CN108710690A - Medical image search method based on geometric verification - Google Patents
Medical image search method based on geometric verification Download PDFInfo
- Publication number
- CN108710690A CN108710690A CN201810494757.9A CN201810494757A CN108710690A CN 108710690 A CN108710690 A CN 108710690A CN 201810494757 A CN201810494757 A CN 201810494757A CN 108710690 A CN108710690 A CN 108710690A
- Authority
- CN
- China
- Prior art keywords
- image
- visual
- feature point
- feature points
- database
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012795 verification Methods 0.000 title claims abstract description 17
- 206010006187 Breast cancer Diseases 0.000 claims abstract description 37
- 208000026310 Breast neoplasm Diseases 0.000 claims abstract description 37
- 230000000007 visual effect Effects 0.000 claims description 67
- 239000013598 vector Substances 0.000 claims description 15
- 238000003064 k means clustering Methods 0.000 claims description 7
- 238000005259 measurement Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 2
- 238000010200 validation analysis Methods 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 210000000481 breast Anatomy 0.000 description 6
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 238000004195 computer-aided diagnosis Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 206010028980 Neoplasm Diseases 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 210000005075 mammary gland Anatomy 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 206010006272 Breast mass Diseases 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 201000007741 female breast cancer Diseases 0.000 description 1
- 201000002276 female breast carcinoma Diseases 0.000 description 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses the medical image search methods based on geometric verification, it is related to technical field of medical image processing, for the problems of the prior art, it is encoded by the spatial relationship obtained between image local feature on the basis of words tree (Vacabulary Tree) and inverted index (Inverted Index), then optimize initial retrieval result to reduce the probability of erroneous matching appearance by the way that whether clarifying space coding meets Geometrical consistency, improve the accuracy of breast cancer image searching result.
Description
Technical Field
The invention relates to the technical field of medical image processing, in particular to a medical image retrieval method based on geometric verification.
Background
The breast cancer is the disease with the highest incidence rate in the female malignant tumor in China and is in the trend of low age. It is difficult to effectively treat it before it spreads in the human body. There are currently a number of breast examination techniques, such as: breast X-ray, ultrasonography, magnetic resonance imaging, and the like. Among them, a mammary gland X-ray film is the most commonly used examination means. On the basis of the above, many Computer Aided Diagnosis (CAD) methods have been proposed to improve the technical level of detecting a tumor (an important index of breast cancer) in a breast X-ray film, and content-based image retrieval is one of them. The widely used retrieval method mainly comprises four steps: preprocessing an X-ray film, segmenting a mammary gland image region, extracting features, clustering and quantizing the features, and measuring the similarity.
With the extensive application and research of image retrieval methods in medical diagnosis, the methods can provide more clinical bases, and some misdiagnoses are reduced and more efficient due to insufficient clinical experience of doctors. Therefore, in the field of female breast cancer treatment, a breast cancer image retrieval method draws wide attention.
In the research of the breast cancer image retrieval method, a plurality of retrieval ideas are proposed. Tourassi G D in 2003 proposed a mutual information (mutual information) based template matching method to determine whether roi (regionof interest) in a breast image describes a true mass. The method is a knowledge-based method, a computer-aided diagnosis system establishes a knowledge database of the ROI of the breast image according to a known reference standard, each ROI in the database is regarded as a template, the computer-aided diagnosis system follows a set of template matching algorithm, mutual information is used as similarity measurement, and the query image and the breast cancer database image are matched. To determine whether the ROI of the query image describes a true tumor. According to the information content of the similar ROI, all similar ROIs in the database are retrieved and ranked, then a decision index (decision index) is calculated according to the inquired best matching result, and the decision index efficiently fuses the similarity measurement and the reference standard of the best matching template into the judgment of whether the tumor exists in the inquired breast image ROI or not.
