CN101346743A - Cross-time and cross-modality medical diagnosis - Google Patents

Abstract

A cross-time and cross-modality inspection method for medical image diagnosis. A first set of medical images of a subject is accessed wherein the first set is captured at a first time period by a first modality. A second set of medical images of the subject is accessed, wherein the second set is captured at a second time period by a second modality. The first and second sets are each comprised of a plurality of medical image. Image registration is performed by mapping the plurality of medical images of the first and second sets to predetermined spatial coordinates. A cross-time image mapping is performed of the first and second sets. Means are provided for interactive cross-time medical image analysis.

Description

Stride the medical diagnosis of time and cross-modality

Technical field

The present invention relates to be used for the digital video processing method of image analysing computer, particularly relate to the time of striding and the cross-modality inspection of heterogeneity tissue in medical image (for example, unusual and normal structure).

Background technology

Clinical use by for example computerized tomograph (CT) scanner realizes the digital imaging technology of medical application in nineteen seventies.From then on, the appearance of the use of x line imaging and digital machine and new iconography (for example, ultrasound wave and Magnetic resonance imaging (MRI)) has promoted the diagnosing image technology together.

Using to health care of digital medical imaging technique brought benefit.For example, employing digital medical imaging and image processing technique have made the angiographic procedures of checking brain, kidney, arm and shank and heart medium vessels be benefited.

By means of digitized video, computerize multidimensional (for example, room and time) image analysing computer becomes possibility.The multiplanar image analysis can in such as body part sequential images swept-volume, change automatically quantize (structure or function), foreign matter location, continue that diagnosis is reproduced and like that etc. the application in use.

In addition, different Medical Imagings has produced the image that bodily fuctions and structure different views are provided, and by means of correct medical image processing software and visualization tool, may strengthen diagnostic accuracy greatly.For example, X computer on line tomography (CT) and Magnetic resonance imaging (MRI) can be showed head configuration, and a spot of function information is provided.The scanning of positron emission tomography (PET) and single photon emission computed tomography (SPECT) shows the brain function aspects, and allow metabolism measure but structure describe poor.In addition, CT and MRI image have been described additional morphological feature.For example, the best results that bone and calcification are seen on the CT image, and MRI can distinguish soft tissue structure better.Many images of some body part are provided usually such as the form of MRI and CT.

As everyone knows, the information from different dimensional (room and time) or form acquisition often is that difference or additional character are arranged.In current clinical setting, this difference or additional image information are the clinical diagnosis environment, and also are the ingredients of widely applying in the plan of surgery and radiation therapy operative procedure and the assessment area.

For using difference or side information effectively, the doctor uses the visual alignment system that the image feature of different dimensional or different shape is superimposed.But, this type of coordination relative to each other of a plurality of images is difficulty especially, wait the medical worker who is well trained also to be difficult to keep and a series of medical images of correct understanding even resemble veteran radiographer, so that can formulate to the optimal methods of treatment of the current medical conditions of patient.

Another problem that the medical worker runs into is mass data and the many image that obtains from current medical image device.The image quantity of collecting in a standard scan can surpass 100 width of cloth, and often quantity is hundreds of width of cloth.Make the medical worker correct check every width of cloth image, this wants a large amount of time, and for many images that current medical skill provides, checks that meticulously all data need the plenty of time.

Therefore, need a kind of effective scheme, can use image processing/computer vision technique to detect automatically/diagnose the illness.

Laid-open U.S. Patents application 2004/0064037 (Smith) relates to the rule-based scheme that is used to handle medical image, and it is incorporated into this paper by reference.Its technology is used pre-designed regular, and the mode of medical image data will be classified or handle to these regular appointments.Designed regular can comprise the rule of selecting from applicable rule, revise/self-defined they to generate new regulation, brand-new rule perhaps is provided.But the technology of Smith is failed teaching and how to be checked patient's the time of striding, cross-modality medical image, so that can carry out reliable and diagnosis and operation plan accurately.

Laid-open U.S. Patents application 2003/0095147 (Daw) relates to the user interface with analysis state designator, and it is incorporated into this paper by reference.Daw has described a kind of medical image processing and visualization method.When image is carried out this type of data analysis, provide the analysis designator in the upper left corner that shows, thereby be provided at carry out or any computer analysis results of executed relevant data and state check indication.The function that the system of Daw provides can't detect and distinguish and contrast medium is had image forming material (tissue) the corresponding image district of different time response automatically.The applicant notices that this type of function is particularly useful to using the MRI radiography to strengthen the pernicious and optimum tumor of breast of diagnostic imaging.

United States Patent (USP) 6353803 (Degani) relates to a kind of equipment and method, and being used for monitoring wherein has liquid flow and have in the system space in time and the system of the feature of change, and this patent is incorporated into this paper by reference.Position preselected in the system comes under observation, so that collect data at two or more time points relevant with system event.These data are indications of certain systematic parameter, and this parameter is along with changing as time of the function of choosing at least two variablees relevant with the elimination behavior with system.

The research of this type of curve/parameter is clinical to be used for identification and to describe pernicious or optimum class tumour, but owing to common good sensitivity but frequent very poor specificity reason, success is changeable always (for example, referring to " MRI of mammary gland " (S.C.Rankin " MRI of the breast ", Br.J.Radiol 73, pp 806-818,2000)).

Though this type systematic can be realized success to a certain degree on its application-specific, the medical image analysis needs a kind of improved digital video processing method, overcomes the problems referred to above and solves above-mentioned material gain needs.

The invention provides a kind of image analysing computer that is used for, and specifically, be used for the method for checking at the time of striding and the cross-modality of medical image heterogeneity tissue (for example, unusual and normal structure).

Summary of the invention

The purpose of this invention is to provide a kind of being used in the time of striding of medical image heterogeneity tissue (for example, unusual and normal structure) and the method for cross-modality inspection.

Any purpose that provides is only enumerated by the illustrated examples mode, and this type of purpose can be the demonstration of the one or more embodiment of the present invention.Those skilled in the art can realize or understand by invention disclosed original attainable other required purpose and advantage.The present invention is by claims definition of enclosing.

The invention provides a kind of being used at the time of striding of medical image heterogeneity tissue (for example, unusual and normal structure) and the image processing/mode identification method of cross-modality inspection.Method may further comprise the steps: stride chronomedicine image order and alternatively tissue property is classified; Time cross-modality image mapping is striden in execution; And carry out interactive mode and stride time cross-modality medical imaging.

According to one aspect of the invention, provide a kind of and striden the method for time check heterogeneity tissue in proper order by striding the chronomedicine image.Method may further comprise the steps: gather a plurality of medical images (for example, injecting before the contrast medium and MRI image afterwards) and stride time series; Carry out a plurality of medical images and stride the inside registration of time series with respect to volume coordinate; Carry out a plurality of medical images and stride the mutual registration of time series with respect to volume coordinate; For a plurality of medical images of registration are striden the time series tissue of different nature of classifying; And for striding time check demonstration classification results.

According to a further aspect in the invention, provide a kind of being used to use the radiography Contrast-enhanced MRI image that increases by other physics or non-physical factor, detected and distinguished the method for abnormal structure automatically.Method may further comprise the steps: gather a plurality of MRI breast image collection; With respect to volume coordinate a plurality of MRI breast images of harmonizing; Distinguish a plurality of MRI breast image collection with reference to the MRI image set, to produce a plurality of difference image set; With the segmentation of a plurality of difference image set, have a plurality of MRI breast images of segmentation intensity pixel with generation; Use the intensity pixel that dynamic system recognizes segmentation, to produce a plurality of dynamic system parameter; And will be categorized as inhomogeneity by a plurality of systematic parameters that other physics or non-physical factor increase.

