image
Manual markings go out the bounding box of organ of interest.Wherein, interested organ is exactly the bounding box for predicting organ, for example, liver
The bounding box etc. of organ.
Step 403: calculating offset of each sampled point relative to the organ bounding box；
In this step, if training image is two-dimensional medical images, offset refers to sampled point to the 4 of bounding box
The offset on a side；If training image is 3 d medical images, offset refers to sampled point to 6 faces of bounding box
Offset, specific calculating process have been known technologies, details are not described herein to those skilled in the art.
Step 404: utilizing each sampled point and the corresponding offset of each sampled point, train on the training image
A recurrence device.
In the step, using each sampled point and each sampled point relative to organ bounding box offset as training number
According to training a recurrence device on current resolution image, training method is not limited to a certain ad hoc approach, optional
Method include support vector regression method, decision tree method etc..
Also referring to Fig. 5, Fig. 5 is a kind of trained another flow chart for returning device provided by the embodiments of the present application, the implementation
Example include:
Step 501: obtaining one group of training image for returning device for training, mark organ packet in every width training image
Enclose box；
Step 502: down-sampled processing at least once being carried out to every width training image, obtains at least one down-sampled figure
Picture；
Wherein, down-sampled processing is exactly that the resolution ratio of image reduces, such as: assuming that original image to be processed resolution ratio (or
Pixel) be 512 × 512, by 1 time two times it is down-sampled after image resolution ratio be 256 × 256, by 2 times two times it is down-sampled after
Picture size be 128 × 128, and so on.
Step 503: the down-sampled image and former training image are ranked up according to the sequence of resolution ratio from low to high,
It is successively handled as follows since the minimum image of resolution ratio:
Step 504: judging whether current training image is lowest resolution image, if so, executing step 505；If
It is no, execute step 506；
In the embodiment, a kind of mode of judgement can judge according to picture size, if current training image
Size is consistent with the size of lowest resolution image, then illustrates that current training image is lowest resolution image；Otherwise, it determines working as
Preceding training image is not lowest resolution image.
Step 505: choosing sampled point in the current training image；Then step 507 is executed；
That is, can uniformly choose and adopt in entire image when current training image is lowest resolution image
Sampling point, it is of course also possible to non-homogeneous selection sampled point, to improve the precision of prediction bounding box.
Step 506: choosing sampled point in preset range around the bounding box；Then step 507 is executed；
When current training image is not lowest resolution image, sampling is only chosen near the bounding box in preset range
Point, because the probability that the point closer apart from bounding box is sampled is higher, that is to say, that sampled point closer from bounding box is more,
Train obtained recurrence device when being predicted in this way, the sampled point of input is closer apart from bounding box, and obtained prediction result is got over
Accurately.
Step 507: calculating offset of each sampled point relative to organ bounding box；
Each sampled point in the step, which can be, chooses sampled point in current training image, be also possible to bounding box
Sampled point is chosen in surrounding preset range.Its specific calculating process has been known technology for those skilled in the art, herein
It repeats no more.
Step 508: utilizing each sampled point and the corresponding offset of each sampled point, train on current training image
A recurrence device.
In the step, calculated each sampled point and each of which sampled point are made relative to the offset of organ bounding box
For training data, a recurrence device on current resolution image is trained, training method is not limited to a certain specific
Method, optional method include support vector regression method, decision tree method etc..
In the embodiment of the present application, when to structures locating is carried out in a width test image, first training returns device, then utilizes
Trained recurrence device predicts offset of the sampled point chosen by iterated revision strategy relative to organ bounding box.In order to
The precision for further increasing the organ bounding box finally predicted, when training returns device, the application uses improved selection side
Formula, i.e., the probability that point closer apart from organ bounding box is sampled are higher, that is to say, that sampled point closer from organ bounding box
It is more, train obtained recurrence device when being predicted in this way, the sampled point of input is closer apart from organ bounding box, and what is obtained is pre-
It is more accurate to survey result.
Also referring to Fig. 6, Fig. 6 is a kind of trained another flow chart for returning device provided by the embodiments of the present application, the implementation
Example can also include judging whether current training image is highest resolution image on the basis of above-mentioned Fig. 5 embodiment.It is specific
The realization process includes:
Step 601 to step 608 is detailed in step 501 to step 508；
Step 609: judging whether current training image is highest resolution image；If so, executing step 610；If
It is no, execute step 611；
Wherein, a kind of judgement is are as follows: if the size of current training image is consistent with the size of highest resolution image,
Illustrate that current training image is highest resolution image；Otherwise, it determines current training image is not highest resolution image.And most
High-definition picture is exactly former training image.
