CN107845106B - Utilize the medical image registration method of improved NNDR strategy - Google Patents
Utilize the medical image registration method of improved NNDR strategy Download PDFInfo
- Publication number
- CN107845106B CN107845106B CN201711143891.6A CN201711143891A CN107845106B CN 107845106 B CN107845106 B CN 107845106B CN 201711143891 A CN201711143891 A CN 201711143891A CN 107845106 B CN107845106 B CN 107845106B
- Authority
- CN
- China
- Prior art keywords
- point
- nndr
- information
- matching
- improved
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
-
- 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
- G06T2207/10108—Single photon emission computed tomography [SPECT]
-
- 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
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of medical image registration methods using improved NNDR strategy, belong to field of medical image registration.The invention proposes the improved matching process INNDR based on NNDR, the case where point of time neighbour correctly matches is increased.The present invention can significantly improve correct matched characteristic point number, to obtain better registration effect.The experimental results showed that the method for the present invention has significant advantage relative to original method and other criterion.The method of the present invention can not only find greater number of matching pair, but also the last matched accuracy of institute can be improved;The present invention tests the method for the present invention with reference picture, and demonstrates matching result using quantitative assessment.The method of the present invention precision with higher and validity as the result is shown.
Description
Technical field
The present invention relates to field of medical image registration, in particular to a kind of medical image using improved NNDR strategy is matched
Quasi- method.
Background technique
In field of medical image processing, with the rapid development of imaging device technology, image registration has been increasingly becoming
One important research direction.Nowadays, computerized tomography (CT), nuclear magnetic resonance image (MRI), single photon emission computer are utilized
The equipment such as tomography (SPECT) and positron radiation tomography (PET) can capture the image of inside of human body form and function.
These images provide effective information during the clinical diagnosis of patient and treatment, to entire medical mistake
Journey brings huge help.But since different imaging devices has respective characteristic, to the obtained figure in same position
As information there is very big difference.For example, CT detected on multiple angle directions X-ray by the attenuation after human body it
Afterwards, the faultage image of body is rebuild using mathematical method.The image obtained in this way can clearly show the internal organs of human body
And the anatomical structure of bone, but it cannot but show the functional information of organ.And PET image is on the contrary.PET detection life
The tomography that rationality active nucleus is distributed in body shows image, is the biochemical imaging to body.PET image is shown
The functional information of body, but image and unintelligible, cannot visually reflect the morphosis of body.
If very big error can be generated by being manually spatially overlapped two kinds of image, so image registration is just
There is important meaning.The purpose of image registration is exactly that the image of different mode is mapped to the same coordinate system by spatial alternation
In, in medical image, the position consistency of the image of corresponding organ in space can be made.Just by obtained image after registration
It can reflect the information of form and function simultaneously, provide more reliably foundation for curative activity.Because image registration is significant excellent
Point, it in lesion localization, PACS system, radiotherapy treatment planning, instruct nerve surgery and check many medical works of therapeutic effect
It has a wide range of applications in work.And good benefit is also achieved in actual application, is provided for doctor's decision abundant
Reliably information.
Since the effect obtained after being registrated to multimodal medical image is fairly obvious, so the research to medical figure registration is
It is necessary to.In image registration, common matching strategy has threshold method, nearest neighbor method (nearest neighbor, NN)
And nearest neighbor distance ratio method (nearest neighbor distance ratio, NNDR).But the most commonly used is nearest
Adjacent ratio method because NNDR in the setting of condition more comprehensively, rationally, and can also confirm it from last matching accuracy
Superiority.But the feature point range that NNDR considers when matching is smaller, has ignored other possible correct matched spies
Sign point.The number for reducing last correct matched characteristic point is resulted in this way, also affects the precision being finally registrated and registration
The effect of image.
Summary of the invention
In order to make up for the deficiencies of the prior art, solve to consider in medical figure registration in the prior art feature point range compared with
It is small, other correct matched characteristic points are had ignored, and cause to reduce last correct matched characteristic point number, influence registration accuracy
The problem of with registration image effect, the present invention provides a kind of medical image registration methods using improved NNDR strategy.
