CN102053963A - Retrieval method of chest X-ray image - Google Patents
Retrieval method of chest X-ray image Download PDFInfo
- Publication number
- CN102053963A CN102053963A CN2009101882450A CN200910188245A CN102053963A CN 102053963 A CN102053963 A CN 102053963A CN 2009101882450 A CN2009101882450 A CN 2009101882450A CN 200910188245 A CN200910188245 A CN 200910188245A CN 102053963 A CN102053963 A CN 102053963A
- Authority
- CN
- China
- Prior art keywords
- image
- chest
- ray image
- semantic
- 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.)
- Granted
Links
Images
Abstract
The invention belongs to the field of computer information retrieval, in particular to a retrieval method of a chest X-ray image, comprising the following steps in sequence: (1) inputting a chest X-ray image; (2) judging whether semantic description exists in an image file or not, and extracting the semantic description in the image file if the semantic description exists, or performing the semantic extraction and the semantic description for the input image if the semantic description does not exist; (3) analyzing the semanteme to form the standard text information; (4) performing the text retrieval; and (5) obtaining a similar image result set. The semantic extraction and the semantic description comprise the following steps in sequence: (a) standardizing the image; (b) partitioning a lung field; (c) performing the regional division according to the anatomical position; (d) extracting textural features; and (e) performing the semantic mapping and the semantic description on the image to be described. By utilizing the method provided by the invention, the manual work and the conditional dependence can be reduced, and the retrieval speed is fast, and the efficiency is high.
Description
Technical field:
The invention belongs to the computer information retrieval field, relate generally to the extraction of the cutting apart of chest x-ray image middle field zone, textural characteristics, the chest x-ray image search method of high-level semantic mapping.
Background technology:
The chest x-ray image is the conventional means of medicals diagnosis on disease such as the heart, lung.Along with the widespread use of computer technology at medical field, and PACS (Picture Archiving and Communication Systems, image archiving and communication system) and HIS (Hospital Information System, hospital information system) application, make chest x-ray image total amount in continuous expansion, effectively retrieval will have important meaning to doctor's training, teaching, scientific research and clinical diagnosis.
Traditional retrieval is to be undertaken by the photographing information of doctor's diagnosis report and image, and this way of search has its inevitable drawback.At first, diagnosis report does not have standardization, and each doctor usually has the expression way of oneself, and the description of same symptom is had multiple vocabulary; Secondly, because doctor's workload is bigger, the description of diagnosis report is generally not detailed, and naked eyes are read sheet and are difficult to provide parameter estimation accurately.Adopt computing machine that the chest x-ray image is carried out the automatic extraction of lung's semanteme and the description that standardizes, will provide bigger convenience to the doctor.This technology is applied in existing P ACS or other related system, will increases its competitive power in market.
The extraction of lung's texture semanteme of chest x-ray image and retrieval technique are mainly by forming cutting apart with several parts such as division, feature extraction, Semantic mapping and text retrievals of the pre-service of DICOM image, lung field zone.Along with the development of Flame Image Process and CBIR technology, the research of above-mentioned each key component all has progress to a certain degree.But, in the application of reality, still have a lot of problems, mainly contain following some:
(1) though the lung field partitioning algorithm of rule-based reasoning meets people's the mode of thinking, can obtain real border, its operation efficiency is low, and combines comparatively inconvenient with follow-up processing procedure.
(2) the lung field dividing region all be with area equate or close be foundation, these division methods have been ignored the positional information of lung field, make the subregion of dividing lose comparability because of anatomical location is different.
(3) present still neither one can accurately extract the algorithm of chest x-ray image middle field textural characteristics.
(4) adopt the main algorithm of present stage realize the time loss of chest x-ray picture retrieval and space consuming can't make us in actual applications accept.
Summary of the invention:
The present invention is intended to overcome the deficiencies in the prior art part and a kind of reduce hand labor and condition dependence is provided, and retrieval rate is fast, the chest x-ray image search method that efficient is high.
For achieving the above object, the present invention is achieved in that
A kind of chest x-ray image search method, can implement successively as follows:
(1) input chest x-ray image;
(2) judge whether there is semantic description in the image file, if then in described image file, extract semantic description; If not, then input picture is carried out extraction of semantics and description;
(3) semanteme resolves to received text information;
(4) carry out text search;
(5) obtain the similar image result set.
