CN100448409C - Three-dimensional ultrasound cardiogram four-cavity section image automatic detection method - Google Patents

Abstract

The invention relates to a method for realizing the three-dimension ultrasonic cardiogram four-chamber section image automatic check, which comprises that 1, extracting one three-dimension data from the end-diastolic phase of heart from the heart full-volume data; 2, extracting 180 section images from the three-dimension data to build an image database; 3, calculating out the similarity between each image of database and the template image, to be selected and using wavelet system statistic character to check out the image with highest similarity in the left image, to record the position in the data as the four-chamber watch position. The invention realizes the automatic search function of four-chamber section; therefore, medical staff can use real-time three-dimension ultrasonic technique to diagnose the patient.

Description

The method that three-dimensional ultrasound cardiogram four-cavity section image detects automatically

Technical field

What the present invention relates to is the method in a kind of Computer Applied Technology field, specifically is the method that a kind of three-dimensional ultrasound cardiogram four-cavity section image detects automatically.

Background technology

The real-time three-dimensional ultrasonic system is the main tool of current diagnosis of heart disease.It has realized observing the function of whole cardiac structure and motion, and can observe the situation of heart inside by selecting arbitrary tangent.At present, the doctor mainly rule of thumb in thtee dimensional echocardiography, manually seeks the distortion that two-dimentional tangent plane is observed heart inside one by one.A lot of bibliographical informations the importances of four chamber tangent planes for diagnosis of heart disease; For the clinical diagnosis of present Using Real-time Three-dimensional Echocardiography, the doctor at first seeks four chamber tangent planes to observe.But, from a large amount of three-D ultrasound datas, seek four chamber tangent planes one by one, for unskilled doctor some difficulties are arranged then, even for skilled doctor, also be a not only time-consuming but also loaded down with trivial details job.

Find by prior art documents, Raj Shekhar etc. are at " IEEE Transaction on MedicalImaging " (Institute of Electrical and Electric Engineers medical imaging journal) (2004, delivered (" Registration of Real-time 3-D Ultrasound Images of the Heart for Novel 3-DStress Echocardiography ") (" the ultrasonic registration of enhancing of heart real-time three-dimensional ultrasonoscopy ") literary composition 23:1141-1149), this article has proposed a kind of three-dimensional registration method based on mutual information, has realized the registration of the three-dimensional kinetocardiogram before and after the injection reinforcing agent.By the three-dimensional ultrasound pattern template of tangent plane position, known four chamber and the registration of any three-dimensional cardiac ultrasonic image, can realize the automatic detection of any four chamber tangent planes in theory.Yet above-mentioned all research work all are based on the registration of the three-dimensional kinetocardiogram data of same human heart, and can only realize the variation registration within the specific limits of two individual data items, tend to lose efficacy for changing two bigger individual data items.The amount of calculation of three-dimensional registration is big, and real-time is relatively poor.

Summary of the invention

The objective of the invention is at the deficiencies in the prior art, the method that a kind of three-dimensional ultrasound cardiogram four-cavity section image detects is automatically proposed, make the doctor to observe heart by four chamber tangent planes easily rapidly, and see the position of telling other important tangent planes fast on the basis in four chambeies.The real-time three-dimensional ultrasound data is positioned on the left and right sides apex of the heart, left and right sides parasternum, the breastbone and the inferior station acquisition of rib by probe usually.Probe positions difference, the data content of collection are also different.Wherein probe is positioned at the total volume data that the left apex of the heart is gathered, and can show whole heart and most of congenital heart disease deformity.The present invention is primarily aimed at this total volume data, carries out the automatic detection of four chamber tangent planes, is implemented in the fast automatic detecting that four chambeies are seen in this kind data of different people.

The present invention is achieved through the following technical solutions, and method step is as follows:

(1) because total volume data generally include tens frame continuous three-dimensional volume datas, and four chamber tangent planes are the most obvious in diastole feature in latter stage, therefore at first need to extract from the total volume data file of gathering diastole three-dimensional data constantly in latter stage.

(2) extract tangent plane the volume data constantly latter stage from diastole, set up the image data base that comprises four chamber tangent planes.Each volume data all comprises data area and background area.It is pyramid that the data area is wherein arranged, and other parts are the background area, and its gray value is 0; From the three-D ultrasound data that the left apex of the heart is gathered, the summit of its pyramid data area is generally the left apex of the heart or left apex of the heart near zone; Characteristics according to such volume data, suppose that summit, pyramidal data area is the apex of the heart, the straight line of crossing this summit and being parallel to rectangular height is a long axis of heart, long axis of heart with apex of the heart place is that rotary middle spindle obtains a series of tangent planes, every degree extracts piece image, each volume data extracts 180 width of cloth images altogether, sets up an image data base.

