CN103366394B - The Direct volume rendering of medical volume data feature abstraction - Google Patents

Abstract

The Direct volume rendering of medical volume data feature abstraction, comprises following five steps: 1) import medical volume data; 2) utilize light projecting algorithm, along radiation direction, volume data is sampled; 3) by abstract for the characteristic area that extracts be sampled point, then calculate the opacity of abstract sampled point according to the abstract sampling number gauge on sampling light; 4) scalar value and the gradient information of sampled point in each characteristic area is analyzed respectively, find the sampled point that scalar value is maximum, its physical location is as the physical location of abstract sampled point, its scalar value is as the scalar value of abstract sampled point, the gradient information simultaneously finding feature inside gradient modulus value maximum is as the normal vector of abstract sampled point, finally, local illumination calculation is also performed for abstract sampled point arranges color; 5) abstract sampled point is merged, form final medical image.

Description

The Direct volume rendering of medical volume data feature abstraction

Technical field

The Direct Volume Rendering Techniques that system of the present invention shows medical volume data characteristic information.

Background technology

Volume visualization technology is widely used in the fields such as numerical simulation in scientific and engineering, medical conditions diagnosis and Atmosphere and Ocean simulation.In volume visualization application, be how the image that details is abundant by the Data Representation of acquisition, how for user provides internal information more accurately more directly perceived to be vital.Along with the development of graphics hardware and volume visualization algorithm, Image Rendering speed goes is fast, and the precision of image is also more and more higher, draws style and also presents variation.Therefore, this technology is more applied in life now, also receives people and more and more pays close attention to.

A kind of important method in volume visualization technology is Direct Volume Rendering Techniques, more traditional iso-surface patch, direct volume drawing is by definition transition function associated with data set, be color and opacity by data-mapping, effectively can show the internal information of 3 d data field, therefore be widely used, become a study hotspot in current volume visualization field.In order to the characteristic information in the effective visualization of 3 d data fields of energy, classify to volume data and draw.Volume data classification has been designed by transition function usually, and drafting is process three-dimensional data being mapped to two dimensional image, and light projecting algorithm becomes the main method of direct volume drawing.

These data transformations, by scalar value is mapped to color value and opacity value, are then figure by one dimension transition function, produce visualization result.But, in the research field of medical data volume drawing, because the identical scalar value of different tissues represents by many medical data collection, the tissue that we are concerned about block by some unessential tissues, this block be difficult to solved by the method for Designing Transfer Function.Moreover design transition function is a very loaded down with trivial details job consuming time, usually can not meet the needs of clinical practice.Therefore, how by characteristic information numerous in three-dimensional data fast and display clearly and become a major issue urgently to be resolved hurrily.

Summary of the invention

In order to overcome the transition function difficult design of existing medical image acquisition technology and regulate shortcoming consuming time, solve the occlusion issue between three-dimensional data different tissues, the invention provides the three-dimensional data object plotting method that the abstract rapidity of a kind of feature based is good, feature is clear, effect is bright.

The technical solution adopted for the present invention to solve the technical problems is: the Direct volume rendering of medical volume data feature abstraction, and described feature abstraction method comprises following five steps:

The Direct volume rendering of medical volume data feature abstraction, described feature abstraction method comprises following five steps:

1) medical volume data is imported;

2) utilize light projecting algorithm, along radiation direction, volume data is sampled.Moving Least is used to reconstruct a scalar curve according to the scalar value on sampling light, then signature analysis is carried out to scalar curve, second derivative is asked to scalar curve, find out all maximum value and minimal value, whether local extremum interval is effective characteristic area then to use the two-valued function based on gradient-norm to judge.Two-valued function based on gradient-norm is as follows:

$\mathrm{\δ}\left(i\right)=\left\{\begin{array}{cc}1,& \left|\right|{\▿f}_j\left|\right|>t,j\∈[{l\min }_i,{l\max }_i]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

Wherein i is the interval (hereinafter referred to as between partial zones) between i-th local minimum and local maximum following closely that sampling light searches out, and j is [lmin between partial zones _i, lmax _i] in sampled point, ▽ f _jfor the gradient information of sampled point j, t is user-defined threshold value, as interval [lmin _i, lmax _i] there is gradient modulus value when being greater than the sampled point of t, δ (i) is 1, is namely a validity feature between this partial zones, otherwise is 0.