To improve the accuracy of the search results, Tourassi added more similarity measurement methods to the breast cancer image ROI template matching in 2007, and studies showed that some similarity measurement methods had higher accuracy and others had higher search efficiency. Narzaez proposed curvature transformation of the ROI of breast mass images in 2012, and then describe the ROI by its marginal distribution. In order to improve the expandability and the retrieval efficiency of the retrieval method, Jiang adds a vocabulary tree and an inverted index method in the research of breast cancer image retrieval. The above-described method has a significant contribution in the research of breast cancer image retrieval methods, but the performance of the retrieval system is affected due to lack of scalability or no consideration of spatial attribute information of image features.
Disclosure of Invention
The embodiment of the invention provides a medical image retrieval method based on geometric verification, which can solve the problems in the prior art.
The invention provides a medical image retrieval method based on geometric verification, which comprises the following steps:
step 1, extracting SIFT characteristics from all images in a breast cancer image database, and storing the SIFT characteristics into an image characteristic database;
step 2, performing layered K-means clustering on the SIFT characteristics of all the extracted database images to generate a breast cancer image visual vocabulary tree;
step 3, establishing an inverted index for the generated breast cancer image visual vocabulary tree;
step 4, describing all breast cancer database images through visual words in a breast cancer image visual vocabulary tree;
step 5, extracting SIFT characteristics of the query image after receiving the query image, quantizing the characteristics of the query image according to the established visual vocabulary tree based on the inverted index, and describing the query image in the form of visual word vectors;
step 6, based on the inverted index, carrying out similarity measurement on the quantized query image and the quantized image in the database to form an initial retrieval result;
step 7, respectively using the matched visual words in the query image and the initial retrieval result image as reference points to define two local circular neighborhoods with the same radius;
and 8, performing geometric verification on the related feature points in the two local circular neighborhoods, and eliminating the feature points which are in error matching to obtain a final retrieval result.
Aiming at the problems, the medical image retrieval method based on geometric verification in the embodiment of the invention obtains the spatial relation codes among the local features of the images on the basis of the vocabulary Tree (Vacabulary Tree) and the Inverted Index (Inverted Index), and then optimizes the initial retrieval result by verifying whether the spatial codes accord with geometric consistency so as to reduce the probability of error matching and improve the accuracy of the retrieval result of the breast cancer image.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a medical image retrieval method based on geometric verification in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, an embodiment of the present invention provides a medical image retrieval method based on geometric verification, including the following steps:
step 1, extracting SIFT characteristics from all images in a breast cancer image database, and storing the SIFT characteristics into an image characteristic database.
And 2, performing layered K-means clustering on the SIFT characteristics of all the extracted database images to generate a breast cancer image visual vocabulary tree.
And 3, establishing an inverted index for the generated breast cancer image visual vocabulary tree.
And 4, describing all breast cancer database images through visual words in the breast cancer image visual vocabulary tree.
And 5, extracting SIFT characteristics of the query image after receiving the query image, quantizing the characteristics of the query image according to the established Visual vocabulary tree based on the inverted index, and describing the query image in a Visual word (Visual Words) vector form.
And 6, based on the inverted index, carrying out similarity measurement on the quantized query image and the quantized image in the database to form an initial retrieval result.
And 7, respectively defining two local circular neighborhoods with the same radius by taking the matched visual words in the query image and the initial retrieval result image as datum points.
And 8, performing geometric verification on the related feature points in the two local circular neighborhoods, and eliminating the feature points which are in error matching to obtain a final retrieval result.
Wherein, the step 2 specifically comprises the following substeps:
substep 201, setting parameters: firstly, a height L and the number K of subtrees of non-leaf nodes, namely the number of cluster centers of each layer, are set for a visual vocabulary tree to be created.
Sub-step 202, generating a sub-tree: and performing K-means clustering on the extracted breast cancer image SIFT feature data set to obtain K clustering centers, wherein the K clustering centers are K sub-nodes of the root node, and the data set is correspondingly divided into K subsets.
Substep 203, a recursive process: and performing K-means clustering operation on the obtained K subsets respectively to obtain K clustering centers of each subset, namely child nodes of the layer of nodes of the vocabulary tree, and iterating to perform the operation until the height of the vocabulary tree reaches L.