Step 610: the recurrence device on the training image that output training obtains.
Step 611: the high level-one image in different resolution of the creation current training image；Return step 604.
In the embodiment, judge whether current training image is highest resolution image, i.e., original training image, if
It is the recurrence device for then exporting training and obtaining, process terminates；Otherwise, high level-one of the creation training image relative to current resolution
Image in different resolution creates mode, for example, if the current resolution of training image is 128 × 128, a high class resolution ratio
Image is 256 × 256.
In the present patent application embodiment, when training returns device and carries out structures locating using trained recurrence device,
It requires to sample a large amount of point in the picture, if the quantity of sampled point is excessive, will affect training and using the effect for returning device
Rate.In order to solve this problem, the embodiment of the present application uses multiresolution strategy, can be entire on lowest resolution image
It is sampled in image range, and is only sampled near bounding box on higher resolution image, to greatly reduce
The calculation amount of this method improves operational efficiency.
Corresponding with the embodiment of the localization method of organ in aforementioned image, present invention also provides organs in image to determine
The embodiment of position device.
The embodiment of the positioning device of organ can be applied in equipment in the application image.Installation practice can pass through
Software realization can also be realized by way of hardware or software and hardware combining.Taking software implementation as an example, it anticipates as a logic
Device in justice is to be read computer program instructions corresponding in nonvolatile memory by the processor of equipment where it
Into memory, operation is formed.For hardware view, as shown in fig. 7, for the positioning device place of organ in the application image
A kind of hardware structure diagram of equipment, in addition to processor shown in Fig. 7, memory, network interface and nonvolatile memory it
Outside, the equipment in embodiment where device can also include other hardware, no longer to this generally according to the actual functional capability of the equipment
It repeats.
Referring to FIG. 8, being the structural schematic diagram of the positioning device of organ in a kind of image provided by the embodiments of the present application；Institute
Stating device includes: first acquisition unit 81, the first selection unit 82, predicting unit 83 and the first determination unit 84, wherein
First acquisition unit 81, for obtaining image to be processed；
First selection unit 82, for choosing sampled point in the image to be processed；
Predicting unit 83, for predicting each sampled point relative to the inclined of organ bounding box using trained recurrence device
Shifting amount；
First determination unit 84, for determining the figure to be processed according to the corresponding offset of each sampled point
The encirclement box position of organ as in.
Optionally, in another embodiment, on the basis of the above embodiments, described device can also wrap the embodiment
Include: first processing units 91 and the second processing unit 92, structural schematic diagram as shown in figure 9,
First processing units 91, for being obtained to described first before first selection unit 82 chooses sampled point
The image to be processed that unit 81 obtains carries out down-sampled processing at least once, obtains at least one down-sampled image；
The second processing unit 92, for the down-sampled image and former image to be processed according to resolution ratio from low to high
Sequence is ranked up, and is successively handled since the minimum image of resolution ratio；
First selection unit 82 is also used to choose sampled point from described the second processing unit 92 treated image.
Optionally, in another embodiment, the embodiment on the basis of the above embodiments, described the second processing unit 92
It include: the first judging unit 921 and map unit 922, wherein its structural schematic diagram is as shown in Figure 10,
First judging unit 921, for judging whether current testing image is lowest resolution image；
First selection unit 82 is also used to determine that the current testing image is in first judging unit 921
When lowest resolution image, sampled point is chosen within the scope of the testing image；
Map unit 922, for determining that the current test image is not minimum resolution in first judging unit 921
When rate image, the bounding box that prediction obtains organ on the low level-one image in different resolution of the current test image is mapped to institute
On the image in different resolution for stating current test image, as the predicted value of bounding box in the current testing image, and in the packet
It encloses and chooses sampled point around box in preset range；
The predicting unit 83 is also used to predict the preset range around the bounding box using trained recurrence device
Offset of the interior each sampled point relative to organ bounding box；
First determination unit 84 is also used to determine the device according to the offset that the predicting unit 83 is predicted
The position of official's bounding box.
