CN100388314C

CN100388314C - System and method for locating compact objects in images

Info

Publication number: CN100388314C
Application number: CNB2004800230065A
Authority: CN
Inventors: M·沃尔夫
Original assignee: Siemens Medical Solutions USA Inc
Current assignee: Siemens Medical Solutions USA Inc
Priority date: 2003-08-14
Filing date: 2004-08-16
Publication date: 2008-05-14
Anticipated expiration: 2024-08-16
Also published as: CN1836253A

Abstract

An image processing system for recognizing image features in three dimensional images, which can be medical images, uses a mask generator (78) for generating masks that are used by a candidate searcher (80) to search for candidate images in the three dimensional image. The candidate searcher (80) applies the mask to a section of a foreground region of the image to determine the presence of a structure/object by counting the number of intersections between the mask and the section of the foreground region.

Description

The system and method that is used for the positioning image compact objects

Cross reference to related application

The application requires in the U.S. Provisional Application No.60/495 of proposition on August 14th, 2003,306 interests, and the copy of this application is in this combination as a reference.

Technical field

The present invention relates to image processing field, particularly detection of three dimensional target in binary picture or marking image.

Background technology

The development of image processing field has surmounted catches image and to the basic function of the image applications base image conversion of catching.Usually compare with three-dimensional (3-D) treatment of picture, the processing complicacy of two dimensional image is less relatively.Suppose that before 3-D shape of identification, the detection of the surface in the 3-D image, shape and size need be handled quantity of parameters, then the identification of 3-D image is the process of a complexity.

Detection of three dimensional in stereo-picture (3-D) target is the task of a complexity.Yet when recognition speed was most important, the identification problem of 3-D was just complicated more.For example, a medical imaging devices in a first aid ward or busy hospital requires the image that obtains is done very to analyze fast, so that physician or surgeon on the basis of the image of having discerned, can produce medical diagnosis and decision apace.

According to the skeleton view that piece image is handled, a desirable 3-D target identification system needs each image-element of full test, just each volume elements (volume element) of image.Yet even all existing volume elements are present in the medium sized relatively image of a width of cloth, this will also be an infeasible technology.Therefore, exist for a system and the needs of method fast, promptly the standard by using an optimization is to select complete tested a small amount of volume elements in treatment step subsequently.

Medical Image Processing is important type during Flame Image Process is used.In medical image analysis, emphasis partly is the medically significant destination object of identification, as knurl, and tumour etc.Typically, the target that is detected has relatively little volume.For example, surveying the lung tubercle automatically in CT (the computable x-ray tomography art scanning) pulmonary scanning is the process of a difficulty.Yet the early diagnosis of lung tubercle helps to detect and treatment lung cancer.Computer assisted nodule detection system should be fast, have high sensitive and high singularity, and this is desirable.

Existing method is used the geometric properties of image object usually, and for example spherical and curvature is discerned tubercle.Yet calculating such feature is a thing of wasting computing time.Typically, to cause the test fully of each volume elements of 3-D medical image be a kind of inefficient technology to the time restriction of medical diagnosis process.

Summary of the invention

Therefore, need be for Medical Image Processing by the system applies fast algorithm with structure special in the detected image and/or target.

The invention provides-kind of an image processing system is used at the 3-D view recognition objective, and this system comprises: the mask generator is used to produce at least one mask; And candidate searcher, be used for searching at least one candidate target similar at least one intended target of at least one width of cloth 3-D view, the presumptive area of the foreground image part of this candidate searcher by mask being covered described 3-D view and this mask is applied to this presumptive area is with the quantity of determining to intersect between mask volume elements and the described presumptive area.

The present invention also provides an image processing system to be used to discern the existence of medical science target, and this system comprises: image capture device is used to catch the 3-D view of an organic part alive; Image processing system, being used for three-dimensional image segmentation is a width of cloth foreground image and a width of cloth background image, image processing system is applied to the zone of a selection of foreground image with at least one mask, with by crossing between the zone of mask and selection counted to determine existing of at least one medical science target.

The present invention also provides a kind of method that is used for detecting at 3-D view target in addition, and this method may further comprise the steps: produce at least one three dimensional mask; And use three dimensional mask in 3-D view, to search the candidate; It is characterized in that the presumptive area of the foreground image part by mask being covered described 3-D view and this mask is applied to this presumptive area is with the quantity of determining to intersect between mask volume elements and the described presumptive area.