3) by abstract for the characteristic area that extracts be sampled point, then calculate the opacity of abstract sampled point according to the abstract sampling number gauge on sampling light.Opacity depending on all abstract sampled points is identical, is set to α, if the feature that a sampling light meets the demands, namely the number of abstract sampled point is S _f, then can obtain amended volume rendering integral formula:

$C\left(r\right)=\underset{k=1}{\overset{S_f}{\mathrm{\Σ}}}c\left(s\left(k\right)\right)\mathrm{\α}{(1-\mathrm{\α})}^{k-1}---\left(2\right)$

Wherein, C (r) is the color value of sampling light r respective pixel; S (k) is the scalar value of a kth abstract sampled point on sampling light, uses the maximum scalar value between partial zones herein; C (s (k)) represents that the scalar value s (k) of a kth abstract sampled point maps the color value obtained by transition function.

Visibility V (s (p)) then apart from viewpoint abstract sampled point p farthest can be expressed as:

$V\left(s\left(p\right)\right)=\mathrm{\α}{(1-\mathrm{\α})}^{S_f-1}---\left(3\right)$

In order to make this abstract sampled point have maximum visual degree, making function get maximal value, by function differentiate, and making the functional value after differentiate equal 0, calculating maximal value finally can obtain the volume rendering integral equation based on local interval analysis:

$C\left(r\right)=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}c\left(s^{\′}\left(i\right)\right)\frac{\mathrm{\δ}\left(i\right)}{S_f}\underset{j=1}{\overset{i-1}{\mathrm{\Π}}}(1-\frac{\mathrm{\δ}\left(i\right)}{S_f})---\left(4\right)$

Wherein, m is between all partial zones of sampling light, s ' (i) is the scalar value between partial zones, uses the maximum value between partial zones, and c (s ' (i)) maps the color value obtained for scalar value s ' (i) by transition function.

Analyze scalar value and the gradient information of sampled point in each characteristic area respectively, find the sampled point that scalar value is maximum, its physical location is as the physical location of abstract sampled point, its scalar value is as the scalar value of abstract sampled point, and the gradient information simultaneously finding feature inside gradient modulus value maximum is as the normal vector of abstract sampled point; Finally, local illumination calculation is also performed for abstract sampled point arranges color;

4) use hsv color model, the degree of depth of feature and form and aspect H are mapped, the depth perception of Enhanced feature.In order to strengthen the interested or important feature of user, introduce importance functions:

f(S _i,G _i)=(1-λ)·S _i+λ·G _i(5)

Wherein, S _irepresent the scalar value of i-th abstract sampled point, G _irepresent the gradient modulus value of i-th abstract sampled point, and scalar value and gradient modulus value be all the value after normalization, λ is user-defined weighting parameters, is used for weighing scalar value and gradient modulus value to threshold function table f (S _i, G _i) contribution amount.Therefore all abstract sampled points importance degree and be:

$F(S,G)=\underset{i=1}{\overset{S_f}{\mathrm{\Σ}}}f(S_i,G_i)---\left(6\right)$

And then the contribution degree of a certain feature in final pixel can be obtained be: if only consider the abstract sampled point on sampling light, then can obtain according to formula (4):

$C\left(r\right)=\underset{i=1}{\overset{S_f}{\mathrm{\Σ}}}c\left(s^{\′}\left(i\right)\right)\frac{1}{S_f}\underset{j=1}{\overset{i-1}{\mathrm{\Π}}}(1-\frac{1}{S_f})---\left(7\right)$

Therefore the integral formula based on abstract sampled point importance functions is:

$C\left(r\right)=\underset{i=1}{\overset{S_f}{\mathrm{\Σ}}}c\left(s^{\′}\left(i\right)\right)w_i\underset{j=1}{\overset{i-1}{\mathrm{\Π}}}(1-w_j)---\left(8\right)$

5) abstract sampled point is merged, form final medical image.

Technical conceive of the present invention is: by the characteristic area on analytical sampling light, extracts each feature, and by they abstract be sampled point; Calculate the opacity of abstract sampled point according to the abstract sampling number gauge on sampling light, make the visibility of distance viewpoint abstract sampled point farthest maximum, then derivation volume rendering integral equation, obtains the volume rendering integral equation based on abstract sampled point; Greatest gradient modulus value in feature is found as the normal vector of abstract sampled point, to calculate local light photograph, drawing image.