As a kind of preferred version, extraction of semantics of the present invention can in turn include the following steps with description:
(a) image standardization;
(b) lung field is cut apart;
(c) carry out area dividing according to anatomical location;
(d) texture feature extraction;
(e) image to be described, is carried out Semantic mapping and description.
As another kind of preferred version, image standardization of the present invention can in turn include the following steps:
(a) compression gray level;
(b) removing irrelevant part such as head, arm and standard is comparable size;
(c) remove noise.
Further, lung field of the present invention is cut apart and can be in turn included the following steps:
(a) set up the shape in lung field zone;
(b) set up the local appearance model;
(c) search for actual cut-point.
Further, area dividing of the present invention in turn includes the following steps:
(a) determine the area dividing point;
(b) generate the area dividing frisket.
In addition, texture feature extraction of the present invention can in turn include the following steps:
(a) extract each regional grey scale difference matrix;
(b) extract statistical characteristic value.
Secondly, Semantic mapping of the present invention can in turn include the following steps with description:
(a) java standard library is carried out artificial semantic description and classification;
(b) input picture is classified and semantic description.
Can add text message when once more, importing the chest x-ray image in the step of the present invention (1).
The standard of in the shape of setting up the lung field zone of the present invention standard being cut apart every width of cloth image in the storehouse is cut apart point set and is regarded a vector as
X＝[x
0，y
0，x
1，y
1，...x
n-1，y
n-1]
T
Wherein n is the sum of cut-point, and x and y represent that respectively the horizontal ordinate of cut-point sets up following shape to M width of cloth image in the java standard library:
Wherein P is the matrix that the proper vector of the covariance matrix of M X vector is formed, and b is the parameter that the different images to input obtains by search iteration, and Y represents active shape.
Of the present invention foundation in the local appearance model, the Grad of 2k+1 of normal direction point of getting each cut-point place profile is as the local feature vectors of this point, described normal direction is determined by its adjacent 2, obtains the mean value of the local feature vectors of M image.
Overall plan of the present invention comprises two parts, is respectively extraction of semantics and description and image retrieval.Extraction of semantics and description part will be presented to the image receiving end of PACS or other related system, also can directly insert DR/CR and take in the workstation.At first image is standardized, is partitioned into lung field and carry out area dividing according to the fixed position, its texture feature vector and normalization are proposed respectively from each zone then, adopt sorter to be classified in each zone at last, semanteme by standard category carries out semantic description, and these information will be stored in DICOM file and the corresponding database.Image retrieval partly is distributed to client, for the user provides multiple image retrieval mode, so that maximum convenience to be provided to the user.
Extraction of semantics comprises the steps: with the description part
Step 1: image standardization
At first gradation of image is quantified as 256 grades, carries out standard chest location (removing head, arm and stomach) then, again degree of comparing adjustment and removal noise with the lower part.
Step 2: lung field is cut apart
(number of cutting apart key point is fixed earlier the expert to be cut apart good java standard library, and each anatomical location of cutting apart key point is fixed) train, form a shape, add up the local feature model of each point then, in image to be split, carry out the unique point search at last, finish cutting apart of lung field zone according to shape and local feature model.
Step 3: area dividing
According to radiodiagnostics, with the lung field zone laterally be divided into interior band, middle band with in addition, vertically be divided into u'eno, middle open country and march off into political wilderness, [is divided 18 zones altogether.Guarantee as far as possible during division that identical anatomical location always is divided in same zone, each Regional Representative the position in the lung field.
Step 4: texture feature extraction
Lung marking is mainly formed by the pulmonary vascular projection, analyze at its feature, to each zone extract respectively that contrast, average gray are poor, entropy and four textural characteristics parameters of energy value (angle second moment).
Step 5: Semantic mapping and description
Adopt the method for classification Semantic mapping.At first adopt the medical terms of standard to carry out artificial semantic description, comprise positional information, lesion information and diagnostic message etc. existing standard storehouse image.Image in the java standard library is sorted out according to every class semanteme of describing by computer program then, same width of cloth image can be grouped in a plurality of classes.Secondly each the lung field zone with image to be described, is mapped in the corresponding java standard library class by sorter, adopts corresponding semanteme that this image is described at last.
Image retrieval is divided into three kinds of retrieval modes according to input condition.