Still have on the other hand according to the present invention, a kind of method that is used for automatic materials classification is provided.Method may further comprise the steps: a plurality of image set of acquisition target in chronological order; With respect to volume coordinate a plurality of image set of harmonizing; Distinguish a plurality of image set with reference to image set, to produce a plurality of difference image set; With the segmentation of a plurality of difference image set, have a plurality of images of segmentation intensity pixel with generation; Use the segmentation intensity pixel that dynamic system recognizes a plurality of images, to produce a plurality of dynamic system parameter; And a plurality of systematic parameters are categorized into different classes.

According to a further aspect in the invention, provide a kind of method of using radiography Contrast-enhanced MRI image to detect abnormal structure.Method may further comprise the steps: gather a plurality of MRI breast image collection in chronological order; With respect to volume coordinate a plurality of MRI breast image collection of harmonizing; Distinguish a plurality of MRI breast image collection with reference to the MRI image set, to produce a plurality of difference image set; With the segmentation of a plurality of difference image set, have a plurality of MRI breast image collection of segmentation intensity pixel with generation; Use the segmentation intensity pixel that dynamic system recognizes a plurality of MRI breast images, to produce a plurality of dynamic system parameter; And a plurality of systematic parameters are categorized into different classes, to detect abnormal structure.

Description of drawings

As shown in drawings, from the more certain illustrated of the following embodiment of the invention, will understand above and other objects of the present invention, characteristic and advantage.Key element among the figure needn't be drawn each other in proportion.

Fig. 1 is the diagram that the dynamic arthrography absorbent properties (curve) of different breast tissues is shown.

Fig. 2 is an image processing system synoptic diagram useful in putting into practice the method according to this invention.

Fig. 3 illustrates the process flow diagram of striding time and cross-modality medical imaging method according to of the present invention.

Fig. 4 illustrates the process flow diagram of striding an embodiment of time tissue property's inspection method according to of the present invention.

Fig. 5 is the process flow diagram that illustrates according to Image registration method of the present invention.

Fig. 6 is the diagram that the Image registration notion is shown.

Fig. 7 illustrates two diagrams of striding the time image sequence.

Fig. 8 is the process flow diagram that illustrates according to an embodiment of automatic abnormal structure detection method of the present invention.

Fig. 9 is the diagram that the dynamic arthrography absorbent properties (curve) of pernicious and benign tumor tissue is shown.

Figure 10 is the synoptic diagram that function response of rank road and system identification notion are shown.

Figure 11 is the process flow diagram that illustrates according to system identification method of the present invention.

Figure 12 illustrates the diagram that the present invention strides time tissue property's inspection visable representation.

Figure 13 is illustrated in the diagram that tissue of different nature is arranged in the medical image.

Figure 14 A-14E illustrates a plurality of diagrams, shows the step of 3D image body of the present invention projection.

Figure 15 is the diagram that the tomography with pixel cloud is shown.

Figure 16 is the projection of a plurality of medical science 3D body.

Figure 17 illustrates the diagram of striding a time and an embodiment of cross-modality medical imaging according to of the present invention.

Embodiment

Be the detailed description with reference to the preferred embodiment of the present invention of accompanying drawing below, identical label identifies identical textural element in each figures of several figures in the accompanying drawing.

The medical imaging that the patient strides time, cross-modality can help to provide reliably and diagnosis and operation plan accurately.For example, X line mammography has limited specificity and susceptibility.It is reported, use X line mammography can leak the cancer of finding 5%-15%.MRI mammography as alternative formation method has higher susceptibility to the tumour greater than 3 millimeters.

As everyone knows, in a single day malignant breast tumor reaches certain stool and urine its oneself blood supply network that begins to grow; This is the mode that cancer can continued growth.In breast MRI scanning, the contrast preparation that is injected in the blood flow can provide relevant blood supply information to breast tissue; Contrast preparation is " demonstration " tumour by the blood vessel network that highlights tumour.Usually to scan several times: once be before contrast preparation injects, and at least once after injection.Compare before the radiography and the image behind the radiography, and highlight different zones.Even it should be understood that if the patient between twice scanning, have slightly mobile, the shape of image or the size can the distortion, thereby cause information dropout.

A kind of contrast preparation that is used for MRI is gadolinium or Gadodiamide, and provides contrast between normal structure and the abnormal structure in brain and health.Gadolinium looks that to resemble water equally limpid, and is inactive.After being injected into it in the blood vessel, gadolinium is built up in the abnormal structure that may influence health or head.Gadolinium can make these abnormal areas in MRI become bright (enhancing).This makes and is easy to see abnormal area.Subsequently, gadolinium is discharged from health by kidney.Gadolinium allows MRI to define abnormal structure with higher sharpness.After using gadolinium, it is more clear that tumour becomes.The definite size of tumour and position are very important in treatment plan and tracking.Gadolinium becomes obviously by making little tumour, is easy to see, also helps to search these tumours.

The dynamic arthrography Contrast-enhanced MRI is used for the breast cancer imaging; Particularly be used for those situations that to make a definite diagnosis based on x line mammography.MRI research can relate to just in time, and contrast preparation (be generally gadolinium and the spray sour grape amine) vein gather the set of T1 weighting MR body with about one minute temporal resolution before injects.The signal that exists contrast preparation to cause observing in search time in the imaging system strengthens.

The research of these signal time curves makes it possible to as shown in Figure 1 owing to different radiography absorbent properties is discerned different types of organizations.Be noted that cancerous tissue is general owing to the microvascular diffusion of the angiogenesis of " penetrating " shows height and fast absorption, and absorption normal and that the adipose tissue demonstration is little.Absorb (dynamically) curve used the pharmacokinetic model match with the physiology correlation parameterization that draws curve (referring to " using the dynamic Gd-DTPA enhancing in the saturated model quantitative analysis tumor of breast " (P.S.Tofts, B.Berkowitz, M.Schnall, " Quantitativeanalysis of dynamic Gd-DTPA enhancement in breast tumours using apermeability model ", Magn Reson Med 33, pp 564-568,1995)).

Fig. 2 is illustrated in and puts into practice image processing system useful in the method according to this invention 10.System 10 comprises the digital MRI image source 100, digitized video memory storage (as, compact disk driver) of MRI scanner for example or like that.Be provided to the image processor 102 of personal computer able to programme for example or such as the digitized video work of treatment station of Sun Sparc workstation from the digitized video of digital MRI image source 100.Image processor 102 can be connected to display 104 (as CRT monitor or other monitor), know operator interfaces such as input media such as keyboard 106 and mouse 108 or other.Image processor 102 also is connected to computer-readable recording medium 107.Image processor 102 will be handled digitized video and be transmitted into output unit 109.Output unit 109 can comprise hard copy printer, long-term image store device, to the connection of another processor, for example be connected to the image telecommunication installation of the Internet or like that.

In the following description, an embodiment will be described as a kind of method.But, in another embodiment, the present invention includes the computer program that is used for detecting digital MRI image abnormal structure according to described method.When description is of the present invention, will be appreciated that computer program of the present invention can be by any computer system utilization of knowing, personal computer as shown in Figure 2.But the computer system of other type can be used for carrying out computer program of the present invention.For example, carry out in the computing machine that method of the present invention can comprise in digital MRI machine or PACS (picture archive and transmission system).Therefore, computer system will will further not discussed herein in detail.