The invention has the beneficial effects as follows: analysiss and abstract is carried out to the characteristic area of sampling light, make the transition function not needing adjustment complexity, can not only characteristic information effectively in display body data, all features met the demands on direction of visual lines can also be shown.Both removed from and repeatedly revised the loaded down with trivial details of transport function, obtain again drafting effect and become clear, feature is result clearly.

Accompanying drawing explanation

Fig. 1 is feature abstraction Direct volume rendering system global structure figure.

Fig. 2 is the program outline flowchart of feature abstraction object plotting method.

Fig. 3 is data sampling and fitted figure.

Fig. 4 is characteristic area extraction figure.

Fig. 5 is scalar value and the normal direction spirogram of feature abstraction sampled point.

Embodiment

Below in conjunction with accompanying drawing 1-5, the invention will be further described.

The Direct volume rendering of the feature abstraction that the present invention proposes, carries out analysis to volume data and extracts feature, calculate the opacity of abstract sampled point, the characteristic information in effective display body data.Method mainly comprises following five steps:

1) medical volume data is imported, as CT, MRI data.In Fig. 1, data importing part uses C Plus Plus to realize, and feature abstraction algorithm part utilizes GLSL language compilation, and whole program realizes on VC platform.In Fig. 2, the step of the Direct volume rendering of feature abstraction has: first import volume data, then samples along radiation direction, simulates level and smooth scalar curve according to sampled point.By finding out greatest gradient modulus value and maximum scalar value determines abstract sampled point in each characteristic area.Finally add illumination, carry out feature enhancing, draw out final image.

2) feature extraction and statistics.According to the discrete scalar value on sampling light, Moving Least is adopted to reconstruct a scalar curve.On scalar curve, multiple crest and the trough light that shows to sample have passed through multiple characteristic area, as shown in Figure 3.Wherein, minimum point, i.e. trough, the often separation of different characteristic.Owing to there is the interference of low-and high-frequency noise in volume data, and also may there is fluctuation in the scalar value in same characteristic area, so not every minimum point is all the separation of feature.This patent is introduced gradient-norm and is differentiated feature separation, extracts the boundary information of feature.Specifically, find extreme point by the method for the first order derivative and second derivative of asking scalar curve, between a certain minimum point and maximum point following closely, then find the sampled point that whether there is gradient modulus value and be greater than the threshold value that certain is set by the user.If exist, illustrating that this region scalar value change is very fast, is the borderline region of an effective feature.For the ease of analytic statistics characteristic area, we use the two-valued function based on gradient-norm:

$\mathrm{\δ}\left(i\right)=\left\{\begin{array}{cc}1,& \left|\right|{\▿f}_j\left|\right|>t,j\∈[{l\min }_i,{l\max }_i]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

Wherein i is i-th local minimum and local maximum interval (hereinafter referred to as between partial zones) that sampling light searches out, and j is [lmin between partial zones _i, lmax _i] in sampled point, ▽ f _jfor the gradient information of sampled point j, t is user-defined threshold value, as interval [lmin _i, lmax _i] there is gradient modulus value when being greater than the sampled point of t, δ (i) is 1, otherwise is 0.According to this two-valued function, we can find all features that light satisfies condition.As shown in Figure 4, scalar curve extracts ABCDE five characteristic areas.

3) opacity of abstract sampled point and illumination.We are sampled point feature abstraction, are referred to as abstract sampled point.In order to make all abstract sampled points visual, the opacity depending on all abstract sampled points is identical, is set to α, if the feature that a sampling light meets the demands, namely the number of abstract sampled point is S _f, then can obtain amended volume rendering integral formula:

$C\left(r\right)=\underset{k=1}{\overset{S_f}{\mathrm{\Σ}}}c\left(s\left(k\right)\right)\mathrm{\α}{(1-\mathrm{\α})}^{k-1}---\left(2\right)$

Wherein, C (r) is the color value of sampling light r respective pixel; S (k) is the scalar value of a kth abstract sampled point on sampling light, uses the maximum scalar value between partial zones herein; C (s (k)) represents that the scalar value s (k) of a kth abstract sampled point maps the color value obtained by transition function.