Optionally, in another embodiment, on the basis of the above embodiments, described device can also wrap the embodiment
Include: second judgment unit, the second determination unit and the first creating unit (not shown) wherein,
Second judgment unit, after the position that the first determination unit 84 determines the organ bounding box is stated for place, judgement
Whether the current test image is highest resolution image；
Second determination unit, for judging that the current test image is highest resolution figure in the second judgment unit
When picture, determining that the organ in the current test image surrounds box position is that final organ surrounds box position；
First creating unit, for judging that the current test image is not highest resolution in the second judgment unit
When image, the high level-one image in different resolution of the current test image is created.
Optionally, in another embodiment, on the basis of the above embodiments, described device can also wrap the embodiment
Include: training unit 11, structural schematic diagram is as shown in figure 11, and Figure 11 is on the basis of Figure 10, wherein
Training unit 11 is connect with the predicting unit 83, returns device for training, the recurrence device is used to prediction samples
Point arrives the offset of organ bounding box.
Optionally, in another embodiment, on the basis of the above embodiments, the training unit 11 wraps the embodiment
It includes: second acquisition unit, marking unit, the second selection unit, computing unit and training subelement (not shown), wherein
Second acquisition unit, for obtain one group for training recurrence device training image；
Marking unit, for marking organ bounding box in every width training image；
Second selection unit, for choosing sampled point in the training image；
Computing unit, for calculating offset of each sampled point relative to the organ bounding box；
Training subelement, for corresponding using each sampled point and the calculated each sampled point of the computing unit
Offset trains a recurrence device on the training image.
Optionally, in another embodiment, on the basis of the above embodiments, described device can also wrap the embodiment
It includes: third processing unit and fourth processing unit (not shown),
Third processing unit is used for before second selection unit chooses sampled point, to every width training image
Down-sampled processing at least once is carried out, at least one down-sampled image is obtained；
Fourth processing unit, for the sequence to the down-sampled image and former training image according to resolution ratio from low to high
It is ranked up, successively carries out judgement processing since the minimum image of resolution ratio.
Optionally, in another embodiment, the embodiment on the basis of the above embodiments, the fourth processing unit packet
It includes: third judging unit and third selection unit (not shown), wherein
Third judging unit, for judging whether current training image is lowest resolution image；
Second selection unit is also used to determine that the current training image is minimum point in the third judging unit
When resolution image, sampled point is chosen in the current training image；
Third selection unit, for determining that the current training image is not minimum resolution in the third judging unit
When rate image, sampled point is chosen in preset range around the bounding box；
The computing unit is also used to calculate each sampled point of the third selection unit selection relative to organ packet
Enclose the offset of box；
The trained subelement is also used to utilize each sampled point and the calculated each sampled point of the computing unit
Corresponding offset trains a recurrence device on the training image.
Optionally, the training unit can also include: the 4th judging unit, output unit and the second creating unit (figure
In do not show), wherein
4th judging unit, for after the trained subelement determines a recurrence device on the training image,
Judge whether the current training image is highest resolution image；
Output unit, it is defeated for when the 4th judging unit judges that current training image is highest resolution image
The recurrence device that the second training subelement training obtains out；
Second creating unit is highest resolution image for judging current training image not in the 4th judging unit
When, create the high level-one image in different resolution of the current training image.
The function of each unit and the realization process of effect are specifically detailed in the above method and correspond to step in above-mentioned apparatus
Realization process, details are not described herein.
As can be seen from the above embodiments, as can be seen that the structures locating method that the application proposes has such as from foregoing description
Under advantage:
Since structures locating method provided by the present application uses iterated revision strategy and traditional structures locating method phase
Than, on the one hand, solve the problems, such as due to sampled point that apart from organ of interest, positioning accuracy is not high when farther out, improves method
Accuracy and robustness；On the other hand, the sample mode for improving the picture point when training returns device, i.e., get over apart from bounding box
Closely, the number of sampled point is more, and when cooperating iterated revision strategy, can be further improved positioning side proposed by the present invention
The precision of method；In another aspect, being directly trained and being tested using original image compared to conventional method, the application uses more points
Resolution strategy, i.e., when training returns device and carries out structures locating using trained recurrence device, in multi-resolution image
Upper progress improves operational efficiency to reduce calculation amount needed for the method for the present invention.