To recognize that also computer program of the present invention can utilize image processing algorithm and the process of knowing.Correspondingly, the explanation of statement will relate to specially form the inventive method part or with more direct those algorithms and the process of cooperating of the inventive method.Therefore, will understand, computer program product embodiments of the present invention can be implemented this paper and clearly not illustrate or describe, but to realizing useful algorithm and process.This type of algorithm and process are conventional algorithm and process, and in the common skill scope of this type of technical field.

The others of this type of algorithm and system and being used to produce and otherwise handle image that relates to or the hardware of cooperating with computer program of the present invention and/or software and clearly do not illustrate or describe at this paper, and can select from this type of algorithm, system, hardware, assembly and key element that technical field is known.

Fig. 3 summarizes the method that heterogeneity tissue in the medical image is striden the medical imaging of time cross-modality that illustrates.More particularly, Fig. 3 illustrates a process flow diagram, shows embodiment of method that the heterogeneity tissue is striden time cross-modality medical imaging.In the embodiment shown in fig. 3, a plurality of multimode medical images have experienced a series of process.These processes are carried out specific function, comprise the medical image of gathering a form, gather the medical image 1204 of another form, the cross-modality image shines upon 1206, the tissue property of striding in the chronomedicine image sequence is carried out arbitrary classification 1202, and carries out interactive mode and stride time cross-modality inspection 1208.

Below each process shown in Figure 3 will be described in more detail.Notice that the medical image sequence of using may or can not be gathered under the situation of introducing contrast medium in step 1202.If, then in step 1202, need not to carry out tissue property's classification not introducing collection medical image sequence under the situation of contrast medium.

Referring now to Fig. 4, summarize the method for striding time check of heterogeneity tissue in the time dependent medical image.Fig. 4 is a process flow diagram, and an embodiment of the time check method of striding of heterogeneity tissue in the medical image of the present invention is shown.In the embodiment shown in fig. 4, a plurality of medical image images are striden time series and have been experienced a series of process 802.Each carries out specific function these processes, as registration 806 between registration in the sequence 804, sequence, performance graph classification 808 and visual and diagnose 810.

To introduce the notion of Image registration now.

Referring now to Fig. 5, there is shown the process flow diagram of conventional Image registration process method.The purpose of Image registration is to determine coordinate (bidimensional image) in a space and the mapping between the coordinate (another bidimensional image) in another space, as corresponding to the some mutual mapping in two spaces of the same unique point of object.The process of determining two mappings between the image coordinate can provide the horizontal shift figure and the perpendicular displacement figure of corresponding point in two images.Vertical and horizontal shift figure is used to make the distortion of one of the image that relates to so that the dislocation between two images is dropped to minimum subsequently.

In the Image registration term, two images that relate in the registration process are called source image 1020 and with reference to image 1022.For ease of discussing, be shown I (x with the source image with reference to shadow table respectively _t, y _t, t) and I (x _T+1, y _T+1, t+l).Symbol x and y are the horizontal coordinate and the vertical coordinates of image coordinate system, and t is image index (image 1, an image 2 etc.).(x=0 y=0) is defined as center at image plane to the initial point of image coordinate system.It should be noted that image coordinate x and y be integer not necessarily.

For ease of realizing that image (or image pixel) is also weaved into index I, and (i, j), wherein, i and j definitely are integers, and for simplicity's sake, parametric t is left in the basket.This expression meets for the matrix in the discrete domain indexs.If image (matrix) highly is h, width is w, and (i, j) corresponding image plane coordinate x and y can be calculated as x=i-(w-1)/2.0, and y=(h-1)/2.0-j then in the position.Column index i from 0 to w-1.Line index j from 0 to h-1.

Usually, registration process will be searched the optimal mapping function phi _T+1(x _t, y _t) (referring to step 1002), make

[x _t+1，y _t+1，1] ^T＝Φ _t+1(x _t，y _t)[x _t，y _t，1] ^T (10-1)

The transforming function transformation function of equation (10-1) is a 3x3 matrix, has the key element shown in the equation (10-2).

$\mathrm{\Φ}=\left[\begin{array}{ccc}{\mathrm{\φ}}_{00}& {\mathrm{\φ}}_{01}& {\mathrm{\φ}}_{02}\\ {\mathrm{\φ}}_{10}& {\mathrm{\φ}}_{11}& {\mathrm{\φ}}_{12}\\ 0& 0& 1\end{array}\right]---(10-2)$

Transformation matrix is formed the gyrator matrix by two parts $\left[\begin{array}{cc}{\mathrm{\φ}}_{00}& {\mathrm{\φ}}_{01}\\ {\mathrm{\φ}}_{10}& {\mathrm{\φ}}_{11}\end{array}\right]$ And translation vector $\left[\begin{array}{c}{\mathrm{\φ}}_{02}\\ {\mathrm{\φ}}_{12}\end{array}\right].$

Notice that transforming function transformation function Φ is overall situation function or local function.Overall situation function Φ is each pixel in the conversion image in an identical manner.Local function phi is based on location of pixels each pixel in the conversion image by different way.For the Image registration task, transforming function transformation function Φ can overall situation function or local function or both combinations.

In fact, transforming function transformation function Φ generates two displacement diagrams (step 1004), X (i, j) and Y (i, j), the information that these two displacement diagrams comprise can be directed to the pixel in the image of source and the reposition of harmonizing with reference to respective pixel position in the image.In other words, in step 1008, the source image will spatially be correlated with, and becomes registration source image 1024.For two displacement diagram X (i, j) and Y (i, j), column index i is from 0 to w-1, and line index j from 0 to h-1.

The demonstration result that dislocation is proofreaied and correct has been shown among Fig. 6.Shown in this figure is source image 1102 and with reference to image 1106.Source image 1102 with reference to image 1106 between different vertical dislocations is arranged.By with step application shown in Figure 5 to these two images, obtained in Fig. 6, being shown the source image that the vertical dislocation of image 1104 has been proofreaied and correct.

Notice that the registration Algorithm of using can be Rigid Registration algorithm, non-rigid registration algorithm or both combinations in calculating image transforming function transformation function Φ.Those skilled in the art will recognize that many registration Algorithm can execute the task, searched and generate required displacement diagram so that proofread and correct the transforming function transformation function Φ that misplaces between two coherent videos.With upper/lower positions, among the Lydia Ng etc. " by the medical visualization of ITK realization " (Medical Visualization with ITK) that the people showed, can find the demonstration registration Algorithm: http://www.itk.org.In addition, those skilled in the art understands, by using any suitable image insertion algorithm, can be by means of displacement diagram implementation space correcting image (for example referring to " robot vision " (" Robot Vision ", by Berthold Klaus Paul Horn, The MITPress Cambridge, Massachusetts)).

Refer again to Fig. 5, by means of the present invention, above-mentioned Image registration process can be considered have input end A (1032), the black surround 1000 of input end B (1034) and output terminal D (1036).The present invention is relevant below has tissue of different nature and strides in the explanation of time check and will use frame 1000.

Now with reference to Fig. 7, registration 806 processes between interior registration 804 of sequence shown in Figure 4 and sequence are described more specifically.

The demonstration MRI image sequence that is used for object (for example, mammary gland) has been shown among Fig. 7.MRI image sequence 704 comprises a plurality of demonstration MRI tomography set 706,708 and 710 of same target (for example, mammary gland).Each MRI tomography set is included as a plurality of tomographies of object (mammary gland) image (xsect).Demonstration tomography shown in Figure 7 is the tomography (image) 712 of set 706, the tomography (image) 714 of set 708 and the tomography (image) 716 of set 710.