Easily know that the visibility V (s (p)) apart from viewpoint abstract sampled point p farthest can be expressed as thus:

$V\left(s\left(p\right)\right)=\mathrm{\α}{(1-\mathrm{\α})}^{S_f-1}---\left(3\right)$

In order to make this abstract sampled point have maximum visual degree, making function get maximal value, by function differentiate, and making the functional value after differentiate equal 0, calculating maximal value finally can obtain the volume rendering integral equation based on local interval analysis:

$C\left(r\right)=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}c\left(s^{\′}\left(i\right)\right)\frac{\mathrm{\δ}\left(i\right)}{S_f}\underset{j=1}{\overset{i-1}{\mathrm{\Π}}}(1-\frac{\mathrm{\δ}\left(i\right)}{S_f})---\left(4\right)$

Wherein, m is that between sampling light all partial zones, s ' (i) is the scalar value between partial zones, and c (s ' (i)) maps the color value obtained for scalar value s ' (i) by transition function.

What abstract sampled point represented is the characteristic area comprising multiple sampled point, containing numerous different gradient information, calculate the illumination of certain abstract sampled point, need in this characteristic area, find the normal vector of gradient as this abstract sampled point that can represent its characteristic information.Because on scalar curve, the maximum of points of characteristic area often can show characteristic information, in accumulation, therefore get the scalar value of maximum scalar value point as abstract sampled point of each characteristic area, simultaneously using this physical location as abstract sampled point.In order to the image making drafting is more level and smooth, between the minimum point and maximum point following closely of characteristic area, find the sampled point that gradient modulus value is maximum, its gradient as the normal direction of this abstract sampled point, for calculating illumination.As shown in Figure 5.Then, adopt local illumination model conventional in direct volume drawing: Blinn-Phong illumination model, calculate illumination, the shape perception of each feature of reinforcement data, strengthen the sense of reality of drawing image.

4) feature strengthens drafting.Although illumination can the shape perception of Enhanced feature, but strengthen not obvious to depth perception, in order to increase the depth information of feature, this patent introduces color mapping techniques, use hsv color model, make the degree of depth and the color map of feature, allow the feature of different depth use different colors to represent, thus make feature have space hierarchy, the depth perception of Enhanced feature.Hsv color model is intuitive nature according to color and a kind of color space created, and is one color model intuitively for user.Wherein form and aspect H, measure by angle, span is 0 ° ~ 360 °, by counterclockwise calculating from redness, can be easy to associate with depth value, along with the increase of the degree of depth, H value gradually changes, the color change gradually of sampled point, therefore, user can tell the degree of depth sequence of each feature very easily.

Meanwhile, drawing the interested feature of user in volume data to strengthen, introducing importance functions:

f(S _i,G _i)=(1-λ)·S _i+λ·G _i（5）

Wherein, S _irepresent the scalar value of i-th abstract sampled point, G _irepresent the gradient modulus value of i-th abstract sampled point, and scalar value and gradient modulus value be all the value after normalization, λ is user-defined weighting parameters, is used for weighing scalar value and gradient modulus value to threshold function table f (S _i, G _i) contribution amount, as when λ=0, just employ maximum scalar value in feature to weigh its importance degree in volume data.Therefore all abstract sampled points importance degree and be:

$F(S,G)=\underset{i=1}{\overset{S_f}{\mathrm{\Σ}}}f(S_i,G_i)---\left(6\right)$

And then the contribution degree of a certain feature in final pixel can be obtained be: if only consider the abstract sampled point on sampling light, then can obtain according to formula (4):

$C\left(r\right)=\underset{i=1}{\overset{S_f}{\mathrm{\Σ}}}c\left(s^{\′}\left(i\right)\right)\frac{1}{S_f}\underset{j=1}{\overset{i-1}{\mathrm{\Π}}}(1-\frac{1}{S_f})---\left(7\right)$

Therefore the integral formula based on abstract sampled point importance functions is:

$C\left(r\right)=\underset{i=1}{\overset{S_f}{\mathrm{\Σ}}}c\left(s^{\′}\left(i\right)\right)w_i\underset{j=1}{\overset{i-1}{\mathrm{\Π}}}(1-w_j)---\left(8\right)$

5) according to light projecting algorithm, by the information fusion of each abstract sampled point, final medical image is formed.