First kind of mode is text retrieval.Select by the direct input text of user or from the text condition options that system provides, retrieve corresponding image according to text semantic.
The second way is the example retrieval, is input as a width of cloth chest x-ray image.Whether the texture semanteme of at first judging this image extracts, if extract, will call extraction of semantics and describing module and this image is carried out extraction of semantics and be stored in the image DICOM file; If extract, will directly from image file, obtain the semantic description document, retrieval obtains the image similar to input picture as input information with it.
The third mode is comprehensive text retrieval and example retrieval.
The present invention is other search method relatively, has following advantage:
The first, reduce hand labor and condition and rely on.This method only needs manually disposable semantic description (this process also can be extracted and be put in order and be generated) to be carried out in the standard picture storehouse of fewer amount from the diagnosis report of standard, and all the other processes are finished automatically by computing machine.Except the input of the Query Information of necessity, avoided other man-machine interaction process, and in whole process, do not needed necessity support of diagnosis report.
The second, retrieval rate is fast.With respect to CBIR, this method adopts text semantic information when retrieval, improved image retrieval speed.With respect to traditional text-based image retrieval, only need get final product disposable execution extraction of semantics of every width of cloth X line image and description on the backstage, retrieval rate approaches text-based image retrieval speed.
The 3rd, the recall precision height.Multiple image retrieval mode is provided, and search condition can be fuzzyyer, also can directly import piece image and carry out the example retrieval.Add positional information and characteristic information in the semanteme, result for retrieval is more accurate.
Description of drawings:
Fig. 1 is module issue synoptic diagram of the present invention;
Fig. 2 is extraction of semantics of the present invention and description process flow diagram;
Fig. 3 a1 is the non-standard rabat of the present invention;
Fig. 3 b1 is the non-standard rabat of the present invention;
Fig. 3 a is the Gray Projection figure of the non-standard rabat of the present invention;
Fig. 3 b is the Gray Projection figure of the non-standard rabat of the present invention;
Fig. 4 a for lung field of the present invention cut apart with area dividing after the frisket that obtains;
Fig. 4 b for lung field of the present invention cut apart with area dividing after the frisket that obtains;
Fig. 4 c for lung field of the present invention cut apart with area dividing after the frisket that obtains;
Fig. 5 example retrieval flow of the present invention figure.
Embodiment:
In conjunction with the accompanying drawings, the present invention has two main modular, is respectively extraction of semantics and describing module, image retrieval module, can be distributed in the medical image related system, and its concrete published method as shown in Figure 1.At first on original database basis of invariable, create a chest x-ray standard picture knowledge base and an image, semantic information bank.Wherein the standard picture knowledge base comprises a certain amount of standard picture semantic information, standard picture is different and different according to categorical measure, about about 100 width of cloth of wherein every class image, every width of cloth image comprises position, symptom, degree etc. by the semantic description that the expert provides standard.The image, semantic information bank is used for storing the semantic information that obtains by extraction of semantics and describing module.Then extraction of semantics and describing module are published to image acquisition end and user side, do not look like to carry out extraction of semantics and description to extracting grapheme, image retrieval module is published to user side, for the user provides the diversification image retrieval.
Extraction of semantics and description flow process comprise 12 steps as shown in Figure 2.Wherein, step 1 to step 3 pair chest x-ray image (DICOM file layout) carries out standardization processing; Step 4 to step 6 is partitioned into the lung field zone according to moving shape model (ASM) thought; Step 7 and step 8 are carried out area dividing by the fixed position to lung field; Step 9 and step 10 are to each extracted region bottom textural characteristics; Step 11 and step 12 pair image carries out Semantic mapping and description.
Step 1: compression gray level
The gray level of chest x-ray image is generally 4096 grades, shows with reducing calculated amount 256 grades of gray level boil down tos for convenience.At first obtain maximum gradation value and minimum gradation value in the original image, obtain the gray-scale value Gcp of each pixel among the image Pcp after the compression then, formula is as follows:
G
cp＝(G
scr-G
min)*255/(G
max-G
min)
Wherein Gscr represents the grey scale pixel value of each point among the original image Pscr, and Gmax and Gmin represent maximum and the minimum gradation value in the original image respectively.