MRI tomography set has and is intended to different time and takes to catch when introducing contrast medium the changes of function of object on space-time.Demonstration time slot between the set of MRI tomography can be 1 minute, 2 minutes and like that.

Stride time check for having tissue of different nature, except that sequence 704, need one or more MRI image sequences of same target (mammary gland).Demonstration MRI sequence 724 is these type of sequences.Sequence 724 was caught in the time different with sequence 704.Demonstration time slot between sequence 724 and the sequence 704 can be some months.

Be similar to sequence 704, sequence 724 comprises a plurality of demonstration MRI tomography set 726,728 and 730 of same target (mammary gland).Each MRI tomography set is included as a plurality of tomographies of object (mammary gland) image (xsect).The demonstration tomography is the tomography (image) 732 of set 726, the tomography (image) 734 of set 728 and the tomography (image) 736 of set 730.

As implied above, the MRI tomography is integrated into different time shootings to catch the changes of function of object on space-time.Demonstration time slot between the set of MRI tomography can be 1 minute, 2 minutes or like that.

Registration (804) is defined as same xsect tomography (image) registration with object in the MRI image sequence of sets in the sequence.For example, the tomography of sequence 704 (image) 712,714 and 716, and the tomography (image) 732,734 and 736 of sequence 724.

In tissue property's inspection method context of image set, registration embodiment in the sequence has been described, this context serves as independent entity.Because inevitably object (for example, mammary gland) motion during the process of catching the MRI image, dislocation appears in the image of object identical cross-section (for example, 712,714 and 716), therefore, registration needs in the sequence occurred.This dislocation can cause makeing mistakes in tissue property's detecting process.

As mentioned above, stride time check, need to obtain same target two or more image sequences (as sequence 704 and 724) at different time for having tissue of different nature.Corresponding tomography (as tomography 712 and 732) misplaces most probably in the different sequences, and has slightly different shape.Therefore, need carry out registration between sequence (806), and this registration is defined as with the same xsect tomography of not homotactic object (image) registration.The demonstration tomography of wanting mutual registration is to being to 712 and 732, to 714 and 734 and to 716 and 736.

Forward Fig. 8 now to, summarized image set and set up the method for knitting character inspection (reaching step 808, the performance graph classification) jointly.Fig. 8 is the process flow diagram that an embodiment of automatic abnormal structure detection method of the present invention is shown.Notice that process flow diagram shown in Figure 8 serves as an independent community that constitutes self-contained process.Therefore, process flow diagram shown in Figure 8 is unintelligible is the expansion of step 808.On the contrary, step 808 and step 804 can make an explanation by using the step shown in Fig. 8 process flow diagram.

In the embodiment shown in fig. 8, before contrast preparation injects and a plurality of MRI breast images set of gathering afterwards experienced a series of process.Each carries out specific function these processes, as adjustment, subduction, segmentation, system identification and classification.In the present invention, abnormal structure's detection task is finished by the dynamic system parameter classification.

In the embodiment shown in fig. 8, first step 202 (relevant with Fig. 3 step 1202 with Fig. 4 step 802) is used for before a contrast medium injects and gathers a plurality of MRI breast image set afterwards.For striding time cross-modality inspection, step 202 repeats to gather with other a plurality of MRI breast images afterwards before being captured in the injection of another time contrast medium.It should be appreciated by those skilled in the art that, for the cross-modality inspection, the medical image sequence that obtains (step 202) at different time can (for example only comprise an image tomography set in each sequence under the situation that contrast medium does not inject, be used for 706 of sequence 704, be used for sequence 726 726).Under last situation (injecting), step 1202 is carried out the tissue property of gathering the medical image sequence and will stride in the chronomedicine image sequence and is classified.Under back one situation (do not have and inject), step 1202 is carried out the collection of medical image sequence; Correspondingly, with omit step 804 and step 808.Following detailed description is applicable to the situation of introducing contrast medium.

With I ₀(x, y z) are illustrated in the breast MRI image that has a plurality of images (tomography) of spatial order before contrast preparation injects and gather, wherein, z ∈ [1 ... S] be the spatial order index, S is the quantity of image in the set, x and y are respectively the horizontal index and the vertical index of image, wherein, x ∈ [1 ... X], y ∈ [1 ... Y].After contrast preparation is introduced, gather a plurality of MRI image set, each set has the image of the same mammary gland equal number (S) of same space order z.A plurality of MRI image set are taken by for example about one minute temporal resolution.This MRI image set can be expressed as I _k(z), wherein, k is the time sequencing index for x, y, k ∈ [1 ... K]; K is a set quantity.Exemplary set be 706,708 and 710 (three set, K=3), perhaps gather 726,728 and 730 (three set, K=3).Set 706 (first set of sequence 704, demonstration tomography I k=1) _k(z) (in the position 1) is tomography 712 for x, y.

The signal that exists contrast preparation to cause observing in the image collection process time in the imaging system strengthens.The research of these signal time curves makes it possible to owing to different radiography absorbent properties is discerned different types of organizations.For detecting abnormal structure automatically, will be in step 204 (reaching registration in step 804 sequence) with respect to volume coordinate x, y will inject K set of the MRI image I that takes the back at contrast preparation _k(x, y z) carry out space adjustment (dislocation is proofreaied and correct) with MRI image reference set.Usually, MRI image reference set is the MRI image set I that takes before contrast preparation injects ₀(x, y, z).The adjustment process guarantees that the pixel that belongs to the same tissue regions of mammary gland has identical x, the y coordinate in all K image set.The adjustment process is carried out as follows:

for k＝1:K

for z＝1:S

align(I _k(x，y，z)，I ₀(x，y，z))

end

end.

By using black surround 1000 (with reference to Fig. 5), I _k(x, y, z) input terminal A (1032), I ₀(x, y, z) input terminal B (1034), and obtain the I of registration at output terminal D (1038) _k(x, y, z) image.

Can be used for realizing adjustment function align (A, demonstration methods B) is a non-rigid registration, this registration is harmonized terminal A and terminal B, and is extensive use of in medical imaging and remote sensing field.The front has been discussed registration process (dislocation is proofreaied and correct).Person of skill in the art will appreciate that also and can use other method for registering.

As described in reference Fig. 1, after contrast preparation injected, the image pixel intensity of different breast tissues increased by different way.This phenomenon shows from injecting the image of taking the back and deducts the image taken will provide the more clear information of relevant image abnormal structure position for the radiographer before injection.This information also can be used for extracting the zone so that detect and distinguish abnormal structure automatically from former MRI breast image.This information obtains in the step 206 of Fig. 8, and this step is carried out and distinguished a plurality of MRI breast image set I _k(x, y, z), k ∈ [1 ..X] with reference to MRI image set, to produce a plurality of difference images set, δ I _k(x, y, z), k ∈ [1 ... K].MRI image set I ₀(x, y z) are selected as the intensity reference image.The difference process is carried out as follows:

for k＝1:K

for z＝1:S

δI _k(x，y，z)＝subtraction(I _k(x，y，z)，I ₀(x，y，z))

end

end

Wherein, (A B) deducts B from A to function subtraction.

In the step 208 of Fig. 3, difference image, δ I _k(x, y z) will carry out fragmenting process, and this process is at first estimated a plurality of difference image set delta I _k(x, y z), and produce a plurality of shade image set M that obtain to issue orders by carrying out _k(x, y, z), k ∈ [1 ... k]:

for k＝1:K

for z＝1:S

for x＝1:X

for y＝1:Y

if?δI _k(x，y，z)＞T

M _k(x，y，z)＝1

end

end

end

end

end

Wherein, shade image set M _k(x, y, z), k ∈ [1 ... K] by zero initialization, T is the STATISTICAL STRENGTH threshold value.The exemplary value of T is an empirical value 10.