Term is explained
1, nearest neighbor distance ratio (NNDR), this is a kind of similitude based on Euclidean distance
Metric algorithm.In image registration work, it is used to measure the similitude between found characteristic point.This algorithm passes through threshold
The number for setting also adjustable last matched characteristic point of value.
2, SIFT, a kind of feature extracting method of classics.Because having scale invariability, rotational invariance and to making an uproar
Sound, the variation of light are all insensitive, obtain good effect.
3, RANSAC, it is a kind of that optimized parameter model is found by continuous iteration, it rejects in one group of data and does not meet optimal mould
The algorithm of type point.It is widely used in image registration and image mosaic.
The technical solution of the present invention is as follows:
A kind of medical image registration method using improved NNDR strategy, comprising steps of
1) two width multi modal medical image to be registered is inputted, wherein a width is reference picture, another width is target image;
2) characteristic point in image subject to registration is extracted using SIFT algorithm, assigns principal direction to each characteristic point, obtains spy
Position, scale and the directional information of point are levied, so that characteristic point has rotational invariance, generates feature point description of 128 dimensions;
3) target image and reference picture Feature Points Matching
A) it setsIndicate r-th of characteristic point in reference picture,Indicate the ith feature point in target image;
B) each is calculatedWith the Euclidean distance of characteristic points all in target image, distance is ranked up;
C) calculate withThe point of arest neighborsAnd the point of secondary neighbourEuclidean distance, meet formula (1) be most
Neighborhood matching M1st;
D) calculate withThe point of secondary neighbourAnd the point of third neighbourEuclidean distance, meeting formula (2) is
Secondary neighborhood matching M2nd;
E) in view of existingNot only the case where matching but also match with the point of secondary neighbour with the point of arest neighbors determines matching to Mt
Are as follows:
Mt=M1st∪M2nd-M2nd∩1st (3)。
It preferably, further include generating new feature point description before target image and reference picture Feature Points Matching
Sub-information.
Further, the method for new feature point description sub-information is generated are as follows:
The information of son will be described with gradient magnitude GM in SIFT algorithm by numerical ordering, and according to being divided into four from big to small
Grade, is arranged the value of each grade, the first order is set as 1, and the second level is set as 0.75, and the third level is set as within the scope of 0-1
0.5, the fourth stage is set as 0.25.
Preferably, the method for generating new feature point description sub-information may be:
The information that son is described in SIFT algorithm is indicated with gradient magnitude GM, increases a kind of new gradient information representation
1 is set as in GO, GO when change of gradient, 0 is set as when gradient is constant;
The GM information of 128 dimensions is directly added with the GO information of 128 dimensions, becomes the gradient information of 256 dimensions.
Preferably, in step 2), SIFT algorithm establishes scale space by gaussian pyramid, then takes Gauss
Difference finds the extreme point in scale space;The unstable fixed point of removal, obtains characteristic point.
Preferably, step 3) determines matching to MtAfterwards, using the characteristic point of RANSAC removal erroneous matching.
The invention has the benefit that
1, the invention proposes the improved matching process INNDR based on NNDR, correct matched spy can be significantly improved
Sign point number, to obtain better registration effect.The experimental results showed that the method for the present invention is relative to original method and other
Criterion have significant advantage.The method of the present invention can not only find greater number of matching pair, but also last institute can be improved
Matched accuracy;
The present invention tests the method for the present invention with reference picture, and demonstrates matching result using quantitative assessment.As a result it shows
Show the method for the present invention precision with higher and validity.
2, method proposed by the present invention can be used for the registration work of multi modal medical image;In Research of Medical multi-mode image,
It can use the information that the method analyzes the same position that different imaging devices obtain, more clearly know the specific feelings of patient
Condition, and help subsequent remedy measures.Furthermore it is possible to which this algorithm to be used for the fusion work to multi modal medical image;Thus
More fully patient information can be obtained, the accuracy for the treatment of is improved.