One is used in the system of 3-D view detection architecture/target three-dimensional image being operated.Image is divided into prospect and background area, and wherein foreground area is the zone that we are concerned about.A mask generator produces three dimensional mask.A candidate searcher also is candidate's generator, the search possible structure/object similar to ferret out in 3-D view.This candidate searcher is used mask to the selection of foreground area, to determine existing by searched structure/object in the image.Each structure/object that is identified is called a candidate.

(intersection) technology that use to intersect is to determine existing of in 3-D view candidate.Use the part of mask to foreground area.The quantity that intersects of counting between this part of mask and foreground area.If zero is arranged or have only one to intersect existence, then deducibility goes out a candidate's existence.This represents the fully masked covering of tubercle or is attached on a wall or the vascular.Yet if intersect more than one, this image section is not a candidate.

In one embodiment, 3-D view is the image of catching from the organism of people or other work.Here, the candidate be one with searched medical structure, lung tubercle for example, knurl etc.Mask is a cuboidal mask typically, and its shape and size are used to cover a single tubercle (or other structure/object).Because only test the volume elements of foreground image areas, rather than the volume elements of entire image, possibility become so detect such tubercle rapidly.In addition, compare with the character of surface of all volume elements in the test pattern, what intersect determines it is a process that calculating strength is less.

Description of drawings

The preferred embodiments of the present invention are illustrated in conjunction with the accompanying drawings, wherein:

Fig. 1 is the synoptic diagram that the lung tubercle with example mask is shown;

Fig. 2 is the process flow diagram that the image processing step that is used to produce the candidate is shown;

Fig. 3 illustrates to handle the process flow diagram that comprises crossing image;

Fig. 4 illustrates the process flow diagram that intersects detection process;

Fig. 5 A shows a diamond shaped mask;

Fig. 5 B shows a square mask;

Fig. 5 C shows a undersized rhombus (parallelogram) mask;

Fig. 5 D shows a large-sized rhombus (parallelogram) mask;

Fig. 6 shows a demonstration system that is used for seeking at 3-D view the candidate; And

Fig. 7 shows a computer system that is used to carry out an existing embodiment who invents.

Embodiment

The preferred embodiments of the present invention will be illustrated in conjunction with the accompanying drawings.

Fig. 1 shows the illustrative figure of the lung tubercle with example mask.At least one embodiment of the present invention can be used for surveying solid-state lung tubercle.Yet those skilled in the art will know that lung's tubercle just is used for an illustrational representative instance, and any structure that has known geometries in stereo-picture can both be detected.For example, can detect knurl, tumour etc.Further, though accompanying drawing is two-dimentional for the purpose of clear displaying, the objective for implementation of the embodiment of the invention is a 3-D view.

The typical image that medical imaging devices captured is a stereo-picture.The xsect 10 of demonstration stereo-picture is the part of pulmonary scanning.Three

illustrative tubercles

12,14 and 16 illustrate with corresponding mask 18-22.From the input picture typical case who wherein obtains xsect 10 is the stereo-picture that a width of cloth is cut apart, and this image can be classified as two zones.First zone is the zone of being concerned about (prospect), and another is unconcerned zone (background).

Next input picture is described.All background regions image elements have a unique value.Foreground region image elements can have a common value, and the volume in this situation is called " two-symbol volume "., each zone in a foreground area or structure have a unique value, and then the volume in this situation is called " mark " or " volume that is labeled ".Further processing at this image is as described below.

Each volume elements that input picture is converted into wherein is marked as prospect or is the image of background.This width of cloth image then is used as the input of further Flame Image Process.All pixels that belong to background are made as 0 value in the image, and all foreground pixels are made as nonzero value, with the prospect and the background area of differentiate between images.

Can locate tubercle by handling this image.Selection has the mask of clear and definite size and Orientation or the existence that the benchmark shape is determined the geometric properties of needs.Here, at least one embodiment, use the existing illustration of lung's tubercle.Therefore use a benchmark shape or mask that is shaped as bounding box.Also can use the mask of other shape, other illustrations are as described below.Those skilled in the art will appreciate that bounding box is an example that is fit to the mask of lung's tubercle geometric configuration.

