CN100561518C - Self-adaptation medical image sequence interpolation method based on area-of-interest - Google Patents

Abstract

The invention discloses a kind of based on interpolation method between the self-adaptation medical image sequence tomography of area-of-interest, this method makes full use of characteristics such as organizing correlativity and voxel correlativity, earlier judge whether interpolation point belongs to area-of-interest, select suitable interpolation method more in view of the above.The present invention has not only considered the variation of gray scale, has also taken into account and has dissected the variation of structure outline, thereby better solved the deficiency of traditional interpolation method, and the picture that new interpolation is gone out more approaches actual conditions.Experiment showed, that visual effect of the present invention and picture quality have had bigger improvement, picture clear-cut, the noise that interpolation goes out be less, no longer include double outline, improved the quality of interpolation, the more important thing is that significantly reduce computing time.Apply the present invention in the system of three-dimensional reconstruction and demonstration, the image boundary that drafting is come out is clear, and the terraced fields effect is reduced to minimum, more realistic demand, thus improve accuracy and the work efficiency that the doctor judges in a big way.

Description

Self-adaptation medical image sequence interpolation method based on area-of-interest

Technical field

The present invention relates to the method for interpolation between a kind of sequence image tomography, relate in particular to a kind of based on interpolation method between the self-adaptation medical image sequence tomography of ROI, promptly the interlayer image that is made new advances by the known sequences image interpolation is core technology during medical image three-dimensional is rebuild.

Background technology

Current society is in the epoch of an information explosion, and people seem in face of usually in boundless and indistinct data ocean and be at a loss, and is difficult to catch for the moment the essence, structure and the rule that are hidden among the data.Visual (Visualization) grows up under this background, and it becomes a kind of form--figure of being accepted and understand by the people of being easy to data conversion.The seventies in 20th century, the successful Application of CT technology in clinical medicine in the new era of having started medical imaging, makes human body carried out noninvasive test and diagnosis becomes a reality.The eighties in 20th century, the successful Application of image technology on clinical medicine that MRI, PET, SPECT etc. are advanced makes that more medical imaging is developed rapidly.In present image medical diagnosis, mainly be to remove to find the pathology body by the two-dimensional slice image of observing a group of CT, MRI, this mainly depends on the sheet experience of reading that the medical personnel enriches image is carried out qualitative analysis.Utilize computer technology that two-dimensional slice image is carried out three-dimensional reconstruction and 3-D display, can assist the medical personnel to carry out qualitative until accurate quantitative analysis to pathology body and other area-of-interests, like this medical personnel see better, more accurate, thereby can improve the accuracy and the real-time of medical diagnosis greatly.

The three-dimensional visualization technique of so-called medical image just is meant to be utilized a series of two-dimensional slice image reconstruction of three-dimensional images model and carries out technology qualitative, quantitative test.Expect the good three-dimensional reconstruction image of effect, need to solve: 1) two dimensional slice data that has collected is carried out suitable pre-service; 2) set up structure, the rule that corresponding model comes expression data; 3) drafting and the demonstration of structure model.But because tomoscans such as CT and MRI are subjected to the restriction of aspect factors such as device hardware condition, security and economy, view data is anisotropic, be that distance between adjacent two faultage images (is got the Cranial Computed Tomography picture, about 2.8mm) distance of adjacent two pixels (is got the Cranial Computed Tomography picture in the same tension fault image, about 0.4mm), this makes the 3D solid and the real-world object that directly reconstruct differ greatly, and display effect is not good.Therefore obtained the three-dimensional data of tissue by a series of two-dimensional ct images, its key is the interlayer interpolation, and promptly interpolation goes out plurality of pictures between two tomography pictures, thereby better carries out three-dimensional reconstruction.The drafting that makes up model comprises that the two dimensional slice data that collects is carried out interpolation to be obtained three-dimensional data and show with showing, wherein to the two dimensional slice data that collects carry out interpolation obtain three-dimensional data be most important also be a step of most critical.Realize that interpolation obtains best intermediate vegetarian refreshments and just can obtain detailed 3 D rendering, make new images (three-dimensional reconstruction) not only on gray-scale value, and organizing more realistic in shape demand.Only obtain the complete description of object that three-dimensional data characterizes, set up virtual tissue and organ, just help the calculating of medical analysis, diagnosis and the 3-dimensional dose field in radiotherapy treatment planning.