Fragmenting process in step 208 is according to shade image M _k(z) non-zero pixels in is with a plurality of MRI breast image set I for x, y _k(z) the image segmentation in is to obtain the intensity pixel of segmentation in a plurality of MRI breast image set images for x, y.Shadow table as a result is shown S _k(x, y, z), k ∈ [1 ... K], staged operation can be expressed as:

for k＝1:K

for z＝1:S

for x＝1:X

for y＝1:Y

if?M _k(x，y，z)＝1

S _k(x，y，z)＝I _k(x，y，z)

end

end

end

end

end

Wherein, image S _k(x, y z) are initialized as zero.

Person of skill in the art will appreciate that, in reality realizes, can ignore the stage that generates the shade image, and can be by carrying out with the realization fragmenting process of issuing orders:

for k＝1:K

for z＝1:S

for x＝1:X

for y＝1:Y

if?δI _k(x，y，z)＞T

S _k(x，y，z)＝I _k(x，y，z)

end

end

end

end

end

Wherein, image S _k(x, y z) are initialized as zero.

The step 210 of Fig. 8 is dynamic system identification steps, and it is described with reference to Fig. 9 and Figure 10.In Fig. 9, show the chart of graph copies shown in Figure 1, but Fig. 9 comprises step function f (t), the curve 302 of insertion and has deleted normal and adipose tissue curve.

The objective of the invention is to detect abnormal structure, and the more important thing is and distinguish pernicious and benign tissue.(annotate: step function f (t) is defined as f (t＜0); F (t 〉=0)=| λ |; λ ≠ 0.) belong to normal and adipose tissue pixel in division step 208 at image S _k(x, y are made as zero in z).Image S _k(z) residual pixel in belongs to pernicious or benign tissue for x, y.

Only by in the assessment of static form, promptly the pixel intensity (intensity) in each image is to be difficult to distinguish malignant tissue and benign tissue.But in dynamic-form, the brightness change has shown the difference between these two types of tissues.As shown in Figure 9, since the time zero, the brightness of malignant tissue (contrast) curve 304, m (t), very fast rising surpasses step function curve 302, and the progressive subsequently step function curve 302 that reaches; And the brightness of benign tissue (contrast) curve 306, b (t) slowly rises for 302 times at the step function curve, and the progressive subsequently step function curve that reaches, f (t), 302.

Those skilled in the art will recognize that brightness (contrast) curve 304, m (t), the step response of similar underdamping dynamic system, and brightness (contrast) curve 306, b (t), the step response of similar overdamping dynamic system.

Figure 10 summarizes the demonstration conventional scheme that shows the behavior of identification dynamic system.For the dynamic system 404 of the unknown, step function 402 is as excitation.The response 406 feedthrough system identification steps 408 of 404 pairs of step functions 402 of dynamic system are so that the dynamic parameter of estimating system 404.

As shown in Figure 8, the system modelling of dynamic system identification 210 can be finished in step 212.The demonstration that Figure 11 shows dynamic system modeling 212 realizes that among the figure, it is shown ARX (returning automatically) model 500 (showing " system identification tool box " referring to The Math Works Lennart Ljung).

Conventional ARX model can be expressed as equation:

y(t)＝G(q)f(t)+H(q)ε(t) (1)

Wherein, G (q) (506) and H (q) (504) are ssystem transfer functions as shown in figure 11, and u (t) (502) is excitation, | ε (t) (508) is disturbance, and y (t) (510) is system's output.As everyone knows, transport function G (q) (506) and H (q) (504) can be according to rational function q ^-1Specify, and specify the molecule and the denominator coefficients of following form:

$G\left(q\right)=q^{-\mathrm{nk}}\frac{B\left(q\right)}{A\left(q\right)}---\left(2\right)$

$H\left(q\right)=\frac{1}{A\left(q\right)}---\left(3\right)$

Wherein, A and B are delay operator q ^-1In polynomial expression:

A(q)＝1+a ₁q ^-1+.......+a _naq ^-na (4)

B(q)＝b ₁+b ₂q ^-1+.......+a _nbq ^-nb+1 (5)

When A and B were polynomial expression, the ARX model of system can obviously be expressed as:

y(t)＝-a ₁y(t-1)-...-a _nay(t-na)+b ₁u(t-nk)+...b _nbu(t-nk-nb+1)+e(t)?(6)

Equation (6) can further be expressed as recurrence:

Wherein

The system identification of coefficient vector θ separate into

$\widehat{\mathrm{\θ}}={\left({\mathrm{\Φ}}^T\mathrm{\Φ}\right)}^{-1}{\mathrm{\Φ}}^{T}Y---\left(8\right)$

Wherein

With

$Y=\left[\begin{array}{c}y\left(t_0\right)\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ y({\mathrm{<figure-callout\; id="0"\; label="t"\; filenames="A20068004931200321.png,A20068004931200401.png"\; state="\{\{state\}\}">t</figure-callout>}}_0+N_t-1)\end{array}\right]---\left(10\right)$

In equation (9) and (10), t ₀Be the data sampling start time, and N _tIt is sample size.

Relevant with brightness (contrast) curve m (t) 304 with brightness (contrast) curve b (t) 306, be respectively

With

Under this particular case, u (t) is a step function.And corresponding separating is

With

Calculating realized that dynamic system discerns 210 steps (and Figure 10 step 408).

Referring again to Fig. 8, be optimum or malignant tumour for will in the MRI image, having high territorial classification (classification step 214) to specific luminance, provide and monitored learning procedure 218.

Be subjected to monitoring study to be defined as a study course, wherein, typical case's set is made up of paired input and required output.Under this MRI image breast tissue classification situation, typical case's input is

With

(or known curve), typical required output are to be respectively applied for pernicious and carcinoid designator O _mAnd O _bIn Fig. 8, step 218 receives M sample breast MRI performance graph with known features (optimum or pernicious) from step 216.The exemplary value of M can be 100.In M curve, M is arranged _mIndividual curve belongs to malignant tumour, and M _bIndividual curve belongs to benign tumour.M _mAnd M _bExample value can be 50 and 50.In step 218, equation (8) is applied to all sample curve has generated M coefficient vector

Wherein, M _mIndividual coefficient vector (is expressed as

I=1 ... M _m) expression has a designator O _mMalignant tumour, and M _bIndividual coefficient vector (is expressed as

I=1 ... M _b) expression has a designator O _bBenign tumour.These have understood coefficient vector

With

Be used to train sorter, and sorter be used for detect or the diagnosis process with the dynamic arthrography curve classification.

Increase specificity (accuracy of distinguishing benign tumour and malignant tumour), other factors (step 220) can be covered training (study) and classification process.As everyone knows, introduce the factor of time, acquisition time and tomography thickness (referring to " breast MRI that radiography strengthens: influence susceptibility and specific factor " (" Contrast-enhanced breast MRI:factors affecting sensitivity andspecificity " such as contrast preparation introducing speed, the relevant radiography of imaging, by CW.Piccoli, Eur.Radiol.7 (Suppl.5), S281-S288 (1997))).

Contrast preparation is introduced velometer be shown α, it is β that the relevant radiography of imaging is introduced time representation, and acquisition time is expressed as γ, and tomography thickness is expressed as δ.These demonstration factors are wanted the attachment coefficient vector

With

Use together, with the training sorter, and the territorial classification that sorter is used for the MRI breast image is pernicious or the benign tumour class.Notice that these demonstration factors should be similar to coefficient vector

With

Quantize in the scope of scope.