In two-dimensional space, mask is a zonule, and all surface point of this zonule is made as 1, and all other non-surface point is made as 0.In 3-dimension, mask is a small size, and this small size all surface point is made as 1, and all other non-surface point is made as 0.Then, this mask moves on entire image.The central point of mask is placed on each pixel that belongs to prospect.The quantity of the mask pixels that intersect counting and foreground point.Can determine how many edges (3-D face) of tubercle in image and reference mask intersect thereafter.

What intersect determines and can number of ways realize.For example, the input picture that covers a definition foreground image when mask is when partly going up, and the special pixel (having nonzero value) that covers the mask on the input picture is compared with the respective pixel in input picture.Carried out on the input picture cover and the processing of mobile mask after, when any nonzero value pixel of mask during, thereby can determine the existence of intersecting corresponding to a nonzero value pixel in input picture.

In this example, think and perhaps be attached to illustrative tubercle or independent on a vascular or the lung wall.Therefore, can not occur intersecting for independent tubercle.For example, tubercle 16 is independent tubercles, and is included in fully in the mask 22, does not therefore intersect.Therefore on the contrary, some tubercle can be attached on a vascular or the wall and only have one to intersect.Yet,, repeat the step that mask covers for different size and Orientations so if there is not or has only a crossing standard not to be satisfied.The result that above-mentioned searching tubercle is handled generates a possibility tuberal area domain list.

In another approach, this technology can be used for reducing the candidate's who produces by algorithms of different quantity.Handle stereo-picture and the tabulation of the position of discovery before that a width of cloth is cut apart together.Mask can directly be placed on the position of previous discovery now.After definite intersecting, can delete list of locations when needed.Below explain above-mentioned processing, it is similar to the operation of searching a width of cloth 3-D view or the previous list of locations of finding of search.

Fig. 2 is the process flow diagram that is used for the image processing step of structure/object detection.Step shown in flow process Figure 24 starts from step 26, obtains stereo-picture herein from vision facilities.This stereo-picture is converted into the image of cutting apart in step 28.In

step

30 and 32 produce the mask of all size and Orientations thereafter.The mask of a plurality of size and Orientations can be used to detect different tubercles and other structure/object in image.In step 34, use mask in image, to seek the candidate of tubercle (or other structure/object).Thereafter, at candidate list of step 36 output.

Fig. 3 handles the process flow diagram that comprises crossing image.Flow process Figure 38 starts from step 40, and in step 40, it is all processed up to all volume elements in the input picture volume to begin an iterative process herein.Carries out image processing on foreground pixel only.In step 42,, then calculate and intersect in step 44 if pixel or volume elements are the prospect type.Whether determine to intersect more than one or be less than one (as zero or one) in step 46.

Zero meet representation mutually tubercle or complete masked restriction of structure/object, and meet representation mutually tubercle or a structure/object are attached on a wall or the vascular.For situation about intersecting more than, as arriving shown in the branch of step 50, this iterative process will be recycled and reused for other volume elements.This process turns back to step 40 in step 50 and carries out iteration, all handles up to the volume elements of all inputs to be over.Be over if all volume elements are all handled, then return a candidate list that is used for tubercle or structure/object in step 52.

Fig. 4 is the process flow diagram of intersection detection process.Produce mask in step 54.Show two exemplary masks in step 54.Mask is cancellate structure (when with a two-dimensional observation), thereby the some of them pixel has null value and some pixels have nonzero value to form mask.In step 56, explain the 3-D characteristic of mask.A 3-D mask will have 6 limits that form its 6 faces.Variable dx, dy and dz represent the distance between mask point and the mask center.

Use an iterative process to handle each volume elements of mask.If mask function m ask (dx, dy, dz) non-0, then carry out searching of (mask) limit in step 62.In step 64, if the value vol of the corresponding volume elements in the image (x+dx, y+dy are 0 z+dz), then this process by iteration to handle next volume elements.Yet, if the value vol of the corresponding volume elements in the image (x+dx, y+dy, z+dz) non-zero, in step 66 with intersect[f] be made as true.If also have remaining volume elements pending in the mask, then this iterative process begins repetition from step 58 once more.