Claims (1)

1. the Direct volume rendering of medical volume data feature abstraction, described feature abstraction method comprises following five steps:

1) medical volume data is imported;

2) utilize light projecting algorithm, along radiation direction, volume data is sampled; Moving Least is used to reconstruct a scalar curve according to the scalar value on sampling light, then signature analysis is carried out to scalar curve, second derivative is asked to scalar curve, find out all maximum value and minimal value, whether local extremum interval is effective characteristic area then to use the two-valued function based on gradient-norm to judge; Two-valued function based on gradient-norm is as follows:

$\mathrm{\δ}\left(i\right)=\left\{\begin{array}{cc}1,& \left|\right|\▿f_j\left|\right|>t,j\∈\[{\mathrm{lmin}}_i,{\mathrm{lmax}}_i\]\\ 0,& otherwise\end{array}\right.---\left(1\right)$

Wherein i is the interval between i-th local minimum and local maximum following closely that sampling light searches out, hereinafter referred to as between partial zones; J is [lmin between partial zones _i, lmax _i] in sampled point, for the gradient information of sampled point j, t is user-defined threshold value, as interval [lmin _i, lmax _i] there is gradient modulus value when being greater than the sampled point of t, δ (i) is 1, is namely a validity feature between this partial zones, otherwise is 0

3) by abstract for the characteristic area that extracts be sampled point, then the opacity that the opacity calculating abstract sampled point according to the abstract sampling number gauge on sampling light looks all abstract sampled point is identical, be set to α, feature light met the demands if one sample, namely the number of abstract sampled point is S _f, then can obtain amended volume rendering integral formula:

$C\left(r\right)=\underset{k=1}{\overset{S_f}{\Σ}}c\left(s\right(k\left)\right)\mathrm{\α}{(1-\mathrm{\α})}^{k-1}---\left(2\right)$

Wherein, C (r) is the color value of sampling light r respective pixel; S (k) is the scalar value of a kth abstract sampled point on sampling light, uses the maximum scalar value between partial zones herein; C (s (k)) represents that the scalar value s (k) of a kth abstract sampled point maps the color value obtained by transition function;

Visibility V (s (p)) then apart from viewpoint abstract sampled point p farthest can be expressed as:

$V\left(s\right(p\left)\right)=\mathrm{\α}{(1-\mathrm{\α})}^{S_f-1}---\left(3\right)$

In order to make this abstract sampled point have maximum visual degree, making function get maximal value, by function differentiate, and making the functional value after differentiate equal 0, calculating maximal value finally can obtain the volume rendering integral equation based on local interval analysis:

$C\left(r\right)=\underset{i=1}{\overset{m}{\Σ}}c\left(s^{\′}\right(i\left)\right)\frac{\mathrm{\δ}\left(i\right)}{S_f}\underset{j=1}{\overset{i-1}{\Π}}(1-\frac{\mathrm{\δ}\left(i\right)}{S_f})---\left(4\right)$

Wherein, m is between all partial zones of sampling light, s'(i) be the scalar value between partial zones, use the maximum value between partial zones, c (s'(i)) be scalar value s'(i) color value obtained is mapped by transition function

Analyze scalar value and the gradient information of sampled point in each characteristic area respectively, find the sampled point that scalar value is maximum, its physical location is as the physical location of abstract sampled point, its scalar value is as the scalar value of abstract sampled point, and the gradient information simultaneously finding feature inside gradient modulus value maximum is as the normal vector of abstract sampled point; Finally, local illumination calculation is also performed for abstract sampled point arranges color;

4) use hsv color model, the degree of depth of feature and form and aspect H are mapped, the depth perception of Enhanced feature; In order to strengthen the interested or important feature of user, introduce importance functions:

f(S _i,G _i)＝(1-λ)·S _i+λ·G _i(5)

Wherein, S _irepresent the scalar value of i-th abstract sampled point, G _irepresent the gradient modulus value of i-th abstract sampled point, and scalar value and gradient modulus value be all the value after normalization, λ is user-defined weighting parameters, is used for weighing scalar value and gradient modulus value to threshold function table f (S _i, G _i) contribution amount therefore all abstract sampled points importance degree and be:

$F(S,G)=\underset{i=1}{\overset{S_f}{\Σ}}f(S_i,G_i)---\left(6\right)$

And then the contribution degree of a certain feature in final pixel can be obtained be: if only consider the abstract sampled point on sampling light, then can obtain according to formula (4):

$C\left(r\right)=\underset{i=1}{\overset{S_f}{\Σ}}c\left(s^{\′}\right(i\left)\right)\frac{1}{S_f}\underset{j=1}{\overset{i-1}{\Π}}(1-\frac{1}{S_f})---\left(7\right)$

Therefore the integral formula based on abstract sampled point importance functions is:

$C\left(r\right)=\underset{i=1}{\overset{S_f}{\Σ}}c\left(s^{\′}\right(i\left)\right)w_i\underset{j=1}{\overset{i-1}{\Π}}(1-w_j)---\left(8\right)$

5) abstract sampled point is merged, form final medical image.