In the prior art, all be to adopt the simplest linear interpolation or point of proximity interpolation in the method for drafting that provides in the kit such as VTK, ITK for example, this 3-D view and real-world object that has just caused drafting to come out differs bigger, and obscurity boundary, has very strong terraced fields effect.Therefore improve or design a kind of suitable medical image three-dimensional and rebuild, and the image boundary that drafting is come out is clear, and the terraced fields effect is reduced to minimum, the interpolation method of more realistic demand is one and possesses very much challenging work.

Although to the two dimensional slice data that collects carry out interpolation obtain three-dimensional data be most important in the drawing three-dimensional also be a step of most critical.Yet the method for discussing this question specially also is applied in the system this method few.The method that has is that the hypothesis gray-scale value is linear change in Z-direction, utilizes a smooth curve to come the match given data, but the interpolation result obscurity boundary.The method that has is by source figure is carried out wavelet transformation, between the corresponding wavelet coefficient of faultage image, carry out intensity and position interpolation, thereby overcome shortcomings such as obscurity boundary, but be about 2000 times of linear interpolation algorithm the computing time of this method, amount of calculation is very big, is difficult to realize in real-time medical image shows.Said method is owing to all deficiencies are difficult to be applied to three-dimensional reconstruction.

The interpolation of two-dimension picture is divided into the inner interpolation of single width picture and the interpolation picture that makes new advances between the multilayer sequence of pictures.For example, Chinese invention patent application CN1750042A discloses a kind of interpolation processing method for digital image, comprises the steps: the white space between pixel on average is divided into a plurality of grids; According to current this grid position, determine the interpolation parameter of each grid with respect to its sampled point; The interpolation parameter compression of all grids with respect to its sampled point stored; Zoomed image is determined the target grid that interpolation point falls into; Location solution according to this target grid compresses interpolation parameter, thereby calculates the interpolation parameter of this grid correspondence; Reach brightness value, calculate the brightness value of this interpolation point according to this interpolation parameter and this sampled point.This method is considered only is inside interpolation to two dimensional image, can not be used for medical image and rebuild.Again for example, Chinese invention patent application CN1722178A discloses a kind of three-dimensional rebuilding method and system of image, this method receives the data for projection from the imaging device of sweep object, identification grips right data for projection altogether corresponding to a projection ray and interpolation is gripped right described data for projection is scanned object with reconstruction image altogether corresponding to described projection ray.But what adopt in this system is the simplest linear interpolation, and therefore the three-dimensional reconstruction effect that obtains is not fully up to expectations in detail, has more serious terraced fields phenomenon.

Summary of the invention

The object of the invention provides a kind of based on interpolation method between the self-adaptation medical image sequence tomography of area-of-interest (ROI), bring drawbacks such as interpolation image obscurity boundary and calculated amount are big in order to overcome existing algorithm, make new images (3-D view) that interpolation goes out not only on gray-scale value, and organizing more realistic in shape demand, so that better carrying out medical image three-dimensional rebuilds

For achieving the above object, the technical solution used in the present invention is: a kind of self-adaptation medical image sequence interpolation method based on area-of-interest, comprise the following steps: that (1) obtains the source data line data pre-service of going forward side by side, and need between every pair of adjacent source picture to determine the number of plies of interpolation; (2) choose a pair of pending adjacent source picture; (3) determine distance between interpolation layer and both sides adjacent source picture; (4) obtain the position of interpolation point, carry out the interpolation operation of interlayer pixel; (5) if untreated interpolation point, repeating step (4) are arranged in the interpolation layer; (6) if untreated interpolation layer is arranged between pending adjacent source picture, repeating step (3) is to (5); (7) if untreated adjacent source picture is arranged, repeating step (2) is realized the interpolation of image thus to (6);

Wherein, the interpolation operation of described step (4) is meant, the position that utilizes the loop program design to obtain interpolation point, again based on the corresponding point in the levels picture to or match point to carrying out the interpolation of interlayer pixel, described interpolation method is:

If a) the interpolation point is positioned at around the picture, belong to marginal point, adopt based on the right gray-scale value of corresponding point up and down and carry out linear interpolation, otherwise execution in step b);

B) utilize the gray-scale value and the difference thereof of levels corresponding point,,, then carry out linear interpolation based on its gray-scale value if do not belong to area-of-interest according to organizing correlation principle to judge whether interpolation point belongs to area-of-interest, otherwise execution in step c);

C) be the center with the levels corresponding point respectively, choose two zones, calculate the coefficient of autocorrelation of organizing in these two zones, if related coefficient is greater than setting threshold, illustrate that two regional integrations are more similar, then carry out Tri linear interpolation based on these two zones, otherwise execution in step d);

D) it is right to seek optimal match point in the levels corresponding region, utilizes the right gray-scale value of corresponding point to carry out linear interpolation.