3, the gradient information for improving characteristic point in this method also has other fields of Medical Image Processing larger
It helps.And these types combines two kinds of gradient informations and improved strategy respectively to have feature, can be according to specific application selection
Some or several combination strategies.
4, RANSAC has been used, the accuracy of Feature Points Matching is improved.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without any creative labor, may be used also for those of ordinary skill in the art
To obtain other drawings based on these drawings.
Fig. 1 is the brain CT image of patient A;
Fig. 2 is the brain MRI image of patient A;
Fig. 3 uses the method for the present invention and existing SIFT method for image in Fig. 1, Fig. 2 for the first evaluation method
Evaluation result comparison diagram with accuracy rate;
Fig. 4 uses the method for the present invention and existing SIFT method for image in Fig. 1, Fig. 2 for second of evaluation method
Evaluation result comparison diagram with accuracy rate.
Specific embodiment
Embodiment 1
A kind of medical image registration method using improved NNDR strategy, specific steps include:
1) two width multi modal medical image to be registered is inputted, wherein a width is as reference picture, another width is as target figure
Picture;
2) characteristic point is extracted, description is established;
SIFT algorithm is taken to extract the characteristic point in image subject to registration, SIFT establishes scale space by gaussian pyramid,
Then difference of Gaussian is taken to find the extreme point in scale space;After removing some unstable fixed points, characteristic point to the end is obtained.
In order to enable characteristic point has rotational invariance, principal direction is assigned to each characteristic point.Three of characteristic point can thus be obtained
Information: position, scale and direction.Key point description of 128 dimensions can thus be generated.
3) new description sub-information.
The information of description is indicated with gradient magnitude (gradient magnitude, GM) in SIFT algorithm, still
Have the defects that certain.Such as a characteristic point adjacent area gradient information there are difference, but last knot
Fruit is the same.So the present invention explores a kind of new gradient information expression (gradient occurrence, GO), GO is just
It is when change of gradient just sets the values to 1, gradient is constant to be set to 0, the amplitude without considering change of gradient.GO can be avoided
The deficiency of GM, but discovery GO was also not highly desirable in some cases again later.Show both gradient informations by experiment
With complementary relationship, thus find combine both gradient informations and new description.
In conjunction with both gradient informations and the method for new description can be any one of following four.
1. last matching is to Mt=MGO∩MGMFor the intersection of GO and the GM matching pair respectively obtained, matched in this way
It is significantly improved to accuracy, but deficiency is matching to fewer;
2. establishing two kinds of description informations for the same characteristic point, there are 8 dimensions of smaller value in corresponding GO value in GM
It is set as 0.Because being influenced when amplitude varies less with regard to very little for GM strategy.
3. the description sub-information of 128 dimensions is divided into 4 grades according to numerical value, in (0,1) range to GM again assignment
Set gradually the value of each grade.The first order is set as 1, and the second level is set as 0.75, the third level 0.5, the fourth stage 0.25.
4. the GM information of 128 dimensions is directly added with the GO information of 128 dimensions, become the gradient information of 256 dimensions.
The experimental results showed that these four tactful accuracys rate all more last than the GM of SIFT script are high, can preferably be tied
Fruit.
4) target image and reference picture Feature Points Matching are obtained using improved NNDR matching algorithm;
It is main for the similarity measurement of key point according to Euclidean distance in image registration field.It common are three kinds of sides
Method, as soon as one is setting threshold value, as long as the Euclidean distance of key point is less than threshold value and regards matching as.WithIt indicates with reference to figure
R-th of characteristic point as in,Indicate the ith feature point in target image.The first matching strategy can indicate are as follows:
But there can be many matchings pair in this way, and be all much erroneous matching.So just producing most later
Neighborhood matching.