Step 2: removing irrelevant part such as head, arm and standard is comparable size
Image Pcp is carried out vertical projection and horizontal projection, as shown in Figure 3, the horizontal coordinate of horizontal ordinate presentation video pixel among Fig. 3 a wherein, the summation of all pixel values of the horizontal coordinate of ordinate presentation video correspondence; The vertical coordinate of horizontal ordinate correspondence image pixel among Fig. 3 b, the summation of all pixel values of the vertical coordinate of ordinate presentation video correspondence.At first determine left and right sides standard lines, in vertical projection, the value that multiply by coefficient (0.8) with the projection average is made a straight line that is parallel to X-axis, with about the outside, two peaks for the first time the horizontal ordinate of two crossing points as the left and right sides standard lines Xleft and the Xright of image.Determine standard lines up and down then, seek the point that satisfies following two conditions in horizontal projection: the X coordinate figure of (1) this point is less than 1/2 of picture altitude, and the Y coordinate figure is less than 0.5 times of the projection average; (2) this point is for trough and Y coordinate figure minimum, with the Yup coordinate of this point last standard lines as image, if can not find, then is the image coboundary.The value of following standard lines Ydown is that above standard lines is a radix, adds that the spacing of left and right sides standard lines multiply by a coefficient a, that is:
Y
down＝Y
up+a(X
right-X
left)
Such four standard liness have all found, and after it is cut out, carry out convergent-divergent according to the size of java standard library and generate standard picture Ps.
Step 3: remove noise
More accurate for latter feature is extracted, image Ps is carried out medium filtering, window width is 7x7, and the pixel in the window is carried out quicksort, and getting its intermediate value is filtered image pixel value, and wherein edge pixel does not deal with.
Step 4: the shape of setting up the lung field zone
The standard that standard is cut apart every width of cloth image in the storehouse is cut apart point set and is regarded a vector as
X＝[x
0，y
0，x
1，y
1，...x
n-1，y
n-1]
T
Wherein n is the sum of cut-point, and x and y represent the horizontal ordinate of cut-point respectively.According to the thought of pivot analysis, M width of cloth image in the java standard library is set up following shape:
Wherein P is the matrix that the proper vector of the covariance matrix of M X vector is formed, and b is the parameter that the different images to input obtains by search iteration, and Y changes, and has represented active shape.
Step 5: set up the local appearance model
The local appearance model is being represented the local feature of each cut-point, and each point is all needed to set up this model.The Grad of 2k+1 of normal direction point of getting this place profile is as the local feature vectors of this point, and normal direction is determined by its adjacent 2.The mean value of local feature vectors of obtaining M image is as the local appearance model.According to thought from coarse to fine, set up many levels of precision local appearance model.
Step 6: search for actual cut-point
The input piece image according to thought from coarse to fine, is set up pyramid model to the hunting zone of each point.Adopting the local appearance model under different accuracy is feature, and the city distance obtains actual cut-point for matching formula search.
Step 7: determine the area dividing point
Cut-point to each lung field is numbered from 0 to n-1, and n is the sum of cut-point, pre-determines the numbering of 8 points on the lung field profile shown in Fig. 4 a according to the feature of each point, and inner four points are got a little and obtained by carrying out equal proportion between 2 of horizontal direction.
Step 8: generate the area dividing frisket
At first generate the unidimensional gray level image Pmask of a width of cloth and image Ps, and background colour is filled to black, promptly gray-scale value is 0.To each lung field zone, from top to down each is filled in image Pmask by the zonule that cut-point and division points surround successively then, filler pixels value from 30 to 210 does not wait, and is spaced apart 10, and filling algorithm adopts the sweep trace filling algorithm.Filling finishes and the filling of mistake might occur, as Fig. 4 b, need revise it.Modification method is, tries to achieve the area-encasing rectangle in each lung field zone earlier, again the zone that surrounds between this rectangle and the cut-point is filled to background colour, and revised frisket is shown in Fig. 4 c.So far, the area dividing frisket generates, and is used for follow-up processing procedure.
Step 9: extract each regional grey scale difference matrix
Obtain in each zone of image Ps that certain deviation amount δ=(r, 2 gray scale difference θ) is the probability P of k
δ(k) (k=0,1 ..., 255), wherein, 2 determination methods in the same area are among the Ps, 2 gray-scale value with its same coordinate in Pmask is identical.