Through the steps, the visual vocabulary tree of the breast cancer image map is finally generated, all leaf nodes of the tree are visual words, and each visual word is an SIFT feature descriptor.
In order to express the contribution of different visual words to the image abridness measure in a quantitative way, a TF-IDF method is used to weight each visual word in the lexical tree. The performance of the retrieval system can be improved by giving different TF-IDF weights to the visual words.
The working principle of TF-IDF is as follows:
first, assuming that k visual words are shared in the visual vocabulary tree, the d-th image I in the breast cancer image database can be obtaineddExpressed as a k-dimensional vector:
vi＝(t1,t2,…,tk)T(1)
vector viOf the ith vector element tiThe expression is as follows:
ti＝tfi,d×idfi(2)
equation (2)) Of (1), tfi,dIndicating the frequency, idf, of the ith visual word in the visual dictionary appearing in the d-th image in the image databaseiDenotes the Inverse file Frequency (inverter Document Frequency), tf of the ith visual word in the visual dictionary in all images in the image databasei,dAnd idfiThe calculation formula is as follows:
n in the formula (3)i,dRepresenting the number of times the ith visual word in the visual dictionary appears in the d image of the image database, ndRepresenting the total number of all visual word occurrences counted in the database image. N in equation (4) is the total number of images in the image database, NiRepresenting the total number of times the ith visual word appears in the image database image. Thus, the image vector viThe ith vector element expression in (1) is as follows:
the operation mechanism of the TF-IDF weighting algorithm can be seen from expression (5), wherein the higher the frequency of a certain visual word in the visual dictionary appearing in one image is, namely ni,dThe larger the value, the word frequency tfi,dThe larger, this means that the visual word is more representative; while the fewer the number of times the visual word appears in the other images in the image database, i.e., NiThe smaller the value, the reverse file frequency idfiThe larger the value, and the higher the degree of discrimination of the visual word.
After a visual vocabulary tree is established for the breast cancer image in the database and the weighting operation is carried out on the visual words in the vocabulary tree, the database image and the query image can be expressed by visual word vectors, each vector element is a value obtained after the visual word appears in the image after the frequency weighting, and the characteristic quantization of the breast cancer image is completed.
The step 7 specifically comprises the following substeps:
substep 701 defining ROI of the query image as IqDefining ROI of an initial search result image as IdQ and D represent SIFT feature point sets on the query image and the initial search result image ROI, respectively, and a set M (Q, D) { (Q, D) of matching feature point pairs (hereinafter simply referred to as matching pairs) can be obtainedi,di)|qi∈Q,di∈D}，qiAnd diRespectively representing mutually matched feature points on the query image and the initial retrieval result image ROI, and selecting one pair of matched feature points as a reference feature point pair (q)i,di) To verify whether the matching pair is a false match.
Substep 702, after selecting a matching pair, defining a local circular neighborhood of the matching pair, wherein the radii of two corresponding circular areas are defined as follows:
wherein scliAnd scli' represents the scale parameters of the query image feature points and the initial search result image feature points in the matching pair, and delta is a parameter set for controlling the local circular neighborhood range.
The step 8 specifically comprises the following substeps:
substep 801, excluding irrelevant feature points outside the circular neighborhood in substep 702, is formulated as follows:
wherein,andrespectively, represent the set of feature points located in a local circular neighborhood on the query image and the initial search image ROI, dist (,) represents the euclidean distance between the relevant feature points and the reference feature points.
Substep 802, extracting the matched related feature points in the two local circular neighborhoods in the following way:
in sub-step 803, since the direction of the selected reference feature point is not necessarily parallel to the coordinate axis direction, a rotation adjustment is required to be made to take the direction of the reference feature point as a new coordinate axis, so that the position of the relevant feature point is rotated. The position of the relevant feature point after the rotation adjustment is calculated as follows:
wherein, thetaiRepresenting the main gradient direction of the fiducial feature point SIFT.