In the technique scheme, described " obtain source data go forward side by side line data pre-service " is meant according to the requirement of drawing, read the multilayer tomography picture file that absorbs by hardware device CT or nuclear magnetic resonance etc. by read module, the interface function that provides by Medical Image Processing and development platform, obtain their performance data, comprise the resolution of number, individual picture of tomography picture, distance and the needed patient information of doctor between two adjacent pictures; At last the tomography picture is carried out necessary pre-service work, comprise level and smooth and denoising etc.Before carrying out interpolation operation, at first to choose two Zhang Yuan's pictures, because the picture file (by the multilayer tomography picture file of picked-ups such as hardware device CT or nuclear magnetic resonance) that is read by read module is a serial picture, a lot of opening arranged, therefore must determine earlier that two adjacent source pictures are at the position of whole picture file, i.e. I _kAnd I _K+1The value of middle k.When between two Zhang Yuan's pictures, carrying out interpolation, must determine earlier interpolation point (picture, or layer) between two pictures of source the position, promptly with I _kApart from d ₁With with I _K+1Apart from d ₂Described area-of-interest is meant the pairing human organ of patient information that the doctor needs, and the zone of loseing interest in then is meant air or the clear and definite unwanted human organ information of other doctor.In each interpolation layer, to carry out a little interpolation; Between two adjacent source pictures,, each layer is carried out interpolation according to the number of plies of required interpolation; In whole serial picture, per two adjacent source pictures are all carried out above-mentioned interpolation operation; Finish the interpolation reconstruction of whole 3-D view thus.In view of the above, can carry out the drafting and the demonstration of three-dimensional body.

In the technique scheme, described source picture is CT or the MRI picture that meets the DICOM form.DICOM is the abbreviation of Digital Imaging and Communications in Medicine, and its literal meaning has comprised the digital imagery and two aspects of communicating by letter of medical science, is the general picture format of present medical domain.

In the technique scheme, described linear interpolation method is, if the right gray-scale value of corresponding point is respectively f up and down _Ijk, f _{Ij (k+1)}Then the linear interpolation formula is f _Ijd=r f _Ijk+ (1-r) f _{Ij (k+1)}, r=d in the formula ₁/ (d ₁+ d ₂), d ₁, d ₂It is respectively the distance of interpolation layer and two adjacent source pictures.

Further technical scheme in the described step c), is chosen the sub-circular zone of two symmetries and is carried out similarity relatively, and described each zone is 3 * 3 pixels, adds the zone of 1 pixel in the middle of every limit; Or be 5 * 5 pixels, add the zone of 3 pixels in the middle of every limit.Because for medical picture, handled generally all is some tissue of human body or diseased region, and the outline line at these positions all is level and smooth, promptly generally present irregular circle or ellipse, institute thinks the better interpolation of acquisition, abandoning tradition is selected the method in two foursquare zones, but adopts the sub-circular zone of selecting two symmetries, can obtain better effect.

In the technique scheme, in the described step b), judge according to organizing correlation principle whether interpolation point belongs to area-of-interest and be meant, at first provide different tissues intensity profile figure, utilize the gray-scale value of interpolation point levels corresponding point and difference thereof to judge that according to gray value profiles the interpolation point belongs to the zone of loseing interest in and still belongs to area-of-interest according to different tissues gray difference in the human body.