Nearest neighbor distance matching strategy is exactly for eachAll calculate it and all characteristic points in target image
Euclidean distance, then select the nearest characteristic point of Euclidean distanceAs withMatched point.It but has ignored so other
Characteristic point, only simply use distance metric, it is last to effect be not very good.NNDR is used in SIFT algorithm
Strategy, it considers the influence of other feature point pair matchings.NNDR is for eachCalculate it and institute in target image
After having the Euclidean distance of characteristic point, distance is ranked up.Then calculate withThe point of arest neighborsAndEurope
Formula distance regards the point of arest neighbors as matched if the ratio of distance is less than or equal to threshold value.
NNDR matching strategy is more preferable relative to other two kinds of obtained effects, so being that image registration field is most extensive now
's.However, it was found that NNDR will likely the range of correct matched characteristic point be located in the nearest point of Euclidean distance.Although it is also examined
The point of Euclidean distance time neighbour has been considered on matched influence, but has ignored and there is correct matched possibility in secondary neighbour
Property.By experimental verification, there is also many correct matched characteristic points in secondary neighbour, so being obtained more by innovatory algorithm
More matchings is to being the purpose of the present invention.Matching to the end to Mt include withM is matched in Euclidean distance arest neighbors1stAlso
There is secondary neighborhood matching M2nd。
Mt=M1st∪M2nd
When correctly matching in finding time neighbour's characteristic point, several method has been measured, has finally been taken in arest neighbors matching
Thought.Consider in the target image withInfluence of the point of third neighbour to secondary neighbour's Feature Points Matching.The condition of setting
Are as follows:
Can thus find Euclidean distance withThe characteristic point of secondary neighbour, but also need establish withUnique match
Characteristic point.Because in the presence ofNot only correctly matched with arest neighbors but also and time neighbour the case where correctly matching, at this moment choose matched spy
Sign point is to for the matched characteristic point of arest neighbors.So finally determining matching pair are as follows:
Mt=M1st∪M2nd-M2nd∩1st
It is significantly improved the experimental results showed that last correct matching logarithm can have.But also existWith target figure
The case where arest neighbors erroneous matching is still correctly matched with secondary neighbour as in, so improved matching strategy is that have great significance
's.
The first accuracy rate evaluation method: the matched accuracy rate of NNDR and INNDR is respectively indicated are as follows:
The matched accuracy rate of NNDR:
Wherein Nt,1stIndicate that arest neighbors is correctly matched to quantity, N in two imageso,1stIndicate two images arest neighbors
Total matching is to quantity.
The matched accuracy rate of INNDR:
Wherein Nt,2ndIndicate that time neighbour correctly matches to quantity, N in two imageso,2ndIndicate two images time neighbour
Total matching is to quantity.
Although sometimesBut all accu2 >=accu1, this, which just can prove that, improves later matching
Strategy is better than original strategy.And it is noted that new correct matching is all greater than the correct matching pair of script to quantity
Quantity.
(5)RANSAC
It obtains matching to after, due to considering the characteristic point of the second neighbour, so increasing matching pair.But the second neighbour
Most of characteristic point accuracy rate that there is no the first neighbours is high, it not is very that this, which results in accuracy rate last sometimes to be promoted,
Obviously.So removing the point of some erroneous matchings using RANSAC, while promoting accuracy rate.After RANSAC, INNDR is obtained
The correct number of pairs arrived is still more than NNDR, and matched accuracy rate also significantly improves.
(6) evaluation criterion
In addition to last accuracy rate mentioned aboveAlso introduce one
Recall vs 1-precision Evaluation Strategy (second of accuracy rate evaluation method).
Wherein correspondences is exactly spatially corresponding correct matched characteristic point number in two images,
Precision is then similar with accuracy.The table being finally made of two parameters can more visually indicate distinct methods
Superiority and inferiority.
Fig. 1, Fig. 2 are respectively the brain CT image and brain MRI image of same patient;
The correct matched characteristic point for using existing SIFT algorithm to obtain is 249 pair, total matching contraposition 484;
Obtained correct matched characteristic point is 290 pair after using the present invention to improve NNDR (INNDR), total matching pair
Position 349.
Fig. 3, Fig. 4 are respectively the result for using the image of Fig. 1, Fig. 2 to evaluate existing SIFT algorithm and the method for the present invention.