Step 10: extract statistical characteristic value
Calculate four eigenwert composition characteristic vectors, be respectively contrast (Contrast), energy (angle second moment ASM), entropy (Entropy) and gray scale difference average (Mean), formula is as follows:
At last it is carried out normalized.
Step 11: java standard library is carried out artificial semantic description and classification
At first adopt standard semantic to be described by doctor or expert to it to the image in the existing standard storehouse, comprise information such as illness title, degree, position, this process will have the auxiliary doctor of special-purpose software or the expert is described.Then semanteme is described by it in the zone of image and classify,, then it is classified as in the normal category if do not have semantic description in certain zone.At last the image in all java standard libraries is cut apart and extracted proper vector.This process is only carried out once, but can constantly increase classification thereafter.
Step 12: input picture is classified and semantic description
Each zone to image to be described, adopts the k-nearest neighbour method to classify, and wherein the k value is (C+1), and C is the classification sum in the java standard library.In order to improve the travelling speed of k-nearest neighbour method, before searching for, earlier the texture feature vector collection of java standard library is created K dimension search tree (K-D tree), statistics searches the classification under the most contiguous C+ 1 width of cloth image, with image division to be described, in classification with maximum images, if more than one of such classification then is divided in a plurality of classifications.Adopt the semantic described of place classification, directly be stored in the DICOM file with the XML form.
Retrieving portion of the present invention is divided into three kinds of modes, specifically implements as follows:
(1) is input as text message
The manner input information is made up of two parts, and a part is that system provides scope, is selected by the user, as positional information etc.; Another part is directly to import search key by the user.This two parts information is converted to normative text information by semantic parser, then it is committed to conventional text search engine and retrieves.
(2) be input as image information
The manner input information is a width of cloth chest x-ray image, and its retrieval flow as shown in Figure 5.Judge at first whether example image had been extracted semantic, promptly check and whether have semantic description in the image file, if there is no, then to carry out extraction of semantics of the present invention and describe part input picture, the extraction of semantics of input picture is come out, be converted to normative text information by semantic parser, then it be committed to conventional text search engine and retrieve.
(3) be input as the integrated information of text and image
The manner input information had both comprised text message, comprised image information again, and need the user to import or select this two category information logical relation (with or etc.) and constraint condition.At first search condition is extracted its semantic description respectively, according to the logical relation of user input and constraint condition the semantic description of this two classes input information is merged then and form new semantic description, the text search engine that at last it is converted to normative text information by semantic parser and is committed to routine is retrieved.
Extraction of semantics of the present invention is with to describe the part travelling speed fast, extraction of semantics efficient height, and the semantic feature of every width of cloth image all only carries out once, do not do repetitive operation.Retrieving portion of the present invention with respect to the tradition retrieval, provides more convenience to the human-computer interaction interface that the user provides a close friend.
Be with being appreciated that, more than about specific descriptions of the present invention, only be used to the present invention is described and be not to be subject to the described technical scheme of the embodiment of the invention, those of ordinary skill in the art is to be understood that, still can make amendment or be equal to replacement the present invention, to reach identical technique effect; Use needs as long as satisfy, all within protection scope of the present invention.
Claims (10)
1. a chest x-ray image search method is characterized in that, implements successively as follows:
(1) input chest x-ray image;
(2) judge whether there is semantic description in the image file, if then in described image file, extract semantic description; If not, then input picture is carried out extraction of semantics and description;
(3) semanteme resolves to received text information;
(4) carry out text search;
(5) obtain the similar image result set.
2. chest x-ray image search method according to claim 1 is characterized in that: described extraction of semantics and description in turn include the following steps:
(a) image standardization;
(b) lung field is cut apart;
(c) carry out area dividing according to anatomical location;
(d) texture feature extraction;
(e) image to be described, is carried out Semantic mapping and description.
3. chest x-ray image search method according to claim 2 is characterized in that: described image standardization in turn includes the following steps:
(a) compression gray level;
(b) removing irrelevant part such as head, arm and standard is comparable size;
(c) remove noise.
4. chest x-ray image search method according to claim 3 is characterized in that: described lung field is cut apart and is in turn included the following steps:
(a) set up the shape in lung field zone;
(b) set up the local appearance model;
(c) search for actual cut-point.
5. chest x-ray image search method according to claim 4 is characterized in that: described area dividing in turn includes the following steps:
(a) determine the area dividing point;
(b) generate the area dividing frisket.