And a substep 804, in order to measure the geometric consistency of the relevant feature points, encoding the geometric position relationship of each pair of matched relevant feature points relative to the respective reference feature points on the divided quadrant circles after the relevant feature points are rotated and adjusted so as to verify whether the relevant feature points fall into the same divided region. The calculation formula is as follows:
c1_ LR (q) if the relevant feature point falls to the left of the reference feature pointi,qj) A value of 0, instead falling to the right, C1_ LR (q)i,qj) The value is 1. Similarly, if the relevant feature point falls below the reference feature point, C1_ UD (q)i,qj) A value of 0, instead falling above, C1_ UD (q)i,qj) The value is 1.
Sub-step 805, in order to generate another 4 equally divided sector areas in the local circular neighborhood, the local circular area is rotated clockwise by an angle of pi/4 around the reference feature point, and then the new position of the relevant feature point becomes:
two new codes C2_ LR (q) are calculatedi,qj) And C2_ UD (q)i,qj)：
The geometric encoding of the relevant feature points relative to the reference feature points in the extracted ROI of the query image is calculatedThe code, in the same way, can also calculate the geometric code of the relevant feature point relative to the reference feature point in the breast cancer image of the initial search result: c1_ LR (d)i,dj)，C1_UD(di,dj)，C2_LR(di,dj) And C2_ UD (d)i,dj)。
Substep 806, performing xor operation on the 8 geometric codes obtained in the geometric coding process:
wherein ⊕ is an exclusive-or operation, and assuming that a relevant feature point in the local circular neighborhood of the reference feature point in the query image ROI is located in the left sectorized region of the reference feature point, as in (17), if the relevant feature point matching it in the breast cancer image of the initial search result is also located in the same-oriented sectorized region of the reference feature point, V1_ LR ((q) is V1 — LR ((q) is a result of the initial searchi,qj),(di,dj) ) a value of 0, otherwise a value of 1.
Then, V1_ LR ((q) for all relevant feature points in the reference feature point local circular region, respectivelyi,qj),(di,dj) And V2_ LR ((q)i,qj),(di,dj) And are) andandand accumulating and then summing, and scoring the rationality of the geometric relationship of the reference feature points, wherein the rationality is expressed as follows:
substep 807 of comparing S _ LR (q)i,di) And S _ UD (q)i,di) The summation yields a final score of the plausibility of the geometric relationship of the reference points, which is expressed as follows:
Sco(qi,di)＝S_LR(qi,di)+S_UD(qi,di) (23)
at this time, a threshold value is setMixing Sco (q)i,di) The comparison with which to exclude false matches is expressed as follows:
wherein,the number of the relevant characteristic points matched with each other in the local circular field of the reference characteristic points is shown, and omega is a weighting coefficient. If the match is correct then C (q)i,di) The value is 0, otherwise the value is 1.
The steps are completed, and a pair of related feature points in the local circular neighborhoods of the visual words matched with each other are obtainedAnd verifying the relation, then verifying the geometrical relation of the related feature points in the local circular neighborhood of all the visual words matched with each other in the breast cancer query image ROI and the initial search result ROI, and verifying all the C (q)i,di) Performing statistics according to C (q) of ROI in each initial search result imagei,di) And the statistical results sort the initial retrieval results in an ascending order again, so that the retrieval results with high matching accuracy are ranked at the front position.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (5)
1. Medical image retrieval method based on geometric verification, characterized in that the method comprises the following steps:
step 1, extracting SIFT characteristics from all images in a breast cancer image database, and storing the SIFT characteristics into an image characteristic database;
step 2, performing layered K-means clustering on the SIFT characteristics of all the extracted database images to generate a breast cancer image visual vocabulary tree;
step 3, establishing an inverted index for the generated breast cancer image visual vocabulary tree;
step 4, describing all breast cancer database images through visual words in a breast cancer image visual vocabulary tree;
step 5, extracting SIFT characteristics of the query image after receiving the query image, quantizing the characteristics of the query image according to the established visual vocabulary tree based on the inverted index, and describing the query image in the form of visual word vectors;
step 6, based on the inverted index, carrying out similarity measurement on the quantized query image and the quantized image in the database to form an initial retrieval result;
step 7, respectively using the matched visual words in the query image and the initial retrieval result image as reference points to define two local circular neighborhoods with the same radius;
and 8, performing geometric verification on the related feature points in the two local circular neighborhoods, and eliminating the feature points which are in error matching to obtain a final retrieval result.