From the figure 3, it may be seen that being directed to different threshold values (threshold), setting range is [0.6,0.99], the method for the present invention
(INNDR) accuracy rate of the result obtained is generally higher than existing SIFT algorithm (NNDR).The improved matching strategy of the present invention
Matched accuracy can be effectively improved.
As shown in Figure 4, using second of evaluation method recall vs 1-precision, the method for the present invention (INNDR) is taken
The effect obtained is substantially better than existing SIFT algorithm (NNDR).
Claims (6)
1. a kind of medical image registration method using improved NNDR strategy, which is characterized in that comprising steps of
1) two width multi modal medical image to be registered is inputted, wherein a width is reference picture, another width is target image;
2) characteristic point in image subject to registration is extracted using SIFT algorithm, assigns principal direction to each characteristic point, obtains characteristic point
Position, scale and directional information so that characteristic point have rotational invariance, generate 128 dimension feature point description son;
3) target image and reference picture Feature Points Matching
A) it setsIndicate r-th of characteristic point in reference picture,Indicate the ith feature point in target image;
B) each is calculatedWith the Euclidean distance of characteristic points all in target image, distance is ranked up;
C) calculate withThe point of arest neighborsAnd the point of secondary neighbourEuclidean distance, meet formula (1) be arest neighbors
Match M1st;
D) calculate withThe point of secondary neighbourAnd the point of third neighbourEuclidean distance, meeting formula (2) is time neighbour
Match M2nd;
E) in view of existingNot only the case where matching but also match with the point of secondary neighbour with the point of arest neighbors determines matching to MtAre as follows:
Mt=M1st∪M2nd-M2nd∩1st (3)。
2. utilizing the medical image registration method of improved NNDR strategy as described in claim 1, it is characterised in that: target image
The feature point description sub-information new with before reference picture Feature Points Matching further including generation.
3. utilizing the medical image registration method of improved NNDR strategy as claimed in claim 2, which is characterized in that generate new
The method of feature point description sub-information are as follows:
The information of son will be described with gradient magnitude GM in SIFT algorithm by numerical ordering, and according to being divided into four grades from big to small,
The value of each grade is set within the scope of 0-1, and the first order is set as 1, and the second level is set as 0.75, and the third level is set as 0.5,
Level Four is set as 0.25.
4. utilizing the medical image registration method of improved NNDR strategy as claimed in claim 2, which is characterized in that generate new
The method of feature point description sub-information are as follows:
The information that son is described in SIFT algorithm is indicated with gradient magnitude GM, increases a kind of new gradient information representation GO, GO
It is set as 1 when middle change of gradient, 0 is set as when gradient is constant;
The GM information of 128 dimensions is directly added with the GO information of 128 dimensions, becomes the gradient information of 256 dimensions.
5. utilizing the medical image registration method of improved NNDR strategy as described in claim any one of 1-4, it is characterised in that:
In step 2), SIFT algorithm establishes scale space by gaussian pyramid, and difference of Gaussian is then taken to find in scale space
Extreme point;The unstable fixed point of removal, obtains characteristic point.