For reaching study and training purpose, made up training data set

{p _jτ _j}，j＝1...l，τj＝{-1，1}，

Wherein, τ _jIt is the class label.

With respect to parameter w, and with respect to undetermined multiple ξ _j〉=0 with its maximization.

After optimization problem solved, the expression formula of w can be represented according to the support vector with nonzero coefficient in the equation (13), and inserted in the equation so that with the lineoid classification, produce the SVM decision function:

$\mathrm{\&Psi;}\left(p_{\mathrm{new}}\right)=(w\&CenterDot;p_{\mathrm{new}}+\mathrm{\&sigma;})=\underset{j=1}{\overset{l_s}{\mathrm{\&Sigma;}}}{\mathrm{\&tau;}}_j{\mathrm{\&xi;}}_j{p}_j\&CenterDot;p_{\mathrm{new}}+\mathrm{\&sigma;}---\left(15\right)$

Wherein, l _sBe the quantity of support vector.New vectorial P _NewBe categorized into one of two classes (pernicious and optimum) and be based on the decision function mark.Person of skill in the art will appreciate that under non-separation situation, can use non-linear SVM.

Image set is set up the said method (and step 804 and step 808) of knitting character and checking jointly and is applied to and waits all to stride the time image sequence such as 704 and 724 to check to stride time tissue property.Be appreciated that in the present invention the time image sequence of striding will be carried out inner registration and mutual registration before entering step 808.The inside registration of sequence and mutually a demonstration of step of registration implementation be earlier with inner registration application to sequence 704, then with mutual registration application to sequence 704 and 724.Person of skill in the art will appreciate that sequence 704 and 724 role are interchangeable.

For the inside registration sequence 704 of this particular exemplary implementation, select the image set cooperation for reference to the image set arbitrarily, for example, gather 706.Set 706 image is input terminal B (Fig. 5 1034) subsequently, other image set (for example, 708 and 710) input terminal A (Fig. 5 1032).The image of registration of image set (708 and 710) obtains at terminal D (Fig. 5 1036).

For the mutual registration of this particular exemplary implementation, the image input terminal A of sequence 724 (Fig. 5 1032), the image input terminal B of sequence 704 (Fig. 5 1034), and

For example, if tumour is pernicious, τ then _j=1, otherwise, τ then _j=-1.Vector $p_j=[\widehat{\mathrm{\&theta;}},\mathrm{\&alpha;},\mathrm{\&beta;},\mathrm{\&gamma;},\mathrm{\&delta;}]$ In computer vision literature, be called proper vector traditionally.Symbol

Representative domain, d are the territory dimensions.For this exemplary scenario, suppose that coefficient vector θ has five key elements, then d=9.Data layout in the equation (11) uses in being monitored learning procedure 218 and classification step 214.Person of skill in the art will appreciate that data vector p _jCan make up in a different manner, and increase with being different from other above-mentioned physics or non-physics numeral key element (factor).

By using dynamic arthrography curve and other physics or non-physical factor, there is the sorter of known type to can be used for finishing difference malignant tumour and carcinoid task.The demonstration sorter is that SVM (support vector machine) is referring to " study course of relevant pattern-recognition support vector machine " (" A Tutorialon Support Vector Machines for Pattern Recognition ", by C.Burges, DataMining and Knowledge Discovery, 2 (2), 1-47,1998, Kluwer AcademicPublisher, Boston), provide information on the following website:

http://ava.technion.ac.il/karniel/CMCC/SVM-tutorial.pdf)。

The sample situation of svm classifier device will be to data training and the classification of expression by two classes of lineoid separation.The lineoid of separating data satisfies

w·p+σ＝0 (12)

Wherein, be dot product.

The purpose of training SVM is to determine free parameter w and σ.Change of scale can be applied to all the time W and σ conversion in case all data observe and to have matched inequality:

${\mathrm{\&tau;}}_j(w\&CenterDot;p_j+\mathrm{\&sigma;})-1\&GreaterEqual;0,\&ForAll;j---\left(13\right)$

Equation (13) can be found the solution by minimizing following glug Lang Ri function:

$L(w,\mathrm{\&xi;})=\frac{1}{2}{\left|\right|w\left|\right|}^2-\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&xi;}}_j\left({\mathrm{\&tau;}}_j(w\&CenterDot;p_j+\mathrm{\&sigma;})\right)---\left(14\right)$

Obtain the image of registration of sequence 724 at lead-out terminal D (Fig. 5 1036).

In step 808 (Fig. 4) when finishing, generate a plurality of performance graphs (being two curves in the current exemplary scenario), be reflected in a plurality of tissue properties of striding time image sequence (for example, two of current exemplary scenario sequences 704 and 724) seizure of a plurality of time instance (being two examples in the current exemplary scenario).As everyone knows, these performance graphs provide the valuable information of relevant patient disease situation (or progress) for the medical worker.

In step 810, visualization tool is used for checking that by the medical worker relevant range (region-of-interest of image) of object is so that make diagnosis better.Figure 12 shows an embodiment of this type of visual facility.

Illustrated among Figure 12 and be applicable to the computer monitor screen 900 that the image processor of putting into practice described method step (can corresponding to the image processor 102 of Fig. 2) is communicated by letter (can corresponding to the display among Fig. 2 104).

On screen 900, shown in the representative image tomography 712 of two shown in the screen left half and 732.For example, tomography 712 is 1 I that stride three set (706,708 and 710) in the locus _k(x, y, 1) | first image of k ∈ [1,2,3]; Tomography 732 is 1 I that stride three set (726,728 and 730) in the locus _k(x, y, 1) | first image of k ∈ [1,2,3]. Breast image 902 and 912 is respectively shown in tomography 712 and 732. Breast image 902 and 912 is images of the same xsect of mammary gland.

In operation, the medical worker browses image (for example, by mobile computer mouse 108 or other user interface), designator 906 is moved on to the top of position 908 in the tomography 712.Simultaneously, the same space position 918 shows ghost image designators 916 (that is, with tomography 712 in 908 identical locus) in tomography 732.Alternative, the top of the also removable designator 916 of user (as user interface) position 918 in the tomography 732, and simultaneously, 918 identical locus 908 demonstration ghost image mouses 906 in tomography 712 and tomography 732.

For arbitrary layout, in the chart 922 of display screen 900, show two performance graphs (solid-line curve 924 and dashed curve 926).Exemplary curves 924 and 926 has reflected the different tissues character two different time mammary gland same points.For example, the image sequence that comprises tomography 712 can be taken in preceding 6 months of the sequence that seizure comprise tomography 732.The medical worker can move on to mouse other position to check the change in (for example, 6 months) inner tissue's character in one period.By means of this visual facility, analysis of disease progression easily.

It will be apparent to one skilled in the art that tissue property can by except that shown in the alternate manner demonstration of dynamic curve diagram 924 and 926.For example, tissue property can be shown by colored angiogram.Those skilled in the art also will understand, and method of the present invention can be handled and a plurality ofly stride the time image sequence, and a plurality of performance graph can show simultaneously so that carry out medical diagnosis.

The classification of heterogeneity tissue makes it possible to generate special graph, as angiogram.In Figure 13, show the demonstration mammary gland angiogram 1300 that comprises suspected tumor zone 1302 and other tissue regions.In this exemplary view, zone 1302 is further to check () region-of-interest (ROI) for example, quantitative analysis, and remaining other zone is regarded as non-ROI.