Figure 12 is also please referred to, is a kind of schematic diagram of application example provided by the present application, based on the more of the embodiment of the present application
Secondary experimental verification shows kidney in the human body CT image obtained using structures locating method provided by the embodiments of the present application in Figure 12
Dirty bounding box, the experimental results showed that, structures locating method described herein can complete kidney in 1 second and surround box position
Prediction, the box of the position of kidney bounding box specifically as indicated by the arrows in the figure, prediction result precision is higher, can satisfy and faces
Bed application demand, has good practicability.
For device embodiment, since it corresponds essentially to embodiment of the method, so related place is referring to method reality
Apply the part explanation of example.The apparatus embodiments described above are merely exemplary, wherein described be used as separation unit
The unit of explanation may or may not be physically separated, and component shown as a unit can be or can also be with
It is not physical unit, it can it is in one place, or may be distributed over multiple network units.It can be according to actual
The purpose for needing to select some or all of the modules therein to realize application scheme.Those of ordinary skill in the art are not paying
Out in the case where creative work, it can understand and implement.
The foregoing is merely the preferred embodiments of the application, not to limit the application, all essences in the application
Within mind and principle, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the application protection.
Claims (16)
1. the localization method of organ in a kind of image characterized by comprising
Obtain image to be processed；
Sampled point is chosen in the image to be processed；
Offset of each sampled point relative to organ bounding box is predicted using trained recurrence device, wherein the offset
Amount refers to sampled point to the offset on each boundary of bounding box；
According to the corresponding offset of each sampled point, the encirclement box position of organ in the image to be processed is determined.
2. the method according to claim 1, wherein being gone back before choosing sampled point in the image to be processed
Include:
Down-sampled processing at least once is carried out to the image to be processed, obtains at least one down-sampled image；
The down-sampled image and former image to be processed are ranked up according to the sequence of resolution ratio from low to high, most from resolution ratio
Low image starts successively to be handled as follows:
Judge whether current testing image is lowest resolution image；
If it is determined that the current testing image is lowest resolution image, then sampling is chosen within the scope of the testing image
Point；
It, then will be in the low level-one of the current test image if it is determined that the current test image is not lowest resolution image
The bounding box that prediction obtains organ on image in different resolution is mapped on the image in different resolution of the current test image, as described
The predicted value of bounding box in current testing image, and sampled point is chosen in preset range around the bounding box.
3. according to the method described in claim 2, it is characterized in that, after determining the encirclement box position, further includes:
Judge whether the current test image is highest resolution image；
If it is, determining that the organ in the current test image surrounds box position as final organ encirclement box position；
If it is not, then the high level-one image in different resolution of the creation current test image.
4. method according to any one of claims 1 to 3, which is characterized in that further include: training returns device, the recurrence
Device is used to prediction samples point to the offset of organ bounding box.
5. according to the method described in claim 4, it is characterized in that, the process that the training returns device includes:
Obtain one group for training recurrence device training image, mark organ bounding box in every width training image；
Sampled point is chosen in the training image；
Calculate offset of each sampled point relative to the organ bounding box；
Using each sampled point and the corresponding offset of each sampled point, a recurrence on the training image is trained
Device.
6. according to the method described in claim 5, it is characterized in that, also being wrapped before choosing sampled point in the training image
It includes:
Down-sampled processing at least once is carried out to every width training image, obtains at least one down-sampled image；
The down-sampled image and former training image are ranked up according to the sequence of resolution ratio from low to high, it is minimum from resolution ratio
Image start successively to be handled as follows:
Judge whether current training image is lowest resolution image；
If it is determined that the current training image is lowest resolution image, then sampling is chosen in the current training image
Point；
If it is determined that the current training image is not lowest resolution image, then around the bounding box in preset range
Choose sampled point.
7. according to the method described in claim 6, it is characterized in that, train on the training image recurrence device it
Afterwards, further includes:
Judge whether the current training image is highest resolution image；
If it is, a recurrence device on the training image that output training obtains；
If it is not, then the high level-one image in different resolution of the creation current training image.
8. the positioning device of organ in a kind of image characterized by comprising
First acquisition unit, for obtaining image to be processed；
First selection unit, for choosing sampled point in the image to be processed；
Predicting unit, for predicting offset of each sampled point relative to organ bounding box using trained recurrence device；
Wherein, the offset refer to sampled point to each boundary of bounding box offset；
First determination unit, for determining device in the image to be processed according to the corresponding offset of each sampled point
The encirclement box position of official.