6. utilizing the medical image registration method of improved NNDR strategy as described in claim any one of 1-4, it is characterised in that:
Step 3) determines matching to MtAfterwards, using the characteristic point of RANSAC removal erroneous matching.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711143891.6A CN107845106B (en) | 2017-11-17 | 2017-11-17 | Utilize the medical image registration method of improved NNDR strategy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711143891.6A CN107845106B (en) | 2017-11-17 | 2017-11-17 | Utilize the medical image registration method of improved NNDR strategy |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107845106A CN107845106A (en) | 2018-03-27 |
CN107845106B true CN107845106B (en) | 2019-07-12 |
Family
ID=61679587
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711143891.6A Active CN107845106B (en) | 2017-11-17 | 2017-11-17 | Utilize the medical image registration method of improved NNDR strategy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107845106B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110287883A (en) * | 2019-06-26 | 2019-09-27 | 山东浪潮人工智能研究院有限公司 | A method of recognition of face is carried out based on nearest neighbor distance ratio method is improved |
CN110472643A (en) * | 2019-08-20 | 2019-11-19 | 山东浪潮人工智能研究院有限公司 | A kind of optical imagery employee's card identification method based on Feature Points Matching |
CN110490268A (en) * | 2019-08-26 | 2019-11-22 | 山东浪潮人工智能研究院有限公司 | A kind of feature matching method of the improvement nearest neighbor distance ratio based on cosine similarity |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
US9147127B2 (en) * | 2013-03-15 | 2015-09-29 | Facebook, Inc. | Verification of user photo IDs |
CN106991695A (en) * | 2017-03-27 | 2017-07-28 | 苏州希格玛科技有限公司 | A kind of method for registering images and device |
-
2017
- 2017-11-17 CN CN201711143891.6A patent/CN107845106B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
US9147127B2 (en) * | 2013-03-15 | 2015-09-29 | Facebook, Inc. | Verification of user photo IDs |
CN106991695A (en) * | 2017-03-27 | 2017-07-28 | 苏州希格玛科技有限公司 | A kind of method for registering images and device |
Also Published As
Publication number | Publication date |
---|---|
CN107845106A (en) | 2018-03-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Shaukat et al. | Artificial neural network based classification of lung nodules in CT images using intensity, shape and texture features | |
Zhang et al. | Multi‐needle localization with attention U‐net in US‐guided HDR prostate brachytherapy | |
US10269122B2 (en) | Knowledge-based automatic image segmentation | |
Zhong et al. | Boosting‐based cascaded convolutional neural networks for the segmentation of CT organs‐at‐risk in nasopharyngeal carcinoma | |
CN106651895B (en) | Method and device for segmenting three-dimensional image | |
KR20220117236A (en) | Automated tumor identification and segmentation using medical images | |
Pang et al. | CTumorGAN: a unified framework for automatic computed tomography tumor segmentation | |
JP2022539162A (en) | Image identification method and related device and equipment | |
JP2013517909A (en) | Image-based global registration applied to bronchoscopy guidance | |
CN109934812A (en) | Image processing method, device, server and storage medium | |
CN107845106B (en) | Utilize the medical image registration method of improved NNDR strategy | |
CN107292884A (en) | The method and device of oedema and hemotoncus in a kind of identification MRI image | |
US20150055848A1 (en) | Mechanism for advanced structure generation and editing | |
JP2019511268A (en) | Determination of rotational orientation in three-dimensional images of deep brain stimulation electrodes | |
CN110310723A (en) | Bone image processing method, electronic equipment and storage medium | |
CN102483862B (en) | To the improvement of Cuved planar reformation | |
CN103140875A (en) | System and method for multi-modality segmentation of internal tissue with live feedback | |
CN103518194A (en) | Knowledge-based automatic image segmentation | |
EP2584970A2 (en) | Mechanism for advanced structure generation and editing | |
US10032316B2 (en) | System for dynamically propagating real-time alterations of medical images | |
CN107977991B (en) | Medical image registration method based on space length Yu data distribution similarity measurement | |
Zeng et al. | TUSPM-NET: A multi-task model for thyroid ultrasound standard plane recognition and detection of key anatomical structures of the thyroid | |
Kieselmann et al. | Auto-segmentation of the parotid glands on MR images of head and neck cancer patients with deep learning strategies | |
CN109214451A (en) | A kind of classification method and equipment of brain exception | |
CN109087357A (en) | Scan orientation method, apparatus, computer equipment and computer readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20221214 Address after: Room 1006, Building C, Yeda Zhigu Comprehensive Building, No. 300, Changjiang Road, Yantai District, China (Shandong) Pilot Free Trade Zone, 264000, Shandong Province Patentee after: Shandong Xinjiu'an Digital Technology Co.,Ltd. Address before: 250000 science and Technology Park of West Xincheng University, Jinan City, Shandong Province Patentee before: Qilu University of Technology |
|
TR01 | Transfer of patent right |