6. chest x-ray image search method according to claim 5 is characterized in that: described texture feature extraction in turn includes the following steps:
(a) extract each regional grey scale difference matrix;
(b) extract statistical characteristic value.
7. chest x-ray image search method according to claim 6 is characterized in that: described Semantic mapping and description in turn include the following steps:
(a) java standard library is carried out artificial semantic description and classification;
(b) input picture is classified and semantic description.
8. according to the arbitrary described chest x-ray image search method of claim 1～7, it is characterized in that: add text message when importing the chest x-ray image in the described step (1) again.
9. chest x-ray image search method according to claim 7 is characterized in that: the standard of in the described shape of setting up the lung field zone standard being cut apart every width of cloth image in the storehouse is cut apart point set and is regarded a vector as
X＝[x
0，y
0，x
1，y
1，...x
n-1，y
n-1]
T
Wherein n is the sum of cut-point, and x and y represent the horizontal ordinate of cut-point respectively
M width of cloth image in the java standard library is set up following shape:
Wherein P is the matrix that the proper vector of the covariance matrix of M X vector is formed, and b is the parameter that the different images to input obtains by search iteration, and Y represents active shape.
10. chest x-ray image search method according to claim 9, it is characterized in that: described foundation in the local appearance model, the Grad of 2k+1 of normal direction point of getting each cut-point place profile is as the local feature vectors of this point, described normal direction is determined by its adjacent 2, obtains the mean value of the local feature vectors of M image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101882450A CN102053963B (en) | 2009-10-28 | 2009-10-28 | Retrieval method of chest X-ray image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101882450A CN102053963B (en) | 2009-10-28 | 2009-10-28 | Retrieval method of chest X-ray image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102053963A true CN102053963A (en) | 2011-05-11 |
| CN102053963B CN102053963B (en) | 2012-08-08 |
Family
ID=43958302
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009101882450A Expired - Fee Related CN102053963B (en) | 2009-10-28 | 2009-10-28 | Retrieval method of chest X-ray image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102053963B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103200861A (en) * | 2011-11-04 | 2013-07-10 | 松下电器产业株式会社 | Similar case retrieval device and similar case retrieval method |
| CN103246888A (en) * | 2012-02-03 | 2013-08-14 | 通用电气公司 | System and method for diagnosing lung disease by computer |
| CN104424629A (en) * | 2013-08-19 | 2015-03-18 | 深圳先进技术研究院 | X-ray chest radiography lung segmentation method and device |
| CN107145910A (en) * | 2017-05-08 | 2017-09-08 | 京东方科技集团股份有限公司 | Performance generation system, its training method and the performance generation method of medical image |
| CN107357898A (en) * | 2017-07-13 | 2017-11-17 | 上海联影医疗科技有限公司 | The method, apparatus and electric terminal of a kind of information search |
| CN109716445A (en) * | 2017-03-10 | 2019-05-03 | 富士通株式会社 | Similar cases image retrieval program, similar cases image retrieving apparatus and similar cases image search method |
| CN109937433A (en) * | 2016-11-08 | 2019-06-25 | 皇家飞利浦有限公司 | The device of detection for the opaque region in radioscopic image |
| WO2019224650A1 (en) * | 2018-05-24 | 2019-11-28 | International Business Machines Corporation | Generating a textual description of an image using domain-independent anomaly analysis |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7739221B2 (en) * | 2006-06-28 | 2010-06-15 | Microsoft Corporation | Visual and multi-dimensional search |
| CN101441658A (en) * | 2008-10-22 | 2009-05-27 | 中国科学院上海技术物理研究所 | Search method and system facing to radiation image in PACS database based on content |
-
2009
- 2009-10-28 CN CN2009101882450A patent/CN102053963B/en not_active Expired - Fee Related
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103200861B (en) * | 2011-11-04 | 2015-10-14 | 松下电器产业株式会社 | Similar cases indexing unit and similar cases search method |
| CN103200861A (en) * | 2011-11-04 | 2013-07-10 | 松下电器产业株式会社 | Similar case retrieval device and similar case retrieval method |
| CN103246888A (en) * | 2012-02-03 | 2013-08-14 | 通用电气公司 | System and method for diagnosing lung disease by computer |