(3) choose secondary four chamber tangent planes in advance as template, the similarity of each width of cloth image and template image in the calculating data storehouse is chosen the highest image of similarity as four chamber tangent planes.At first according to the tangent plane picture among vertex position alignment template image and each width of cloth data base; Then, utilize mutual information measure to carry out elementary coupling, some images lower in the elimination image data base with template similarity based on gray scale; Utilize the wavelet coefficient statistical nature further to detect and the highest piece image of template image similarity in remaining image again, this image is considered to four chamber tangent plane pictures among this data base, writes down its position in this volume data and is sight position, four chambeies.

Described utilization is carried out elementary coupling based on the mutual information measure of gray scale, is specially: the public intersecting area of two images is divided into 2 * 2 parts, and the correlation coefficient of the entropy of each part is total

(with) estimate as two the similar of width of cloth image.Wherein the correlation coefficient computing formula of each part entropy is as follows:

$\mathrm{ECC}(A,B)=\frac{2I(A,B)}{H\left(A\right)+H\left(B\right)},$ (A, value B) is big more, represents that two width of cloth images are similar more for ECC.

Wherein: $H\left(A\right)=-\underset{i}{\mathrm{\Σ}}{p}_A\left(i\right)\log p_A\left(i\right),$

$H\left(B\right)=-\underset{j}{\mathrm{\Σ}}{p}_B\left(j\right)\log p_B\left(j\right),$

$H(A,B)=-\underset{i,j}{\mathrm{\Σ}}{p}_{\mathrm{AB}}(i,j)\log p_{\mathrm{AB}}(i,j),$

I(A，B)＝H(A)+H(B)-H(A，B)。

The described wavelet coefficient statistical nature that utilizes further detects and the highest piece image of template image similarity in remaining image, be specially: each width of cloth image and template image that alignment is remaining, the public part of asking two width of cloth images to intersect, the small echo horizontal direction high frequency coefficient feature of calculating intersection:

$E=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\left|x(m,n)\right|,$

$V=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}|x{(m,n)}^2-E^2|.$

Similarity is defined as: and Sim (A, B)=(E _A-E _B) ²+ (V _A-V _B) ², (A, B) the more little presentation video of value is similar more for Sim.(A, B) tangent plane of minima is the four chamber tangent planes the most similar to template to Sim, is four chamber tangent planes in this volume data.

Only can realize that by the conventional three-dimensional method for registering four chamber tangent planes detect in the volume data of the same sequence of same human heart, amount of calculation is big, and the present invention has realized the automatic searching of four chamber tangent planes in the thtee dimensional echocardiography that the left apex of the heart of different people gathers, amount of calculation is little, real-time is good, the diagnostic work that has greatly made things convenient for the doctor to utilize thtee dimensional echocardiography to carry out.Method among the present invention experimentizes to the volume data in 29 diastoles of 16 people latter stage, and the experiment accuracy reaches 97%.

Description of drawings

Fig. 1. the three-dimensional data sketch map that the left apex of the heart is gathered.

Fig. 2. volume data and tangent plane position view.

Fig. 3. tangent plane is at XOY face inner projection sketch map.

The specific embodiment

Below in conjunction with accompanying drawing one embodiment of the invention are elaborated: present embodiment is being to implement under the prerequisite with the technical solution of the present invention; provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.

Below do further detailed narration with the total volume data instance of arbitrary left apex of the heart collection:

(1) three-dimensional matrice (matrix) of PhilipsSonos7500 type real-time three-dimensional diasonograph probe is positioned at left apex of the heart collection Full-volume data, extract wherein the 7th frame diastole three-dimensional data constantly in latter stage, this volume data size is 144 * 160 * 208.

(2) Fig. 1 is the thtee dimensional echocardiography sketch map that the left apex of the heart is gathered, and each volume data portion comprises data area and background area, and wherein the data area is intermediary pyramid part, and other parts are the background area, and the background area gray value is 0.Three-D ultrasound data from left apex of the heart collection, the summit of its pyramid data area is generally the left apex of the heart or left apex of the heart near zone, according to the volume data own characteristic, with the summit of pyramid data area as zero, set up rectangular coordinate system as shown in Figure 2, Fig. 3 is tangent plane projection on the XOY face; With the Z axle is rotary middle spindle, and every degree (θ=0,1,2 ... 179) extract a width of cloth tangent plane, extract 180 width of cloth images altogether and set up an image data base.