2. The medical image retrieval method based on geometric verification as claimed in claim 1, wherein the step 2 specifically comprises the following sub-steps:
substep 201, setting parameters: firstly, setting a height L and the number K of subtrees of non-leaf nodes, namely the number of clustering centers of each layer, for a visual vocabulary tree to be created;
sub-step 202, generating a sub-tree: performing K-means clustering on the extracted breast cancer image SIFT feature data set to obtain K clustering centers, wherein the K clustering centers are K sub-nodes of the root node, and the data set is correspondingly divided into K sub-sets;
substep 203, a recursive process: and performing K-means clustering operation on the obtained K subsets respectively to obtain K clustering centers of each subset, namely child nodes of the layer of nodes of the vocabulary tree, and iterating to perform the operation until the height of the vocabulary tree reaches L.
3. The geometric validation-based medical image retrieval method of claim 2, wherein each visual word in the lexical tree is weighted using a TF-IDF method to quantitatively express the magnitude of the contribution of different visual words to the image measure of abridness, the TF-IDF method comprising:
the vision vocabulary tree shares k vision words, and the d-th image I in the breast cancer image databasedExpressed as a k-dimensional vector:
vi＝(t1,t2,…,tk)T(1)
vector viOf the ith vector element tiThe expression is as follows:
ti＝tfi,d×idfi(2)
in the formula (2), tfi,dIndicating the frequency, idf, of the ith visual word in the visual dictionary appearing in the d-th image in the image databaseiRepresenting the inverse file frequency, tf, of the ith visual word in the visual dictionary in all images in the image databasei,dAnd idfiThe calculation formula is as follows:
n in the formula (3)i,dRepresenting the number of times the ith visual word in the visual dictionary appears in the d image of the image database, ndRepresenting the total number of all visual word occurrences counted in the database image of the d-th image, N in equation (4) being the total number of images in the database image, NiRepresents the total number of times the ith visual word appears in the image database image, and thus, the image vector viThe ith vector element expression in (1) is as follows:
the higher the frequency of a visual word in the visual dictionary appearing in an image, i.e. ni,dThe larger the value, the word frequency tfi,dThe larger the representation of the viewThe more representative the sense word is; while the fewer the number of times the visual word appears in the other images in the image database, i.e., NiThe smaller the value, the reverse file frequency idfiThe larger the value, and the higher the degree of discrimination of the visual word.
4. The medical image retrieval method based on geometric verification as claimed in claim 1, wherein step 7 specifically includes the following sub-steps:
substep 701 defining ROI of the query image as IqDefining ROI of an initial search result image as IdQ and D respectively represent SIFT feature point sets on the query image and the initial search result image ROI, and a set M (Q, D) of matched feature point pairs is obtainedi,di)|qi∈Q,di∈D}，qiAnd diRespectively representing mutually matched feature points on the query image and the initial retrieval result image ROI, and selecting one pair of matched feature points as a reference feature point pair (q)i,di) To verify whether the matching pair is a false match;
substep 702, after selecting a matching pair, defining a local circular neighborhood of the matching pair, wherein the radii of two corresponding circular areas are defined as follows:
wherein scliAnd scli' represents the scale parameters of the query image feature points and the initial search result image feature points in the matching pair, and delta is a parameter set for controlling the local circular neighborhood range.