In the described step c), respectively with levels corresponding point V (i, j, k) and V (i, j are the center k+1), choose two regional A _k, and A _K+1, calculate the coefficient of autocorrelation of organizing in these two zones

$\mathrm{\ρ}(A_k,A_{k+1})=\frac{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}(f_{\mathrm{ijk}}-\stackrel{\‾}{f_k})(f_{\mathrm{ij}(k+1)}-\stackrel{\‾}{f_{(k+1)}})}{\sqrt{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{(f_{\mathrm{ijk}}-\stackrel{\‾}{f_k})}^2}\sqrt{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}}{(f_{\mathrm{ij}(k+1)}-\stackrel{\‾}{f_{(k+1)}})}^2},$

In the formula, A _kAnd A _K+1For respectively with levels corresponding point V (i, j, k) and V (i, j, the k+1) zone of choosing for the center, f _Ijk, f _{Ij (k+1)}Be a V (i, j, k) and V (i, j, gray-scale value k+1), f _k, f _(k+1)Be average gray, N is the number of above-mentioned institute favored area interior pixel; If related coefficient greater than preset threshold ρ 0, illustrates that two regional integrations are more similar, belong to identical tissue, then based on these two regional A _k, and A _K+1Carry out Tri linear interpolation.

In the described step d), optimal match point is to meeting following condition, and match point is close to gray-scale value; Match point is close to the rate of gray level of point around reaching; Match point is close to the grey scale change direction of point around reaching; The right distance of match point is less.

Described optimal match point is limited by following formula,

$R=\underset{0\&le;i\&le;N-1}{\mathrm{<figure-callout\; id="0"\; label="min"\; filenames="C2007100238720017H2.png,C2007100238720017H3.png"\; state="\{\{state\}\}">min</figure-callout>}}\left\{R(V_{\mathrm{ki}},V_{(k+1)i})\right\}$

$=\underset{0\&le;i\&le;N-1}{\mathrm{<figure-callout\; id="0"\; label="min"\; filenames="C2007100238720017H2.png,C2007100238720017H3.png"\; state="\{\{state\}\}">min</figure-callout>}}\{Q(V_{\mathrm{ki}},V_{(k+1)i})\&CenterDot;E(V_{\mathrm{ki}},V_{(k+1)i})\}$

Wherein,

Q(V _ki，V _(k+1)i)＝α·Δf+β·ΔK+γ·Δθ

＝α·|f _ki-f _(K+1)i|+β·|K _ki-K _(k+1)i|+γ·Δθ _i

$E(V_{\mathrm{ki}},V_{(k+1)i})=\frac{1}{\mathrm{\&lambda;}}\&CenterDot;\exp \left(\mathrm{\&lambda;}\sqrt{[{(x_{\mathrm{ki}}-x_{(k+1)i})}^2+{(y_{\mathrm{ki}}-y_{(k+1)i})}^2]}\right)$

(i＝0，1，...，N-1)

In the formula, R represents to weigh the generic function of two some matching degrees up and down; V _KiAnd V _{(k+1) i}Be illustrated respectively in two favored area A of institute up and down _k, and A _K+1In the i that compares right to putting, f _KiAnd f _{(k+1) i}Be its gray-scale value, K _KiAnd K _{(k+1) i}Expression point V _KiAnd V _{(k+i) i}The maximum slope at some place, Δ θ _iRepresent that aforementioned two maximum slopes form the angle between the space line; Q (V _Ki, V _{(k+1) i}) expression weighs up and down the function that two points meet first three some matching degree described in the claim 7; E (V _Ki, V _{(k+1) i}) expression weighs up and down that two points meet the 4th point described in the claim 7: the function that the right distance of match point should be less, i.e. exponential distribution function; x _KiAnd x _{(k+1) i}, y _KiAnd y _{(k+1) i}Represent some V respectively _KiAnd V _{(k+1) i}Horizontal ordinate and ordinate; α, beta, gamma, λ are weighting function, in order to regulate the influence of each component to generic function.

Selecting of each weighting function can be obtained by experiment, and a kind of optional scheme is α=8, β=1, γ=0.5, λ=1.Because it is bigger to seek the calculated amount of optimal match point, is faster and better acquisition interpolation, can in realizing, system adopt difference method.