Because above-mentioned technology is only counted and the intersecting of frame face, thereby improved the speed of identification tubercle or other structure/object.Volume elements in the frame need not be considered.For each frame (having fixed size and rotation), have only the coordinate on the face to be calculated once.During detecting step, the position of volume elements and these coordinates are compared.

Fig. 5 A-5D shows dissimilar masks.Fig. 5 A shows the mask of a rhombus (parallelogram); Fig. 5 B shows a square mask; Fig. 5 C shows a undersized rhombus (parallelogram) mask; Fig. 5 D shows rhombus (parallelogram) mask of a large-size.Means of mask dimensions and direction can change.In the following description, for task of explanation is considered the two-dimensional surface of mask, yet mask itself is to have a plurality of three dimensional mask as mentioned above.

At first, consider the change of mask aspect direction.Consideration is two typical directions of form with diamond shaped parallelogram and square.Diamond shaped mask (Fig. 5 A) only intersects in a place with tubercle (pointing out to reach the purpose of outstanding this tubercle with *), and the tubercle remainder drops in the mask.Yet a foursquare mask (Fig. 5 B) intersects in two places with tubercle.Therefore, the geometry of the direction of mask and tubercle will be determined to determine intersecting of the best.

The size of mask is another key element that deterministic process is intersected in influence.Fig. 5 C shows a small diamond shaped mask, and the tubercle in itself and the image has only one to intersect.Yet the diamond shaped mask of a large-size had more than one intersect.Therefore, select mask size an of the best and the efficient that the mask direction will determine to intersect deterministic process.

Fig. 6 shows a demonstration system that is used for seeking at 3-D view the candidate.In at least one embodiment, image processing system 70 comprises image capture device 72.Image segmentation module 74 produces the piece image of the prospect and the background that have a separation, wherein the zone of prospect for being concerned about.Image processing system 70 can be by an inner insertion equipment, and the part of an external unit or a software configuration realizes.Image processing system can be realized above-mentioned part or all of image processing algorithm.

Characteristics of image hunting system 76 comprises a mask generation module 78, and this module produces the mask of describing in detail above.The mask that candidate's search module 80 utilizes mask generation module 78 to be produced finds intersecting of split image that mask and image segmentation module 74 produced.Based on intersecting, candidate's search module 80 is determined and a pixel that structure/object is complementary will being searched.For further subsequent treatment, output module 82 is exported to module 84 with the candidate list of finding.For example, subsequent treatment can comprise the analysis completely to each candidate, perhaps on display show candidate for the diagnosis medical problem.

Fig. 6 is an exemplary electronic system, and it can be used in realizes at least one embodiment of the present invention.Can be understood that the present invention can various forms of hardware, software, firmware, special purpose processors, perhaps their combination realizes.In at least one embodiment, the present invention can form of software be embodied as the application program that clearly is included on the program storage device.This application program can to a machine upload and by execution, wherein this machine can calculate and have any suitable structure.

Next the explanation of the lung images of scanning is described.In the lung images of width of cloth scanning, carry out the search of lung's tubercle.Whole lung is a foreground area, and mask is applied to this lung areas.When a given mask and tubercle image crossing is zero, because tubercle is enclosed in the mask, so tubercle exists.Even there is one to intersect, because a tubercle may be attached on a wall or the vascular, the then also deducibility existence that goes out a tubercle.

By only searching the specific region of foreground area, the fast searching of the candidate image that the form with medical image feature (for example, lung's tubercle) is occurred becomes possibility.Other alternative ways that each volume elements in the image is all carried out test are time-consuming operations, and therefore only searching the specific region can be faster.Intersect by counting only and to search, improved the speed of searching, this is because search space has been dwindled, and calculates from the angle of calculating that to intersect be a fast relatively process.

