CN102622759B - A kind of combination gray scale and the medical image registration method of geological information - Google Patents

Abstract

A kind of combination gray scale of the present invention is related to technical field of medical image processing with the medical image registration method of geological information.A kind of combination gray scale and the medical image registration method of geological information, described gray scale is the value in digital medical image in each pixel/voxel, is distributed as gradation of image information；Described geological information is the unit vector of the geometry normal direction for characterizing each point, the half-tone information of medical image is characterized by using grey value profile, the geological information of medical image is characterized using image method vector distribution, finally optimizes the similarity of gray scale and geometrical information using weighting mutual information.Present invention uses existing mutual information method and seeking gradient method, and the association relationship for obtaining is carried out plus asks entirely excellent, facilitated implementation in engineer applied；Take full advantage of the advantage of gray scale and geological information so that less in registration process fall into Local Extremum, with more preferable robustness；Experiment shows that the inventive method all obtains good precision.

Description

A kind of combination gray scale and the medical image registration method of geological information

Technical field

A kind of combination gray scale of the present invention is related to Medical Image Processing neck with the medical image registration method of geological information Domain.

Technical background

Modern medicine has been increasingly dependent on the assistance of armarium, and medical imaging device therein is then wherein most For important equipment, run through from whole processes such as medical diagnosiss, surgical navigational, postoperative effect judges.And with medical imaging skill The fast development of art, various imaging techniques provide more rich, more accurate imaging effect to doctors and patients.But different imaging skills Art is typically based on its image-forming principle and has the advantages which is unique, and there is also which specifically limits to, perfect and general into As technological means do not occur now yet, different imaging techniques have quality under different application scenarios, to different tissues device The imaging of official is also each has something to recommend him.Medical image fusion is by the different images synthesis according to the characteristics of each imaging technique after co-registration The process of new image is obtained, is a critically important technology in present Medical Image Processing.Except above-mentioned on medical diagnosiss Application, other one big application of Medical image fusion is in computer assisted surgery and treatment（IGST, Image Guided Surgery and Therapy）On.Computer assisted surgery and treatment are the cross disciplines of multiple research fields, help at which Under, doctor can carry out preoperative diagnosis, the planning of operation, the real-time navigation in art and fixed by means of computer and navigator Position, to accomplish accurate operation and Minimally Invasive Surgery as much as possible.Wherein, fixed in real time in the preoperative surgery planning and art being related to Position navigation, technology all based on image co-registration.

Medical image fusion is the process that several medical images are carried out informix, but, its combined process must So can not possibly arbitrarily, carry out fusion with having corresponding relation between medical image just meaningful, and several medical images are pressed Real body organizational information when which is imaged is set up the technology of correct corresponding relation and is referred to as medical figure registration （Medical Image Registration）.Therefore, the core of Medical image fusion is medical figure registration.

Medical image registration method can be divided into the algorithm based on gradation of image and base by its medical image for utilizing Algorithm in image geometry information.Algorithm based on half-tone information mainly has gray value difference averaging method, correlation coefficient process and mutually Information law.Algorithm based on geological information mainly has mensuration, contour line of gradient difference method, normal direction etc..But research worker is in application Find during various algorithms：Merely using a kind of quantity of information therein can not registering particularly well image, for example：For low resolution The image of rate（Such as PET）For, its half-tone information amount is fewer, is unable to reach good stability using half-tone information merely. Although having had the registering thought of the combination half-tone information and geological information including proposition including applicant, method to be in document The method that two category informations are passed through vector dot merely is combined together rigidly, is the thought for comparing early stage, and of the invention More efficiently, new technical scheme is proposed on this basis.

Content of the invention

It is an object of the invention to：The medical image registration method of a kind of combination gray scale and geological information is provided, using doctor The gray scale in image and geometrical information is learned, registration is carried out to many sub-pictures.