Because the utilization of technique scheme, the present invention compared with prior art has following advantage:

1. the present invention organically combines multiple different interpolation method, and selects interpolation method adaptively according to the character of picture itself, thereby realizes handling respectively to doctor's area-of-interest with to the interpolation of general area;

2. experiment showed, of the present invention is feasible based on interpolation method between the self-adaptation medical image sequence tomography of area-of-interest, and has higher precision;

3. apply the present invention in the three-dimensional reconstruction system, can improve interpolation speed and render speed and do not influence interpolation and draw effect, thereby adapted to the requirement of real-time, improved doctor's work efficiency;

4. the present invention changes twice multiplication of linear interpolation formula into multiplication one time through conversion, and the multiplication number of times reduces to original half like this, has significantly reduced calculated amount;

5. the present invention has not only considered the variation of gray scale, has also taken into account and has dissected the variation of structure outline, thereby solved the deficiency of traditional interpolation method better, and the picture that new interpolation is gone out more approaches actual conditions.Experiment showed, that visual effect of the present invention and picture quality have had bigger improvement, picture clear-cut, the noise that interpolation goes out be less, no longer include double outline, improved the quality of interpolation, the more important thing is that significantly reduce computing time.Apply the present invention in the system of three-dimensional reconstruction and demonstration, the image boundary that drafting is come out is clear, and the terraced fields effect is reduced to minimum, more realistic demand, thus improve accuracy and the work efficiency that the doctor judges in a big way.

Description of drawings

Accompanying drawing 1 is the self-adaptation medical image sequence interpolation method system diagram based on area-of-interest of the embodiment of the invention one;

Accompanying drawing 2 is FB(flow block) of interpolation method part among the embodiment one;

Accompanying drawing 3 is comparison diagrams of the image that goes out of two Zhang Yuan's image files and interpolation;

Accompanying drawing 4 is to organize intensity profile figure in the human body head;

Accompanying drawing 5 is to organize intensity profile figure in the human body basin bone;

Accompanying drawing 6 is that synoptic diagram is selected in the zone after improving;

Accompanying drawing 7 is to carry out the synoptic diagram that match point is determined based on the zone of selecting among the embodiment one.

Embodiment

Technical scheme for a better understanding of the present invention is further described the present invention below in conjunction with drawings and Examples:

Embodiment: referring to accompanying drawing 1, for the present invention proposes a kind of self-adaptation medical image sequence interpolation method overall system frame diagram based on area-of-interest, data file (picture file) is CT or the MRI picture that meets the DICOM form.

[1] obtaining according to the requirement of drawing of source data read the multilayer tomography picture file that is absorbed by hardware device CT or nuclear magnetic resonance etc. by the file read module, and obtained their performance data by interface function.The number of getting the tomography picture in this example is 227, and the resolution of individual picture is that 512 * 512, two distances between the adjacent picture are 2.8mm, and the needed patient information of doctor.At last the tomography picture is carried out necessary pre-service work, comprise level and smooth and denoising etc.;

[2] determine to obtain 227 layers of tomography picture altogether in this example of position of source picture, nethermost one deck as the 1st layer, and the position is decided to be Z=0 with this layer picture on the Z axle, upwards by that analogy.Obtain nethermost two-layer picture earlier, carry out interpolation operation between layers 1 and 2, order upwards again;

[3] position of determining interpolation point obtains in [2] after the two-layer adjacent picture, seeks the position at ground floor interpolation picture place again between this two-layer adjacent picture, promptly with I _kApart from d ₁With with I _K+1Apart from d ₂Be in the middle of two-layer tomography picture, to begin interpolation, d when promptly beginning among the present invention ₁=d ₂=Δ z/2, other picture of interpolation up and down respectively again.

[4] interpolation operation utilize loop program design obtain interpolation point V (x, y, position d), based on the corresponding point in the levels picture (or match point to) carried out the interpolation of interlayer pixel again:

[4-1] if the processing of marginal point utilize loop program design obtain interpolation point V (x, y d) be positioned at around the picture (x, y=0, N-1), then directly employing based on about the right gray-scale value f of corresponding point _Ijk, f _{Ij (k+1)}Carry out linear interpolation, otherwise go to [4-2];

Linear interpolation formula: f _Ijd=r f _Ijk+ (1-r) f _{Ij (k+1)}, (r=d ₁/ (d ₁+ d ₂).Among the present invention with linear interpolation formula: f _A=rf ₁+ (1-r) f ₂, (r=d ₁/ (d ₁+ d ₂) change into: f _A=f ₂+ r (f ₁-f ₂), all the other do same operation, and the multiplication number of times reduces to original half like this, has significantly reduced calculated amount.