With reference to figure 7, according to one embodiment of present invention, be used to realize that computer system 101 of the present invention can comprise CPU (central processing unit) (CPU) 102, storer 103 and I/O (I/O) interface 104 and other parts.Computer system 101 is passed through the various input equipments 106 of I/O interface display 105 of 104 couplings and for example mouse and keyboard usually.Auxiliary circuit can comprise as circuit such as cache memory, power supply, clock circuit and communication buss.Storer 103 can comprise random-access memory (ram), ROM (read-only memory) (ROM), disc driver, tape drive etc., perhaps their combination.The present invention can a routine 107 form realize that this routine stores and is carried out the signal that sends from signal source 108 to handle by CPU102 in storer 103.Similarly, computer system 101 is one by computer system, becomes the computer system that specific purpose is arranged when it carries out routine 107 of the present invention.

Computer platform 101 comprises operating system and micro-instruction code equally.But the part of various program described herein and function micro-instruction code, or the part of application program (or their combination), wherein micro-instruction code or application program are carried out by operating system.In addition, various other peripherals can be connected to computer platform, as an additional data storage device and a printing device.

Can further be understood that: because some the element system parts described and the realization of method step available software in conjunction with the accompanying drawings, so the actual connection between system unit (or treatment step) may be different, this depends on the mode to the present invention's programming.According to instruction of the present invention provided herein, those of ordinary skill in the related art can draw these or similar the invention process or structure.

Specifically illustrate and described the present invention though combine exemplary embodiment of the present invention, those skilled in the art may appreciate that, under the situation that does not depart from the spirit and scope of the present invention, can make various changes aspect form and the details.

Claims

1. an image processing system is used at the 3-D view recognition objective, and this system comprises:

The mask generator is used to produce at least one mask;

And

Candidate searcher, be used for searching at least one candidate target similar at least one intended target of at least one width of cloth 3-D view, the presumptive area of the foreground image part of this candidate searcher by mask being covered described 3-D view and this mask is applied to this presumptive area is with the quantity of determining to intersect between mask volume elements and the described presumptive area.

2. image processing system as claimed in claim 1 further comprises:

Image capture module is used to catch 3-D view;

The image segmentation module is used to cut apart the 3-D view of catching; And

Output module is used to export the candidate's that candidate searcher finds tabulation.

3. image processing system as claimed in claim 1, wherein the image segmentation module is separated into a width of cloth foreground image and a width of cloth background image with 3-D view.

4. image processing system as claimed in claim 3, wherein the candidate image search engine is operated on above-mentioned foreground image.

5. image processing system as claimed in claim 1, wherein 3-D view is a width of cloth stereo-picture.

6. image processing system as claimed in claim 1, wherein 3-D view is a width of cloth medical image, and predetermined target is the organic part of a work.

7. image processing system as claimed in claim 1, wherein

Mask comprises:

One width of cloth 3-D view comprises a plurality of volume elements and has a plurality of that wherein at least one volume elements has a mask value.

8. image processing system as claimed in claim 7, wherein mask is a cube.

9. image processing system as claimed in claim 7, wherein the face of mask is a parallelogram.

10. image processing system as claimed in claim 9, wherein, when the quantity that intersects was less than two, candidate searcher was determined at least one candidate's existence.

11. an image processing system is used to discern the existence of medical science target, this system comprises:

Image capture device is used to catch the 3-D view of an organic part alive;

Image processing system, being used for three-dimensional image segmentation is a width of cloth foreground image and a width of cloth background image, image processing system is applied to the zone of a selection of foreground image with at least one mask, with by crossing between the zone of mask and selection counted to determine existing of at least one medical science target.

12. system as claimed in claim 11, if wherein crossingly be less than two in addition to what the counting that intersects was determined to be counted, then image processing system determine the medical science target really real storage exist.

13. system as claimed in claim 11, wherein image processing system further comprises:

The mask generator is used to produce three dimensional mask.

14. system as claimed in claim 13, wherein, mask is a cube.

15. a method that is used for detecting at 3-D view target, this method may further comprise the steps:

Produce at least one three dimensional mask; And

Use three dimensional mask in 3-D view, to search the candidate;

It is characterized in that the presumptive area of the foreground image part by mask being covered described 3-D view and this mask is applied to this presumptive area is with the quantity of determining to intersect between mask volume elements and the described presumptive area.

16. method as claimed in claim 15 further comprises

Utilize image capture device to catch 3-D view; And

Three-dimensional image segmentation is become foreground area and background area.

17. method as claimed in claim 16, wherein, the step of search further comprises:

In the foreground area of 3-D view, search the candidate.