The purpose of the present invention is realized by following proposal：A kind of combination gray scale and the medical figure registration side of geological information Method, described gray scale are the values in digital medical image in each pixel/voxel, and grey value profile is gradation of image information；Institute The geological information that states is the unit vector of the geometry normal direction for characterizing each point, wherein, characterizes doctor by using grey value profile The half-tone information of image is learned, the geological information of medical image is characterized using image method vector distribution, finally using weighting mutual trust Cease to optimize the similarity of gray scale and geometrical information, realize that medical figure registration, step are：

The first step, asks for normal vector to reference picture and floating image；

Second step, asks for the mutual information of the component to normal vector；

3rd step, obtains normal vector mutual information to the mutual information summation of the component of each normal vector；

4th step, asks for the mutual information to reference picture and floating image intensity profile and is multiplied by the gray scale factor, to obtain Gray scale mutual information；

The normal vector mutual information that tries to achieve is obtained final measure value with the summation of gray scale mutual information by the 5th step；

Wherein, the first step to the method that reference picture and floating image ask for normal vector is：First to reference image R and conversion Floating image T (F) afterwards asks for water rogulator；Then, the gradient of two width images is asked for using gaussian kernel；After again, by asked for Gradient is normalized and obtains normal vector；Finally, by three methods of the floating image T (F) after the reference image R that tries to achieve and conversion The component of vector is stored respectively in three width image CR_m, CF_mWherein, m=0,1,2, represent x, tri- components of y, z respectively；

Second step asks for the mutual information method of the component to normal vector：The image of each normal vector component is asked for mutually Information MI_m=MI(CR_m,CF_m), wherein, mutual information MI comes from information-theoretical instrument, and its formula is：

Wherein H (A) and H (B) are the edge entropy of two signals A, B respectively, and H (A, B) is both combination entropies, combination entropy Computing formula be：

.

The following computing formula of measure value of the medical image registration method of a kind of described combination gray scale and geological information Description：

In formula, the distribution of the gray value of image is directly divided with the gray scale of another image as the source of gradation of image information Cloth does mutual information computing, obtains gray scale mutual information item after being multiplied by gray scale factor W；

By reference picture and floating image after a vector filter device, image method vector distribution is obtainedNV _AkWithNV _Bk, Wherein k=0,1,2, for 3-D view, its normal vector contains x, and tri- components of y, z, when two width images are registered, will be corresponded to Normal vector component distribution between be calculated geometry mutual information item；

Finally, all of mutual information is carried out suing for peace and obtains final test value.

Medical image registration method of the present invention with reference to gray scale with geological information is concretely comprised the following steps：

Func is to floating image T after reference image R and conversion（F）Ask for estimating Measure：

Ask for R and T（F）Normal direction spirogram NV（R）, NV（T（F））

Measure=0

For each normal direction spirogram each component i:

From NV（R）, NV（T（F））The middle image CR for extracting i-th component_i、CF_i,

Ask for CR_iAnd CF_iMutual information mi=MI (CR_i, CF_i)

Measure+=mi

END For each

Ask for R and T（F）Mutual information mi=MI (R, T（F）)

END Func.

Advantages of the present invention has：（1）Facilitate implementation in engineer applied, used existing mutual information method and sought gradient Method, and the association relationship for obtaining is added.（2）There is more preferable robustness, except believing using gray scale in registration process Breath, also uses the geological information of image, takes full advantage of the advantage of gray scale and geological information so that less in registration process Fall into Local Extremum.（3）Have compared with the much the same precision of mutual information.Experiment shows, either with mode or anomalous mode state number According to the inventive method all obtains good precision.

Description of the drawings

Accompanying drawing 1：The flow chart of method for registering of the present invention.

Specific embodiment

Below in conjunction with the accompanying drawings embodiments of the invention are elaborated.

Concretely comprise the following steps in conjunction with medical image registration method of the gray scale with geological information：It is Pentuim M in CPU 1.8GHz, video card are NVIDIA Geforce 6000, and under the conditions of inside saving as the computer hardware of 2.0GB, programming language is C++ Realize under environment：

Wherein, described gray scale is the value in digital medical image in each pixel/voxel, is distributed as gradation of image letter Breath；Described geological information is the unit vector of the geometry normal direction for characterizing each point, and feature is：Come by using grey value profile The half-tone information of medical image is characterized, and the geological information of medical image is characterized using image method vector distribution, finally using mutual Information is weighing the similarity of gray scale and geometrical information.