[4-2] if it is not point around the picture that interpolation point is obtained in the processing of non-marginal point 1, then utilize levels corresponding point V (i, j, k) and V (i, j, gray-scale value f k+1) _Ijk, f _{Ij (k+1)}And difference | f _Ijk-f _{Ij (k+1)}|, according to organizing correlation principle (shown in Figure 4) to judge whether interpolation point belongs to RONI,, then adopt based on its gray-scale value f if belong to RONI _Ijk, f _{Ij (k+1)}Carry out linear interpolation, otherwise forward [4-3] to;

Organize correlation principle: according to different tissues gray difference in the human body and different tissues intensity profile figure, interpolation voxel V (i judges according to gray value profiles (referring to accompanying drawing 4, accompanying drawing 5) earlier in system of the present invention before carrying out correlation calculations, j, d) belong to RONI (as air), still belong to ROI (as tissue).When belonging to ROI, to judge also up and down in the faultage image (i, j) whether two of place zones belong to identical tissue to pixel.

The processing 2 of [4-3] non-marginal point respectively with levels corresponding point V (i, j, k) and V (i, j are the center k+1), choose two regional A _k, and A _K+1, the zone be shaped as shown in Figure 6 in a kind of, calculate the coefficient of autocorrelation of organizing in these two zones, if related coefficient is greater than a certain threshold value ρ 0, illustrate that two regional integrations are more similar, then adopt based on these two regional A _k, and A _K+1Carry out Tri linear interpolation, otherwise forward [4-4] to;

Organize the coefficient of autocorrelation computing formula:

$\mathrm{\&rho;}(A_k,A_{k+1})=\frac{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}(f_{\mathrm{ijk}}-\stackrel{\&OverBar;}{f_k})(f_{\mathrm{ij}(k+1)}-\stackrel{\&OverBar;}{f_{(k+1)}})}{\sqrt{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{(f_{\mathrm{ijk}}-\stackrel{\&OverBar;}{f_k})}^2}\sqrt{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}}{(f_{\mathrm{ij}(k+1)}-\stackrel{\&OverBar;}{f_{(k+1)}})}^2},$

Variable is explained as follows in detail in the formula:

(1) A _k, and A _K+1For respectively with levels corresponding point V (i, j, k) and V (i, j, k+1) zone of choosing for the center (as Fig. 5);

(2) f _Ijk, f _{Ij (k+1)}Be a V (i, j, k) and V (i, j, gray-scale value k+1), f _k, f _(k+1)Be average gray,

(3) N is the number of above-mentioned institute favored area interior pixel;

The selection in zone: because for medical picture, handled generally all is some tissue of human body or diseased region, and the outline line at these positions all is level and smooth, promptly general irregular circle of one-tenth or ellipse, institute thinks the better interpolation of acquisition, the present invention adopts the sub-circular zone of selecting two symmetries, as Fig. 5 (pixel of a box indicating).

Described Tri linear interpolation is: (as accompanying drawing 6) chooses V (x, y, z) (x=i-1, i+1 in the sub-circular zone of two symmetries up and down; Y=j-1, j+1; Z=k, k+1) 8 points are earlier by a V (x, j-1, and z) (x=i-1, gray-scale value linear interpolation i+1) goes out a V (i, j-1, z) (z=k, gray-scale value k+1), by a V (x, j+1, z) (x=i-1, i+1) gray-scale value linear interpolation goes out a V (i, j+1, z) (z=k, k+1) gray-scale value is then by a V (i, y, z) (y=j-1, gray-scale value linear interpolation j+1) goes out a V (i, j, z) (z=k, gray-scale value k+1) is at last by a V (i, j, and z) (z=k, gray-scale value linear interpolation k+1) goes out a V (i, j, gray-scale value d) promptly carries out the cubic curve interpolation.

The processing 3 of [4-4] non-marginal point is at levels corresponding region A _k, and A _K+1In to seek optimal match point right, utilize the right gray-scale value f of corresponding point _Ijk, f _{Ij (k+1)}Carry out linear interpolation.Match point synoptic diagram (as shown in Figure 6).Optimal match point is to satisfying 4 points: match point should be close to gray-scale value; Match point should be close to the rate of gray level of point around reaching; Match point should be close to the grey scale change direction of point around reaching; The right distance of match point should be shorter.Therefore the optimal match point formula that adopts among the present invention is:

$R=\underset{0\&le;i\&le;N-1}{\mathrm{<figure-callout\; id="0"\; label="min"\; filenames="C2007100238720017H2.png,C2007100238720017H3.png"\; state="\{\{state\}\}">min</figure-callout>}}\left\{R(V_{\mathrm{ki}},V_{(k+1)i})\right\}$