9. device according to claim 8, which is characterized in that further include:
First processing units, for being obtained to the first acquisition unit before first selection unit chooses sampled point
Image to be processed carry out down-sampled processing at least once, obtain at least one down-sampled image；
The second processing unit, for the down-sampled image and former image to be processed according to resolution ratio sequence from low to high into
Row sequence, and judgement processing is successively carried out since the minimum image of resolution ratio.
10. device according to claim 9, which is characterized in that described the second processing unit includes:
First judging unit, for judging whether current testing image is lowest resolution image；
First selection unit is also used to determine that the current testing image is lowest resolution in first judging unit
When image, sampled point is chosen within the scope of the testing image；
Map unit, for when first judging unit determines that the current test image is not lowest resolution image,
The bounding box that prediction obtains organ on the low level-one image in different resolution of the current test image is mapped to the current survey
Attempt on the image in different resolution of picture, as the predicted value of bounding box in the current testing image, and around the bounding box
Sampled point is chosen in preset range；
The predicting unit is also used to predict using trained recurrence device every in preset range around the bounding box
Offset of a sampled point relative to organ bounding box.
11. device according to claim 10, which is characterized in that further include:
Second judgment unit, after the position that the first determination unit 84 determines the organ bounding box is stated for place, described in judgement
Whether current test image is highest resolution image；
Second determination unit, for judging that the current test image is highest resolution image in the second judgment unit
When, determining that the organ in the current test image surrounds box position is that final organ surrounds box position；
First creating unit, for judging that the current test image is not highest resolution image in the second judgment unit
When, create the high level-one image in different resolution of the current test image.
12. device according to any one of claims 8 to 11, which is characterized in that further include:
Training unit is connect with the predicting unit, returns device for training, the recurrence device is used to prediction samples point to organ
The offset of bounding box.
13. device according to claim 12, which is characterized in that the training unit includes:
Second acquisition unit, for obtain one group for training recurrence device training image；
Marking unit, for marking organ bounding box in every width training image；
Second selection unit, for choosing sampled point in the training image；
Computing unit, for calculating offset of each sampled point relative to the organ bounding box；
Training subelement, for utilizing each sampled point and the corresponding offset of the calculated each sampled point of the computing unit
Amount, trains a recurrence device on the training image.
14. device according to claim 13, which is characterized in that further include:
Third processing unit, for being carried out to every width training image before second selection unit chooses sampled point
Down-sampled processing at least once obtains at least one down-sampled image；
Fourth processing unit, for being carried out to the down-sampled image and former training image according to the sequence of resolution ratio from low to high
Sequence, successively carries out judgement processing since the minimum image of resolution ratio.
15. device according to claim 14, which is characterized in that the fourth processing unit includes:
Third judging unit, for judging whether current training image is lowest resolution image；
Second selection unit is also used to determine that the current training image is lowest resolution in the third judging unit
When image, sampled point is chosen in the current training image；
Third selection unit, for determining that the current training image is not lowest resolution figure in the third judging unit
When picture, sampled point is chosen in preset range around the bounding box；
The computing unit is also used to calculate each sampled point of the third selection unit selection relative to organ bounding box
Offset.