5. The medical image retrieval method based on geometric verification as claimed in claim 4, wherein the step 8 specifically comprises the following sub-steps:
substep 801, excluding irrelevant feature points outside the circular neighborhood in substep 702, is formulated as follows:
wherein,andrespectively representing feature point sets positioned in local circular neighborhoods on the query image and the initial retrieval image ROI, dist (,) representing the Euclidean distance between the related feature point and the reference feature point;
substep 802, extracting the matched related feature points in the two local circular neighborhoods in the following way:
in sub-step 803, the direction of the reference feature point is used as a new coordinate axis, and the position of the relevant feature point after the rotation adjustment is calculated as follows:
wherein, thetaiRepresenting the main gradient direction of the reference feature point SIFT;
and a substep 804 of encoding the geometric position relationship of each pair of matched related feature points relative to the respective reference feature points on the divided quadrant circle to verify whether the matched related feature points fall into the same divided region, wherein the calculation formula is as follows:
c1_ LR (q) if the relevant feature point falls to the left of the reference feature pointi,qj) A value of 0, instead falling to the right, C1_ LR (q)i,qj) The value is 1, and similarly, if the relevant feature point falls below the reference feature point, C1_ UD (q)i,qj) A value of 0, instead falling above, C1_ UD (q)i,qj) A value of 1;
substep 805, rotating the local circular region clockwise by an angle of pi/4 around the reference feature point to generate another 4 equally divided sector regions in the local circular neighborhood then the new positions of the relevant feature points become:
two new codes C2_ LR (q) are calculatedi,qj) And C2_ UD (q)i,qj)：
In the same manner, the geometric code of the relevant feature point relative to the reference feature point is calculated in the breast cancer image obtained by the initial search: c1_ LR (d)i,dj)，C1_UD(di,dj)，C2_LR(di,dj) And C2_ UD (d)i,dj)；
Substep 806, performing xor operation on the 8 geometric codes obtained in the geometric coding process:
v1_ LR ((q) for all relevant feature points in the local circular region of the reference feature pointi,qj),(di,dj) And V2_ LR ((q)i,qj),(di,dj) And are) andandand accumulating and then summing, and scoring the rationality of the geometric relationship of the reference feature points, wherein the rationality is expressed as follows:
substep 807 of comparing S _ LR (q)i,di) And S _ UD (q)i,di) The summation yields a final score of the plausibility of the geometric relationship of the reference points, which is expressed as follows:
Sco(qi,di)＝S_LR(qi,di)+S_UD(qi,di) (23)
setting a threshold valueMixing Sco (q)i,di) The comparison with which to exclude false matches is expressed as follows:
wherein,the number of the related characteristic points matched with each other in the local circular field of the reference characteristic point, omega is a weighting coefficient, and if the matching is correct, C (q)i,di) The value is 0, otherwise the value is 1;
the steps finish the verification of the geometric relationship of the relevant feature points in the local circular neighborhood of the matched visual words, then verify the geometric relationship of the relevant feature points in the local circular neighborhood of all the matched visual words in the breast cancer query image ROI and the initial retrieval result ROI, and verify all C (q)i,di) Performing statistics according to C (q) of ROI in each initial search result imagei,di) And the statistical results sort the initial retrieval results in an ascending order again, so that the retrieval results with high matching accuracy are ranked at the front position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810494757.9A CN108710690A (en) | 2018-05-22 | 2018-05-22 | Medical image search method based on geometric verification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810494757.9A CN108710690A (en) | 2018-05-22 | 2018-05-22 | Medical image search method based on geometric verification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108710690A true CN108710690A (en) | 2018-10-26 |
Family
ID=63868614
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810494757.9A Pending CN108710690A (en) | 2018-05-22 | 2018-05-22 | Medical image search method based on geometric verification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108710690A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110263198A (en) * | 2019-06-27 | 2019-09-20 | 安徽淘云科技有限公司 | A kind of search method and device |