$=\underset{0\&le;i\&le;N-1}{\mathrm{<figure-callout\; id="0"\; label="min"\; filenames="C2007100238720017H2.png,C2007100238720017H3.png"\; state="\{\{state\}\}">min</figure-callout>}}\{Q(V_{\mathrm{ki}},V_{(k+1)i})\&CenterDot;E(V_{\mathrm{ki}},V_{(k+1)i})\}$

Wherein

Q(V _ki，V _(k+1)i)＝α·Δf+β·ΔK+γ·Δθ

＝α·|f _ki-f _(K+1)i|+β·|K _ki-K _(k+1)i|+γ·Δθ _i

$E(V_{\mathrm{ki}},V_{(k+1)i})=\frac{1}{\mathrm{\&lambda;}}\&CenterDot;\exp \left(\mathrm{\&lambda;}\sqrt{[{(x_{\mathrm{ki}}-x_{(k+1)i})}^2+{(y_{\mathrm{ki}}-y_{(k+1)i})}^2]}\right)$

(i＝0，1，...，N-1)

Argument is explained as follows in detail in the formula:

(1) R represents to weigh the generic function of two some matching degrees up and down;

(2) V _KiAnd V _{(k+1) i}Be illustrated respectively in two favored area A of institute up and down _k, and A _K+1In the i that compares right to putting, f _KiAnd f _{(k+1) i}Be its gray-scale value, K _KiAnd K _{(k+1) i}Expression point V _KiAnd V _{(k+1) i}The maximum slope at some place, Δ θ _iRepresent that aforementioned two maximum slopes form the angle between the space line;

(3) Q (V _Ki, V _{(k+1) i}) first three puts the function of matching degree described in two points [4-4] up and down in the expression measurement;

(4) E (V _Ki, V _{(k+1) i}) expression weighs the 4th point described in two points [4-4] up and down: the function that the right distance of match point should be less, i.e. exponential distribution function.x _KiAnd x _{(k+1) i}, y _KiAnd y _{(k+1) i}Represent some V respectively _KiAnd V _{(k+1) i}Horizontal ordinate and ordinate;

(5) α, beta, gamma, λ are weighting function, can regulate the influence of each component to generic function;

Parameter alpha in the native system example=8, β=1, γ=0.5, λ=1.In the technical scheme, the calculated amount of seeking optimal match point is bigger, can not satisfy the real-time that present doctor requires preferably, thereby, be faster and better acquisition interpolation, in realizing, system adopts difference method.

[5] judge that the present invention utilizes the loop program design to obtain interpolation point V (x, y, position d), based on the corresponding point in the levels picture (or match point to) carried out the interpolation of interlayer pixel again, by x=0, y=0 place beginning interpolation, up to x=N-1, y=N-1 stops.Judge whether that interpolation finishes behind the intact point of interpolation,, go to [3] if do not have, otherwise go to [6];

[6] layer judges among the present invention it is to begin interpolation, d when promptly beginning in the middle of two-layer tomography picture ₁=d ₂=Δ z/2, after the middle layer interpolation is finished, other picture of interpolation up and down respectively again.At first take upwards (agreeing also can be downward earlier), interpolation goes out the picture of other layers, in the middle layer to after all interpolation finishes between last layer, again by the downward interpolation in middle layer, judge whether at last all layers between two adjacent source pictures to be carried out interpolation, if no, go to [3], otherwise go to [6];

[7] space judges that among the present invention be to obtain nethermost two-layer picture earlier, carries out interpolation operation between layers 1 and 2, and order upwards again; After interpolation finishes between the layers 1 and 2, carry out the interpolation operation between the layers 2 and 3 again, by that analogy.Judge whether that at last all layers carry out interpolation between any two adjacent source pictures,, go to [2] if do not have, otherwise go to [7];

[8] carrying out three-dimensional body draws and demonstration.Such three-dimensional body is that auxiliary medical personnel carries out qualitative basis until accurate quantitative analysis to pathology body and other interesting areas, on this basis, just help medical analysis, diagnosis and the calculating of the 3-dimensional dose field in radiotherapy treatment planning and the measurement of diseased region etc.