In other cases, can there be a plurality of ROI to analyze.But for simplicity, mammary gland angiogram 1300 will be used to describe the present invention and stride the process that the time cross-modality is checked.It will be apparent to one skilled in the art that method of the present invention is applicable to other imaging form (PET, CT, US and like that), other signal (information) form and/or other disease.

Refer again to Fig. 3, discuss method step 1204,1206 and 1208 by specific detail now.

In breast cancer diagnosis, X line mammography has limited specificity and susceptibility.As the MRI mammography of alternative formation method to higher susceptibility being arranged greater than certain size tumor.For medical worker and researchist, check that X line mammography and MRI image are useful to obtain side information.For example, microcalcifications is by orthovoltage x-ray image capturing best results.

In Fig. 3, step 1202 has been gathered and has been striden time MRI image sequence as a form, and step 1204 collection is striden time X line mammography as another form.Show two demonstration X line mammography images 705 and 725 with striding time MRI sequence 704 and 724 in Fig. 7, these two images are being collected the about identical time shooting of sequence 794 with 724 respectively.These two mammography images will be used for the cross-modality analysis in step 1206 and 1208.

Be appreciated that X line mammography image 705 and 725 is three dimensional object (for example, mammary gland) projections, and image sequence 704 and 724 is by forming as the two-dimentional tomography of three dimensional object (mammary gland) xsect image.

For ease of the cross-modality such as data such as 705,725,704 and 724 is detected, the step 1206 of Figure 12 is mapped to a morphological data (image) of higher dimension (MRI sequence 704 and 724) another morphological data of lower dimension (X line image 705 and 725).

With reference to figure shown in Figure 14 A-14E, the mapping process (step 1206) of a form of higher dimension to another form of lower dimension described.

In Figure 14 A, only show the demonstration MRI tomography set 1402 that is similar among Fig. 7 image set such as 706 or 726.For ease of discussing, set 1402 has three tomographies 1403,1404 and 1405 that have breast image 1406,1407 and 1408 respectively.Usually, the 3 D medical device for image produces the image tomography, and wherein, the distance in tomography between the neighbor is often separated less than the center to center tomography.Therefore, the system dimension is not isotropic usually, and this is not desirable in most of medical image analytical applications.Therefore, taked to carry out the step that tomography inserts in the present invention, be isotropy or be enough to, so that can carry out the cross-modality mapping effectively near isotropy so that the image of gathering is gathered (as gathering 1402).Correspondingly, what comprise in the present invention is tomography insertion method, and this method generates the new tomography of any amount between two existing tomographies, so that can obtain isotropy.The formula that tomography inserts can be expressed as:

I _int＝βI _1-β(i→j)+(1-β)I _β(j→i) (16)

Wherein, I _IntBe to insert tomography, I _1-β(i → j) and I _β(j → i) generates I _IntTwo middle phantoms.Tomography I _1-β(i → j) and I _β(j → i) the puppet intersection method for registering by two former adjacent tomography I (i) and I (j) obtains.The action that factor beta and two middle phantoms of 1-β control internally interrupt layer.The effect of subscript β and 1-β will be below partial dislocation figure and pseudo-registration process understand in discussing.For example, in shown in Figure 14 B, tomography 1413 is demonstration I _Int, tomography 1403 is demonstration I (i), and tomography 1404 is demonstration I (j).

The pseudo-method for registering that intersects of two tomographies that tomography of the present invention inserts is described now.

Review is used for the equation (10) of Image registration, transforming function transformation function Φ generate two displacement diagram X (i, j) and Y (i, j), the information that these two displacement diagrams comprise can be directed to the pixel in the image of source and reposition with reference to respective pixel position adjustment in the image.

Generating insertion tomography I _IntIn the practice of (as tomography 1413), partial dislocation figure X has been introduced in the somewhere between two tomographies (as 1403 and 1404) _α(i, j) and Y _α(i, j).Partial dislocation figure is directed to the pixel among source image (tomography) I (x) at the source image pixel and the reposition in somewhere between the respective pixel position in reference to image I (y).Partial dislocation figure X _α(i, j) and Y _α(i, j)) is calculated as by the predetermined factor α of particular value:

Y _α(i，j)＝αY(i，j)

X _α(i，j)＝αX(i，j)

Wherein, 0＜α≤1.

The partial dislocation figure that generates is used to make intermediate imagery (tomography) I of source deformation of image to obtain to be calculated as follows subsequently _α(x → y):

I _α(x→y)＝align _partial(I(x)，I(y)，α)，

Wherein, align _Partial(I (x), I (y) are by using controlled variable α α), with former displacement diagram X (i, j) and Y (i j) is revised as X _α(i, j) and Y _α(i j), thereby carries out the source image and with reference to the function of the pseudo-registration of image (adjustment).

By the different controlled variable as shown in equation (16), the pseudo-registration of above-mentioned process is applied to former tomography (for example, 1403 and 1404) again.This has generated the middle phantom (for example, 1413) with two former fault information.Therefore each former tomography serves as with reference to image and source image.Therefore, adopted term " the pseudo-registration that intersects ".

In Figure 14 B, tomography set 1412 illustrates two and inserts tomography 1413 (having former tomography 1403 and 1404) and 1414 (having former tomography 1404 and 1405).Because every pair of former tomography has only one to insert tomography, therefore, for this exemplary scenario, parameter beta is chosen as 0.5.Usually, parameter beta is calculated as β _k=k/ (N+1), wherein, k ∈ [1,2 ... N], and N is required insertion tomography quantity.Equation (16) becomes:

$I_{\mathrm{int}}^{{\mathrm{\&beta;}}_k}={\mathrm{\&beta;}}_k{I}_{1-{\mathrm{\&beta;}}_k}(i\&RightArrow;j)+(1-{\mathrm{\&beta;}}_k)I_{{\mathrm{\&beta;}}_k}(j\&RightArrow;i).$

Breast image 1406,1407 and 1408 illustrated examples are shown in the tomography set 1402.The breast image 1415 that inserts in tomography set 1412 shows the insertion mammary gland of size in the middle of 1406 and 1407.In tomography set 1412, between former tomography 1404 and 1405, there is another to insert tomography 1414.

For ease of discussing, tomography set 1412 is included in the insertion tomography of right quantity between every pair of former tomography, so that satisfy the requirement of isotropy voxel.Therefore, tomography is gathered the three-dimensional MRI body of 1412 indicated objects (mammary gland), need be mapped to more low-dimensional (2D) space so that check with object (mammary gland) expression (X line) in two-dimensional space.Represent that from higher-dimension more the mapping that low-dimensional is more represented relates to the projection of paying close attention to phase.For the demonstration breast examination, usually accept be mutually X line mammography a pin (CC) mutually with skew back (ML) mutually.

Though digital MRI body (for example, insert tomography set 1412) available, but what the medical worker paid close attention to is can be by the body around axle 1417 scissors faulies set 1412, and throw this body along direction 1419 (parallel) or direction 1421 (vertical) subsequently with fault surface with the tomography vertical margin, can obtain any phase (comprise CC with ML mutually).Notice that axle 1417 is roughly parallel with the apical margin or the root edge of tomography, and be center ideally by this body.In fact, axle 1417 is by the center of practical object (mammary gland) body, and this is because the center of subject needn't be consistent with the center of tomography body usually.Argumentation is searched the method for rotation center (object centers) in the back.