| CN104424629B (en) * | 2013-08-19 | 2018-01-26 | 深圳先进技术研究院 | A kind of x-ray chest radiograph lung segmentation method and apparatus |
| CN104424629A (en) * | 2013-08-19 | 2015-03-18 | 深圳先进技术研究院 | X-ray chest radiography lung segmentation method and device |
| CN109937433A (en) * | 2016-11-08 | 2019-06-25 | 皇家飞利浦有限公司 | The device of detection for the opaque region in radioscopic image |
| CN109937433B (en) * | 2016-11-08 | 2024-01-16 | 皇家飞利浦有限公司 | Device for detecting opaque regions in an X-ray image |
| CN109716445A (en) * | 2017-03-10 | 2019-05-03 | 富士通株式会社 | Similar cases image retrieval program, similar cases image retrieving apparatus and similar cases image search method |
| CN107145910A (en) * | 2017-05-08 | 2017-09-08 | 京东方科技集团股份有限公司 | Performance generation system, its training method and the performance generation method of medical image |
| US11024066B2 (en) | 2017-05-08 | 2021-06-01 | Boe Technology Group Co., Ltd. | Presentation generating system for medical images, training method thereof and presentation generating method |
| CN107357898A (en) * | 2017-07-13 | 2017-11-17 | 上海联影医疗科技有限公司 | The method, apparatus and electric terminal of a kind of information search |
| WO2019224650A1 (en) * | 2018-05-24 | 2019-11-28 | International Business Machines Corporation | Generating a textual description of an image using domain-independent anomaly analysis |
| US10685172B2 (en) | 2018-05-24 | 2020-06-16 | International Business Machines Corporation | Generating a textual description of an image using domain-independent anomaly analysis |
| CN112136155A (en) * | 2018-05-24 | 2020-12-25 | 国际商业机器公司 | Generating textual descriptions of images using domain-independent anomaly analysis |
| GB2588547A (en) * | 2018-05-24 | 2021-04-28 | Ibm | Generating A Textual Description Of An Image Using Domain-Independent Anomaly Analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102053963B (en) | 2012-08-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102053963B (en) | Retrieval method of chest X-ray image | |
| Hassanien | Fuzzy rough sets hybrid scheme for breast cancer detection | |
| ElAlami | A novel image retrieval model based on the most relevant features | |
| dos Santos et al. | Interactive multiscale classification of high-resolution remote sensing images | |
| CN104881680A (en) | Alzheimer's disease and mild cognitive impairment identification method based on two-dimension features and three-dimension features | |
| CN103530882B (en) | Improved image segmentation method based on picture and color texture features | |
| CN104217213B (en) | A kind of medical image multistage sorting technique based on symmetric theory | |
| Guedri et al. | Indexing and images retrieval by content | |
| CN101211344A (en) | Text message ergodic rapid four-dimensional visualization method | |
| Puig et al. | Application-independent feature selection for texture classification | |
| CN105740378A (en) | Digital pathology whole slice image retrieval method | |
| Kak et al. | Content-based image retrieval from large medical databases | |
| CN111797267A (en) | Medical image retrieval method and system, electronic device and storage medium | |
| Xu et al. | Identification of benign and malignant lung nodules in CT images based on ensemble learning method | |
| CN101609451A (en) | Sort out the relevant feedback of identification based on the medical image fuzzy region feature and estimate method | |
| Marinai et al. | Tree clustering for layout-based document image retrieval | |
| Fan et al. | Pulmonary nodule detection using improved faster R-CNN and 3D Resnet | |
| Stanley et al. | An image feature-based approach to automatically find images for application to clinical decision support | |
| Manikandan et al. | A mathematical approach for feature selection and image retrieval of ultra sound kidney image databases | |
| Huang et al. | Automatic image annotation using multi-object identification | |
| CN106570124A (en) | Remote sensing image semantic retrieval method and remote sensing image semantic retrieval system based on object level association rule | |
| Scott et al. | EBS kd tree: An entropy balanced statistical kd tree for image databases with ground-truth labels | |
| Xie | Research on application of ultrasound medical imaging technology in big data mining of regional medical imaging | |
| Long et al. | A prototype content-based image retrieval system for spine X-rays | |
| Jaime-Castillo et al. | A system to perform cbir on x-ray images using soft computing techniques |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120808 Termination date: 20141028 |
|
| EXPY | Termination of patent right or utility model |