The tangent plane extraction algorithm:

As zero, set up rectangular coordinate system with the pyramid summit, tangent plane is projected as straight line such as AB on the XOY face, and establishing AB and X-axis angle is θ (0≤θ＜180), and every degree is got a tangent plane, then tangental equation: y=tan (θ) x.On the XOY face, obtain A, B coordinate, can get the unfolded image width: $L=\stackrel{\‾}{\mathrm{AB}}=\sqrt{{(x2-x1)}^2+{(y2-y1)}^2}.$ Given θ and l (l for some P the subpoint of XOY face along AB to a some A (l≤L), arbitrfary point P satisfies on the tangent plane for x1, distance y1):

$\left\{\begin{array}{c}x=x1+l\cos \mathrm{\&theta;}\\ y=y1+l\sin \mathrm{\&theta;}\end{array}\right.,(0\&le;\mathrm{\&theta;}<180)$

Obtain the coordinate of size for being had a few on the arbitrarily angled tangent plane of L * n thus, wherein n is the height of volume data z direction.Extract 180 width of cloth tangent plane pictures altogether and form an image data base.

θ=0 tangent plane leaching process:

When θ=0, in the XOY plane, tangential equation y=0 asks the intersection point on tangent line and the volume data border on the XOY face, the A point coordinates $\left\{\begin{array}{c}x1=-80\\ y1=0\end{array}\right.,$ The B point coordinates $\left\{\begin{array}{c}x2=80\\ y2=0\end{array}\right.;$ Calculate $L=\stackrel{\&OverBar;}{\mathrm{AB}}=\sqrt{{(x2-x1)}^2+{(y2-y1)}^2}=160;$ Calculate the coordinate of each point on the tangent plane $\left\{\begin{array}{c}x=x1+l\cos \mathrm{\&theta;}=-80+l\\ y=y1+l\sin \mathrm{\&theta;}=0\end{array}\right.,(l=\mathrm{0,1},...160).$

As above, get θ=0,1,2 successively ... 179 extract tangent plane.

(3) choose secondary four chamber tangent planes in advance as template image, in the data base, seek the image the most similar as four chamber tangent planes to template image.At first according to the tangent plane picture among vertex position alignment template image and each width of cloth data base; Calculate the public intersection of each width of cloth tangent plane picture and template image then, should be divided into 2 * 2 parts by public intersecting area, the correlation coefficient summation of the entropy of each part is estimated as two the similar of width of cloth image.

Tangent plane is similar estimates for template tangent plane and θ=0 degree:

ECC(A，B)＝1.3579；

Respectively calculation template tangent plane and θ=0,1,2 successively ... 179 degree the similar of tangent plane are estimated, and (A, B) bigger 60 width of cloth tangent planes are as further image library to be matched to get ECC.Each width of cloth image and template image that alignment is remaining, calculate the small echo horizontal direction high frequency coefficient feature of their public intersection:

The template tangent plane:

$E=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}\left|x(m,n)\right|=-0.0185,$

$V=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}|x{(m,n)}^2-E^2|=8.5657.$

θ=0 degree tangent plane:

$E=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}\left|x(m,n)\right|=-0.0521$

$V=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}|x{(m,n)}^2-E^2|=7.5847$

Similarity: Sim (A, B)=(E _A-E _B) ²+ (V _A-V _B) ²=0.9635,

Successively respectively the Sim of calculation template tangent plane and 60 width of cloth tangent planes (A, B), (A, it is similar more B) to be worth more little presentation video for Sim.(A, B) minima is 0.9635 to Sim, is the tangent plane at 0 degree place, is four chamber tangent planes in this volume data.

Because four chamber tangent planes are determined by doctor's subjective experience usually, can't be quantified as an exact position, thus we with 3 doctors respectively the meansigma methods m of the position of the four chamber tangent planes that in this 29 individual data items, search out of 3 independence and standard variance sd as the correctness of evaluation result.Tangent plane position, four chambeies is represented by angle:

$\mathrm{angle}=\left\{\begin{array}{c}\mathrm{\&theta;}(0\&le;\mathrm{\&theta;}<90)\\ \mathrm{\&theta;}-180(90\&le;\mathrm{\&theta;}<180)\end{array}\right.,(-90\&le;\mathrm{angle}<90)$

As m-sd-5≤angle≤m+sd+5, tangent plane position, detected four chamber all is considered to correct.This volume data doctor testing result m=-1.2, sd=5.6, present embodiment experimental result angle=0 satisfies above-mentioned requirements.Volume data to 29 diastoles of 16 people latter stages experimentizes, and the experiment accuracy reaches 97%.