Quantitive error analysis are carried out using BrainWeb data bases, wherein, the image in BrainWeb data bases is basis The image of the advance registration that the image-forming principle of MRI is obtained using the simulation of nuclear magnetic resonance, NMR simulator, as the goldstandard of quantitative analyses, A pre-transform is carried out to piece image therein in advance, makes that two width images are artificial must to be shifted by；

BrainWeb data bases contain the MRI image of different relaxation times and parameter, including T1, tri- sequence chart of T2, PD, As image is obtained by computer simulation, so these images are all to have identical coordinate system, it is to be registered Image because BrainWeb data bases have this characteristic, using which as accuracy of quantitative analysis is carried out to originally estimating, select T1 and PD sequence chart removes noise as reference picture and floating image, and picture size is all 181x217x181, image voxel At intervals of 1mm × 1mm × 1mm；

Before registering platform is entered, floating image first carries out pre-transform, the one-to-one image of two width is lost Go correspondence, but our known its not corresponding spatial alternations, i.e. standard deviation；Then again by reference picture and conversion in advance The floating image that crosses is input in registering platform, carries out registration using estimating, such as the flow chart institute of Fig. 1 method for registering of the present invention Show, image carries out registration by following steps：

The first step, asks for normal vector to reference picture and floating image；

Second step, after a vector filter device, image obtains its normal vector distribution, for 3-D view, its normal direction Amount asks for the mutual information of component 1, component 2 and component 3 to normal vector, when two width containing three components 1, component 2 and component 3s Image is registered, will be calculated mutual information between the component distribution of corresponding normal vector；

3rd step, obtains normal vector mutual information to the mutual information summation of the component of each normal vector；

4th step, asks for the mutual information to reference picture and floating image intensity profile and is multiplied by gray scale factor R；

The normal vector mutual information that tries to achieve is obtained final measure value with the summation of gray scale mutual information by the 5th step.

The conversion that registration is obtained using registration after terminating carries out error quantitative assessment with standard deviation before；The present embodiment In, we are averaged using the parameters of conversion and standard deviation are carried out difference simply, are asking registration result and standard During difference between deviation using phase add operation, formula it is：

Wherein, p_sAnd p_s' be respectively the conversion and standard deviation obtained by registration s-th parameter, N be conversion parameter The conversion obtained by registration is carried out addition with each corresponding parameter of standard deviation and seeks absolute value by number.

But, the conversion obtained by registration should have opposite effect with standard deviation, and for example we apply one in advance Floating image is given to the change of left, then if registration success, then the conversion after registration should necessarily make floating figure As to right translation, therefore rather than subtract each other using phase add operation when the difference between registration result and standard deviation is sought.Three-dimensional rigid body Conversion has 6 parameters.When asking registration result with goldstandard difference using phase add operation.

The trueness error of acquisition is as follows：

1. BrainWeb Images Registrations of table are compared with mutual information

The present embodiment is implemented under premised on technical solution of the present invention, gives detailed embodiment and concrete Operating process, but protection scope of the present invention is not limited to the above embodiments.

Claims (4)

1. the medical image registration method of a kind of combination gray scale and geological information, described gray scale is each in digital medical image Gray value in pixel/voxel, is distributed as gradation of image information；Described geological information is the geometry of each point on phenogram picture The unit vector information of normal direction, it is characterised in that：The half-tone information of medical image is characterized by using intensity profile, using figure The geological information of medical image is characterized as normal vector distribution, finally optimizes gray scale and geometrical information using weighting mutual information Similarity, realize that the registration between medical image, step are：

The first step, asks for normal vector to reference picture and floating image；

Second step, asks for the mutual information of the component to normal vector；

3rd step, obtains normal vector mutual information to the mutual information summation of the component of each normal vector；

4th step, asks for the mutual information to reference picture and floating image intensity profile and is multiplied by the gray scale factor, to obtain gray scale Mutual information；

The normal vector mutual information that tries to achieve is obtained final measure value with the summation of gray scale mutual information by the 5th step；

Wherein,

The first step to the method that reference picture and floating image ask for normal vector is：First to the floating after reference image R and conversion Image T (F) asks for water rogulator；Then, the gradient of two width images is asked for using gaussian kernel；After again, the gradient that asks for is carried out Normalization obtains normal vector；Finally, dividing three normal vectors of the floating image T (F) after the reference image R that tries to achieve and conversion Amount is stored respectively in three width image CR_m, CF_mWherein, m=0,1,2, represent x, tri- components of y, z respectively；

Second step asks for the mutual information method of the component to normal vector：Mutual information is asked for the image of each normal vector component MI_m=MI (CR_m,CF_m), wherein, mutual information MI comes from information-theoretical instrument, and its formula is：

$\begin{array}{c}MI\left(A,B\right)=H\left(A\right)+H\left(B\right)+H\left(A,B\right)\\ ={\mathrm{\Σ}}_i{f}_i\log \frac{1}{f_i}+{\mathrm{\Σ}}_j{f}_j\log \frac{1}{f_j}-{\mathrm{\Σ}}_i{\mathrm{\Σ}}_{j}p\left(i,j\right)\log p\left(i,j\right)\end{array},$