| CN111341437A (en) * | 2020-02-21 | 2020-06-26 | 山东大学齐鲁医院 | Digestive tract disease judgment auxiliary system based on tongue image |
| CN113378842A (en) * | 2021-05-18 | 2021-09-10 | 浙江大学 | Recommendation method based on segmented image feature extraction |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104615642A (en) * | 2014-12-17 | 2015-05-13 | 吉林大学 | Space verification wrong matching detection method based on local neighborhood constrains |
| CN107045520A (en) * | 2016-11-29 | 2017-08-15 | 江南大学 | A kind of vehicle image search method that words tree is weighted based on positional information |
| CN107145829A (en) * | 2017-04-07 | 2017-09-08 | 电子科技大学 | A kind of vena metacarpea recognition methods for merging textural characteristics and scale invariant feature |
-
2018
- 2018-05-22 CN CN201810494757.9A patent/CN108710690A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104615642A (en) * | 2014-12-17 | 2015-05-13 | 吉林大学 | Space verification wrong matching detection method based on local neighborhood constrains |
| CN107045520A (en) * | 2016-11-29 | 2017-08-15 | 江南大学 | A kind of vehicle image search method that words tree is weighted based on positional information |
| CN107145829A (en) * | 2017-04-07 | 2017-09-08 | 电子科技大学 | A kind of vena metacarpea recognition methods for merging textural characteristics and scale invariant feature |
Non-Patent Citations (1)
| Title |
|---|
| QINGLIANG LI等: "Computer-aided diagnosis of mammographic masses using geometric verification-based image retrieval", 《MEDICAL IMAGING 2017: COMPUTER-AIDED DIAGNOSIS》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110263198A (en) * | 2019-06-27 | 2019-09-20 | 安徽淘云科技有限公司 | A kind of search method and device |
| CN111341437A (en) * | 2020-02-21 | 2020-06-26 | 山东大学齐鲁医院 | Digestive tract disease judgment auxiliary system based on tongue image |
| CN113378842A (en) * | 2021-05-18 | 2021-09-10 | 浙江大学 | Recommendation method based on segmented image feature extraction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Soltaninejad et al. | Automated brain tumour detection and segmentation using superpixel-based extremely randomized trees in FLAIR MRI | |
| Demyanov et al. | Classification of dermoscopy patterns using deep convolutional neural networks | |
| Zulqarnain Gilani et al. | Shape-based automatic detection of a large number of 3D facial landmarks | |
| Chen et al. | Automated contouring error detection based on supervised geometric attribute distribution models for radiation therapy: a general strategy | |
| Mastouri et al. | Deep learning-based CAD schemes for the detection and classification of lung nodules from CT images: A survey | |
| de Sousa Costa et al. | Classification of malignant and benign lung nodules using taxonomic diversity index and phylogenetic distance | |
| Ahmad et al. | [Retracted] Efficient Liver Segmentation from Computed Tomography Images Using Deep Learning | |
| CN108710690A (en) | Medical image search method based on geometric verification | |
| Ahmad et al. | SiNC: Saliency-injected neural codes for representation and efficient retrieval of medical radiographs | |
| Kuang et al. | Unsupervised multi-discriminator generative adversarial network for lung nodule malignancy classification | |
| Zhang et al. | A novel algorithm for breast mass classification in digital mammography based on feature fusion | |
| Felix et al. | Using 3D texture and margin sharpness features on classification of small pulmonary nodules | |
| Patil et al. | Improved region growing segmentation for breast cancer detection: progression of optimized fuzzy classifier | |
| Tao et al. | NSCR‐Based DenseNet for Lung Tumor Recognition Using Chest CT Image | |
| Jaffar et al. | Ensemble classification of pulmonary nodules using gradient intensity feature descriptor and differential evolution | |
| Aman et al. | Content-based image retrieval on CT colonography using rotation and scale invariant features and bag-of-words model | |
| Dong et al. | An improved YOLOv5 network for lung nodule detection | |
| Wang et al. | Multi-view fusion segmentation for brain glioma on CT images | |
| CN109886091B (en) | Three-dimensional facial expression recognition method based on weighted local rotation mode | |
| Yu et al. | Tnn: Tree neural network for airway anatomical labeling | |
| Li et al. | Improved cost-sensitive multikernel learning support vector machine algorithm based on particle swarm optimization in pulmonary nodule recognition | |
| CN110264562A (en) | Skull model characteristic point automatic calibration method | |
| Limberger et al. | Shrec'17 track: Point-cloud shape retrieval of non-rigid toys | |
| Meng et al. | Research on the ROI registration algorithm of the cardiac CT image time series | |
| CN114821157A (en) | Multi-modal image classification method based on hybrid model network |
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 |