Claims (3)

1, the automatic method that detects of a kind of three-dimensional ultrasound cardiogram four-cavity section image is characterized in that, comprises the steps:

(1) from the total volume data file of left apex of the heart collection, extracts diastole three-dimensional data constantly in latter stage;

(2) extract tangent plane the volume data constantly latter stage from diastole, set up the image data base that comprises four chamber tangent planes;

(3) at first according to the tangent plane picture among vertex position alignment template image and each width of cloth data base; Then, utilize mutual information measure to carry out elementary coupling, some images lower in the elimination image data base with the template image similarity based on gray scale; Utilize the wavelet coefficient statistical nature to carry out similarity calculating in the E and the V feature of the high frequency coefficient of the wavelet decomposition horizontal direction of remaining image again, detect and the highest piece image of template image similarity, detailed process is:

Each width of cloth image and template image that alignment is remaining, the public part of asking two width of cloth images to intersect, the small echo horizontal direction high frequency coefficient feature of calculating intersection:

$E=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}\left|x(m,n)\right|,$

$V=\frac{1}{\mathrm{MN}}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}|x{(m,n)}^2-E^2|$

Similarity is defined as: and Sim (A, B)=(E _A-E _B) ²+ (V _A-V _B) ², Sim (A, B) the more little presentation video of value is similar more, Sim (A, B) tangent plane of minima is the four chamber tangent planes the most similar to template, is four chamber tangent planes in this volume data;

The image that obtains according to said process is considered to four chamber tangent plane pictures among this data base, writes down its position in this volume data and is four chambeies and sees the position.

2, the automatic method that detects of three-dimensional ultrasound cardiogram four-cavity section image according to claim 1 is characterized in that described step (2) is specially:

Tangent plane is projected as straight line AB on the XOY face, establishing AB and X-axis angle is θ (0≤θ＜180), and A and B select the intersection point into tangent plane border on the XOY face at projection line on the XOY face and volume data, and every degree is got a tangent plane, then tangental equation: y=tan (θ) x; On the XOY face, obtain A (x1, y1), B (x2, y2) coordinate, the unfolded image width: $L=\stackrel{\&OverBar;}{\mathrm{AB}}=\sqrt{{(x2-x1)}^2+{(y2-y1)}^2},$ Given θ and l, l for some P the subpoint of XOY face along AB to a some A (x1, distance y1), l≤L, arbitrfary point P satisfies on the tangent plane:

$\left\{\begin{array}{c}x=x1+l\cos \mathrm{\&theta;}\\ y=y1+l\sin \mathrm{\&theta;}\end{array}\right.,(0\&le;\mathrm{\&theta;}<180)$

Obtain the coordinate of size for being had a few on the arbitrarily angled tangent plane of L * n thus, wherein n is the height of volume data z direction, extracts 180 width of cloth tangent plane pictures altogether and forms an image data base.

3, the automatic method that detects of three-dimensional ultrasound cardiogram four-cavity section image according to claim 1 is characterized in that described utilization is carried out elementary coupling based on the mutual information measure of gray scale, and is specific as follows:

The public intersecting area of two images is divided into 2 * 2 parts, and with the correlation coefficient summation of the entropy of each part similarity as two width of cloth images, wherein the correlation coefficient computing formula of each part entropy is as follows:

$\mathrm{ECC}(A,B)=\frac{2I(A,B)}{H\left(A\right)+H\left(B\right)},$ ECC (A, value B) is big more, represents that two width of cloth images are similar more;

Wherein: $H\left(A\right)=-\underset{i}{\mathrm{\&Sigma;}}{p}_A\left(i\right)\log p_A\left(i\right),$

$H\left(B\right)=-\underset{j}{\mathrm{\&Sigma;}}{p}_B\left(j\right)\log p_B\left(j\right),$

$H(A,B)=-\underset{i,j}{\mathrm{\&Sigma;}}{p}_{\mathrm{AB}}(i,j)\log p_{\mathrm{AB}}(i,j),$

I(A，B)＝H(A)+H(B)-H(A，B)。

Images