In the above-mentioned embodiment, I _kAnd I _K+1Be adjacent k layer and k+1 tomographic image, I _dBe the interpolation picture, this paper I _kRepresent the k tomographic image, (x, y k) are positioned at (x, pixel y), f in the expression k tomographic image to V _IjkRepresent to be positioned in the k tomographic image (i, the gray-scale value of pixel j), Δ z and Δ are respectively the distance between any two neighbors in a distance between adjacent tomography picture and the pictures, V (i, j, k) and V (i, j k+1) are and interpolation point x the point that the y value equates, f _Ijk, f _{Ij (k+1)}Be a V (i, j, k) and V (i, j, gray-scale value k+1), A _k, A _K+1Be with these 2 be the pending zone at center, N is the number of regional interior pixel.

Claims (3)

1. self-adaptation medical image sequence interpolation method based on area-of-interest comprises the following steps: that (1) obtains the source data line data pre-service of going forward side by side, and need between every pair of adjacent source picture to determine the number of plies of interpolation; (2) choose a pair of pending adjacent source picture; (3) determine distance between interpolation layer and both sides adjacent source picture; (4) obtain the position of interpolation point, carry out the interpolation operation of interlayer pixel; (5) if untreated interpolation point, repeating step (4) are arranged in the interpolation layer; (6) if untreated interpolation layer is arranged between pending adjacent source picture, repeating step (3) is to (5); (7) if untreated adjacent source picture is arranged, repeating step (2) is realized the interpolation of image thus to (6); It is characterized in that:

The interpolation operation of described step (4) is meant, the position that utilizes the loop program design to obtain interpolation point, again based on the corresponding point in the levels picture to or match point to carrying out the interpolation of interlayer pixel, interpolation method is:

If a) the interpolation point is positioned at around the picture, belong to marginal point, adopt based on the right gray-scale value of corresponding point up and down and carry out linear interpolation, otherwise execution in step b);

B) utilize the gray-scale value and the difference thereof of levels corresponding point,,, then carry out linear interpolation based on its gray-scale value if do not belong to area-of-interest according to organizing correlation principle to judge whether interpolation point belongs to area-of-interest, otherwise execution in step c);

C) be the center with the levels corresponding point respectively, choose two zones, calculate the coefficient of autocorrelation of organizing in these two zones, if related coefficient is greater than setting threshold, illustrate that two regional integrations are more similar, then carry out Tri linear interpolation based on these two zones, otherwise execution in step d);

D) it is right to seek optimal match point in the levels corresponding region, utilizes the right gray-scale value of corresponding point to carry out linear interpolation.

2. the self-adaptation medical image sequence interpolation method based on area-of-interest according to claim 1, it is characterized in that: in the described step b), judge according to organizing correlation principle whether interpolation point belongs to area-of-interest and be meant, at first provide different tissues intensity profile figure, utilize the gray-scale value of interpolation point levels corresponding point and difference thereof to judge that according to gray value profiles the interpolation point belongs to the zone of loseing interest in and still belongs to area-of-interest according to different tissues gray difference in the human body.

3. the self-adaptation medical image sequence interpolation method based on area-of-interest according to claim 1 is characterized in that: in the described step c), respectively with levels corresponding point V (i, j, k) and V (i, j are the center k+1), choose two regional A _k, and A _K+1, calculate the coefficient of autocorrelation of organizing in these two zones

$\mathrm{\&rho;}(A_k,A_{k+1})=\frac{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}(f_{\mathrm{ijk}}-\stackrel{\&OverBar;}{f_k})(f_{\mathrm{ij}(k+1)}-\stackrel{\&OverBar;}{f_{(k+1)}})}{\sqrt{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{(f_{\mathrm{ijk}}-\stackrel{\&OverBar;}{f_k})}^2}\sqrt{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{(f_{\mathrm{ij}(k+1)}-\stackrel{\&OverBar;}{f_{(k+1)}})}^2}},$

In the formula, A _kAnd A _K+1For respectively with levels corresponding point V (i, j, k) and V (i, j, the k+1) zone of choosing for the center, f _Ijk, f _{Ij (k+1)}Be a V (i, j, k) and V (i, j, gray-scale value k+1), f _k, f _(k+1)Be average gray, N is the number of above-mentioned institute favored area interior pixel; If related coefficient greater than preset threshold ρ 0, illustrates that two regional integrations are more similar, belong to identical tissue, then based on these two regional A _k, and A _K+1Carry out Tri linear interpolation.

Images