Wherein H (A) and H (B) are the edge entropy of two signals A, B respectively, and H (A, B) is both combination entropies, the meter of combination entropy Calculating formula is：

$H(A,B)=-\underset{i,j}{\Σ}p(i,j)\log p(i,j).$

2. the medical image registration method of a kind of combination gray scale according to claim 1 and geological information, it is characterised in that The measure value of method for registering is described with following computing formula：

$\begin{array}{c}Measure\left(A,B\right)=MI\left(A,B\right)\·W+\underset{k=1}{\overset{n}{\mathrm{\Σ}}}MI\left({\mathrm{NV}}_{Ak},{\mathrm{NV}}_{Bk}\right)\\ =\left(H\left(A\right)+H\left(B\right)-H\left(A,B\right)\right)\·W+\\ \underset{k=1}{\overset{n}{\mathrm{\Σ}}}\left(H\left({\mathrm{NV}}_{Ak}\right)+H\left({\mathrm{NV}}_{Bk}\right)-H\left({\mathrm{NV}}_{Ak},{\mathrm{NV}}_{Bk}\right)\right),\end{array}$

In formula, the distribution of the gray value of image is directly done with the intensity profile of another image as the source of gradation of image information Mutual information computing, obtains gray scale mutual information item after being multiplied by gray scale factor W；

By image after a vector filter device, image method vector distribution NV is obtained_AkAnd NV_Bk, wherein k=0,1,2, for three Dimension image, its normal vector contain x, and tri- components of y, z, when two width images are registered, the component of corresponding normal vector are distributed Between be calculated geometry mutual information item；

Finally, all of mutual information is carried out suing for peace and obtains final test value.

3. the medical image registration method of a kind of combination gray scale according to claim 1 and geological information, it is characterised in that Func is asked for estimating Measure to floating image T (F) after reference image R and conversion：

Ask for normal direction spirogram NV (R) of R and T (F), NV (T (F))

Measure=0

For each normal direction spirogram each component m:

The image CR of m-th component is extracted from NV (R), NV (T (F))_m、CF_m,

Ask for CR_mAnd CF_mMutual information mi=MI (CR_m, CF_m)

Measure+=mi

END For each

Ask for the mutual information mi=MI (R, T (F)) of R and T (F)

END Func.

4. the medical image registration method of a kind of combination gray scale according to claim 1 and geological information, it is characterised in that Concretely comprise the following steps：It is Pentuim M 1.8GHz in CPU, video card is NVIDIA Geforce 6000, inside saves as the meter of 2.0GB Under calculation machine hardware condition, programming language is to realize under C++ environment：

Quantitive error analysis are carried out using BrainWeb data bases, wherein, the image in BrainWeb data bases is according to MRI The image of the advance registration that image-forming principle is obtained using the simulation of nuclear magnetic resonance, NMR simulator, as the goldstandard of quantitative analyses, in advance A pre-transform is carried out to piece image therein, makes that two width images are artificial must to be shifted by；BrainWeb data bases contain difference Relaxation time and the MRI image of parameter, including T1, tri- sequence chart of T2, PD, as image is obtained by computer simulation Arrive, so these images are all to have identical coordinate system, it is the image being registered, from T1 and PD sequence chart as ginseng Examine image and floating image, remove noise, picture size is all 181x217x181, image voxel at intervals of 1mm × 1mm × 1mm；

Before registering platform is entered, floating image first carries out pre-transform, the one-to-one image of two width is lost right Ying Xing, it is known that its not corresponding spatial alternation, i.e. standard deviation；Then again by reference picture and floating image transformed in advance Be input in registering platform, registration is carried out using estimating, conversion and standard deviation before that registration is obtained using registration after terminating Difference carries out error quantitative assessment；The parameters of conversion and standard deviation are carried out difference to average, registration result and mark is being asked During difference between quasi- deviation using phase add operation, formula it is：

$ERR=\frac{1}{N}\underset{s=0}{\overset{N}{\Σ}}|P_s+{P_s}^{\′}|,$

Wherein, p_sAnd p_s' be respectively the conversion and standard deviation obtained by registration s-th parameter, N be conversion number of parameters, The conversion obtained by registration is carried out addition with each corresponding parameter of standard deviation and seeks absolute value.