With reference to Figure 14 C, be by the new layering of body weight around a demonstration methods of axle 1417 scissors fault bodies with tomography set 1412, wherein, tomography (for example, 1423,1424 and 1425 of set 1422) is vertical with spools 1417 as a result.Rotate each tomography 1423,1424 and 1425 and roughly be equal to scissors fault set 1412.These new tomographies (1423,1424 and 1425) intersect with tomography 1403,1413,1404,1414 and 1405.Become lines in 1423,1424 and 1425 such as breast images such as 1406,1415,1407 and 1408.Shown in Figure 14 D, projection tomography 1423,1424 and 1425 has the figure of putting shown in the image 1,433 1432 with generation on direction 1419. Projection tomography 1423,1424 and 1425 has the figure 1434 of lines shown in the image 1432 with generation on direction 1421.

In a more general case, can be before carrying out projection tomography being gathered 1412 rotates in the field of view angle mode around axle 1443,1444 and 1445 (referring to the figure 1442 of Figure 14 E).Those skilled in the art will know that another selection of throwing from arbitrarily angled acquisition is that imaginary projector is placed 3d space, and rotate imaginary projector around axle 1443,1444 and 1445 when the static tomography set 1412 of execution is throwed.

With reference to Figure 15, the method for searching rotation center (object centers) is described.Intersecting of tomography 1423 and breast image (1406,1415,1407 etc.) can produce pixel cloud 1602.Center (the O of cloud 1602 ₁, O ₂) be calculated as:

o ₁＝m ₁₀/m ₀₀

o ₂＝m ₀₁/m ₀₀

Wherein, moment m _PqBe calculated as:

$m_{\mathrm{pq}}=\underset{-\&infin;}{\overset{\&infin;}{\&Integral;}}\underset{-\&infin;}{\overset{\&infin;}{\&Integral;}}{c}_1^p{c}_2^{q}f({\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="A20068004931200421.png"\; state="\{\{state\}\}">c</figure-callout>}}_1,c_2)d{c}_1\mathrm{<figure-callout\; id="2"\; label="d\; c"\; filenames="A20068004931200321.png"\; state="\{\{state\}\}">d</figure-callout>}{c}_{}{}_2$

Wherein, the f (c in the cloud 1602 ₁, c ₂)=1 is 0 in other position, and C ₁And C ₂In this uses is image coordinate (Figure 15).

For example, Figure 16 is illustrated in three projections that tomography inserts back MRI mammary gland body.Image 1533 is the projections along direction 1419, and image 1544 is the projections along direction 1421.There is another projection 1555 in the direction of not discussing along the front 1417.In fact, some medical workers are considered as required at least projecting direction with this.

After the 3D body was projected the 2D space, mapping process carried out registration with result's projection (for example, image 1533 and 1544) and the image of directly gathering (as X line mammography 705) in the 2D space.Notice that the 3D body that relates to can be former tomography that is untreated (as tomography set 796) or the 3D body be made up of angiogenesis image (as Figure 130 0) in mapping (projection and registration).

For time cross-modality inspection is striden in execution, the projection of 3D body (as tomography set 706 or 716) need be carried out registration with image (as 705 or 725).In addition, 706 and 726 projection needs mutual registration.In addition, image 705 and 725 also needs mutual registration.These essential registrations help carrying out interactive mode and stride time cross-modality inspection in step 1208, this uses the demonstration situation to explain below.

The computer monitor screen 900 of communicating by letter with the image processor (102) of carrying out step noted earlier (it can corresponding to the display 104 of Fig. 2) has been shown among Figure 17.What show on screen 900 is that two representativenesses are striden time 2D image (X line mammary gland figure) 705 and 725.For example, image 705 is image 725 seizure in preceding 6 months of catching for same target (mammary gland). Image 705 and 725 is mutual registration after collection.What show on screen 900 in addition, is two and strides time MRI body projection 1705 and 1725.The actual example of projection has been shown among Figure 16.Mesosome projection 1705 and 1725 mutual registrations when striding similarly.In addition, they also with 795 and 725 registrations.

In the demonstration of striding time cross-modality inspection, the medical worker moves on to designator 1706 (as the mouse that provides by user interface) top of mammary gland 1702 certain position 1708 in the image 705.Roughly at one time, be presented at the position 1718 of mammary gland 1712 in the image 725 such as circle 1716 marks such as grade, 1708 in this position and the image 705 is in the same space position.In addition, circle 1726 is presented at around the locus 1728 of mammary gland 1722 in the image 1705, and 1708 in the same space position in this position and the image 705.Also have, circle 1736 appears at around the locus 1738 of mammary gland 1732 in the image 1705, and 1708 in this position and the image 705 is in the same space position.In fact, the medical worker can select to stride the focus (locations/regions) in the arbitrary image (tomography) that relates in the time cross-modality inspection, corresponding point (zone) will highlight by mark (as circle or square or other shape) in all other images (tomography), so that carry out pathological analysis.As shown in Figure 9, may need two performance graphs (924 solid lines and 926 dotted lines) to appear in the shielding 900, wherein, exemplary curves 924 and 926 is reflected in the different tissues character of the same point of two different time mammary gland.

The time cross-modality inspection method of striding of the present invention can realize in CAD (computer-aided diagnosis) workstation independently or in PACS (picture archive and transmission system).Check result can send by the network link of safety or the radio communication of safety.

Theme of the present invention relates to digitized video to be handled and computer vision technique, this is appreciated that also can understand object, attribute or condition to the people thus specifies useful connotation with identification for expression is handled digitized video with digital form, and utilizes the result's who obtains in digitized video is further handled technology subsequently.

The computer program of carrying out the inventive method can be stored in the computer-readable recording medium.This medium for example can comprise: such as the magnetic storage medium or the tape of disk (as hard disk drive or floppy disk); Such as optical storage mediums such as CD, light belt or machine readable barcode; Such as random-access memory (ram) or ROM (read-only memory) solid state electronic memory device such as (ROM); Or be used for any other physical unit or the medium of storage computation machine program.The computer program of carrying out the inventive method also can be stored in by the Internet or other communication media and be connected on the computer-readable recording medium of image processor.Those skilled in the art will recognize easily that the equivalent of this type of computer program also can make up in hardware.

Claims (6)

1. method that the medical image that is used to time of striding and cross-modality is analyzed comprises:

Visit is according to first set of first form at the medical image of the research object of very first time section seizure;

Visit is according to second set of second form at the medical image of the described research object of second time period seizure, and described first and second set respectively comprise a plurality of medical images;

Be mapped to the predetermined space coordinate by described a plurality of medical images, carry out Image registration described first and second set;

Carry out the time of the striding image mapping of described first and second set; And

Be provided for interactive mode and stride the parts of chronomedicine image analysing computer.

2. the method for claim 1, the described step of wherein carrying out Image registration comprises:

Carry out the inside registration of described a plurality of medical images of described first and second set; And

Carry out the mutual registration of described a plurality of medical images of described first and second set.

3. the method for claim 1 comprises that also at least one tissue property of the described images of carrying out described first and second set checks.

4. the method for claim 1 also comprises:

Visit the reference set of the medical image of described research object;

Distinguish described first and second set and described reference set, comprise the difference image set of a plurality of images with generation;

With described a plurality of image segmentations of described difference image set, a plurality of images that have segmentation intensity pixel with generation;

System identification is applied to have segmentation intensity pixel described a plurality of images to generate a plurality of systematic parameters; And

With described a plurality of systematic parameter classification.

5. method as claimed in claim 4 also comprises: before with described a plurality of systematic parameter classification, increase described systematic parameter by physics or non-physical factor.

6. the method for claim 1 also comprises: after carrying out Image registration, with tissue typing of different nature.

Images