16. device according to claim 15, which is characterized in that the training unit further include:
4th judging unit, for judging institute after the trained subelement determines a recurrence device on the experienced image
State whether current training image is highest resolution image；
Output unit, for exporting institute when the 4th judging unit judges that current training image is highest resolution image
State the recurrence device that the second training subelement training obtains；
Second creating unit, for when it is highest resolution image that the 4th judging unit, which judges current training image not,
Create the high level-one image in different resolution of the current training image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610902595.9A CN106530280B (en) | 2016-10-17 | 2016-10-17 | The localization method and device of organ in a kind of image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610902595.9A CN106530280B (en) | 2016-10-17 | 2016-10-17 | The localization method and device of organ in a kind of image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106530280A CN106530280A (en) | 2017-03-22 |
| CN106530280B true CN106530280B (en) | 2019-06-11 |
Family
ID=58332368
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610902595.9A Expired - Fee Related CN106530280B (en) | 2016-10-17 | 2016-10-17 | The localization method and device of organ in a kind of image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106530280B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018227449A1 (en) | 2017-06-15 | 2018-12-20 | Shanghai United Imaging Healthcare Co., Ltd. | Imaging systems and methods thereof |
| CN107680134B (en) * | 2017-09-29 | 2020-06-12 | 东软医疗系统股份有限公司 | Spine calibration method, device and equipment in medical image |
| CN107742312A (en) * | 2017-10-09 | 2018-02-27 | 沈阳东软医疗系统有限公司 | A kind of method and apparatus that key point is positioned in medical image |
| CN113780477B (en) * | 2021-10-11 | 2022-07-22 | 深圳硅基智能科技有限公司 | Method and device for measuring fundus image based on deep learning of tight frame mark |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105975979A (en) * | 2016-04-22 | 2016-09-28 | 浙江大学 | Instrument detection method based on machine vision |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3017400B1 (en) * | 2013-07-02 | 2019-05-15 | Surgical Information Sciences Inc. | Method and system for a brain image pipeline and brain image region location and shape prediction |
| US9704300B2 (en) * | 2015-03-06 | 2017-07-11 | Siemens Medical Solutions Usa, Inc. | Detection of anatomy orientation using learning-based regression |
-
2016
- 2016-10-17 CN CN201610902595.9A patent/CN106530280B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105975979A (en) * | 2016-04-22 | 2016-09-28 | 浙江大学 | Instrument detection method based on machine vision |
Non-Patent Citations (2)
| Title |
|---|
| "一种运动目标多特征点的鲁棒跟踪方法研究";张泽旭等;《数据采集与处理》;20031231;第18卷(第4期);论文第423-427页 |
| "基于在线回归学习的轮廓跟踪算法";沈宋衍等;《计算机工程》;20160531;第42卷(第5期);论文第230-234页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106530280A (en) | 2017-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106530280B (en) | The localization method and device of organ in a kind of image | |
| US8121362B2 (en) | Registration of medical images using learned-based matching functions | |
| CN105027163B (en) | Scanning area determining device | |
| CN109872333A (en) | Medical image dividing method, device, computer equipment and storage medium | |
| KR20220117236A (en) | Automated tumor identification and segmentation using medical images | |
| CN110363802B (en) | Prostate image registration system and method based on automatic segmentation and pelvis alignment | |
| Lukashevich et al. | Medical image registration based on SURF detector | |
| CN109754447A (en) | Image generating method, device, equipment and storage medium | |
| CN107545309A (en) | Scored using the picture quality of depth generation machine learning model | |
| Wells III | Statistical object recognition | |
| US20120013710A1 (en) | System and method for geometric modeling using multiple data acquisition means | |
| CN107209240A (en) | For the automatically scanning planning of follow-up magnetic resonance imaging | |
| CN110223389B (en) | Scene modeling method, system and device fusing image and laser data | |
| CN111488872B (en) | Image detection method, image detection device, computer equipment and storage medium | |
| US20220335600A1 (en) | Method, device, and storage medium for lesion segmentation and recist diameter prediction via click-driven attention and dual-path connection | |
| Herl et al. | Scanning trajectory optimisation using a quantitative Tuybased local quality estimation for robot-based X-ray computed tomography | |
| GB2472142A (en) | Method and apparatus for the registration of medical images | |
| CN111833237A (en) | Image registration method based on convolutional neural network and local homography transformation | |
| CN109901123A (en) | Transducer calibration method, device, computer equipment and storage medium | |
| Joris et al. | HemoVision: An automated and virtual approach to bloodstain pattern analysis | |
| CN113116377B (en) | Ultrasonic imaging navigation method, ultrasonic equipment and storage medium | |
| US20220301156A1 (en) | Method and system for annotation efficient learning for medical image analysis | |
| CN109920002B (en) | Characteristic point positioning method in cephalometry image based on three-dimensional random forest model | |
| CN113538372B (en) | Three-dimensional target detection method and device, computer equipment and storage medium | |
| Shahedi et al. | Spatially varying accuracy and reproducibility of prostate segmentation in magnetic resonance images using manual and semiautomated methods |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 110167 No. 177-1 Innovation Road, Hunnan District, Shenyang City, Liaoning Province Applicant after: Shenyang Neusoft Medical Systems Co.,Ltd. Address before: Hunnan New Century Road 110179 Shenyang city of Liaoning Province, No. 16 Applicant before: SHENYANG NEUSOFT MEDICAL SYSTEMS Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190611 |