CN106127753A - CT image body surface handmarking's extraction method in a kind of surgical operation - Google Patents

Abstract

The present invention relates to CT image body surface handmarking's extraction method in a kind of surgical operation, comprise the following steps: S1, reading three-dimensional CT images, and CT image is carried out body surface segmentation, it is thus achieved that three-dimensional body surface curved surface；S2, on the axle position two-dimensional silhouette image that three-dimensional body surface curved surface is corresponding the localized indentation convexity of the three-dimensional body surface curved surface of detection, and automatically extract the singular point through patched two-dimensional silhouette, determine the Probability Area that handmarking puts；S3, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, three dimensions formed the cluster of singular point set, and is determined the substantially three-dimensional space position of handmarking's candidate point by cluster centre；S4, by the bone segmentation image of CT image, handmarking's candidate point is verified, determine the precise area that handmarking puts；S5, output handmarking's point.The present invention is not required to use any three-dimensional traversal coupling or three dimensions Filtering Processing, and the extraction rate of labelling point is the fastest.

Description

CT image body surface handmarking's extraction method in a kind of surgical operation

Technical field

The present invention relates to Medical Imaging Technology field, in particular, relate to CT image body surface in a kind of surgical operation Handmarking's extraction method.

Background technology

In computer-assisted surgery and Image-guided surgery, image registration is an important ring.Due to art In three-dimensional real-time imaging be not easy to obtain, the most generally obtain bidimensional image by X-ray fast imaging, and such two dimension Precisely manipulation and the identification of space anatomical structure in the most unfavorable Rhizoma Atractylodis Macrocephalae of spatial information of image shortage three-dimensional data.By preoperative Registration between the bidimensional image obtained in real time in volume data and art provides real-time three-dimensional spatial information can to operation, thus assists Surgical operation precisely operates.It is the Image registration of two and three dimensions that registration between X-ray and CT image belongs to, and Image registration is general It is divided into feature based and based on gray scale two types.The registration of feature based relies primarily on handmarking's point, or extraction group Knit Morphologic Characteristics, by these features coupling between different modalities, thus realize the registration of two and three dimensions image.Special The extraction levied and coupling, be very important two links in the registration process of feature based.Image registration based on gray scale Need not extract image feature, but it is universal relatively to make to registrate speed owing to relying on the search of a large amount of iteration optimization in calculating process Slowly, the requirement of real-time of registration in art cannot often be met.Therefore, during real-time surgical navigational, joining of feature based Quasi-method is widely adopted.In the method for registering of feature based, feature extraction is important link, and handmarking is owing to having The shape of priori and imaging characteristic, be widely used in real-time operation guiding system based on two dimension three-dimensional registration.

Body surface handmarking extracts the cutting techniques being normally based on characteristics of image.Owing to handmarking is at three-D CT imaging In have and the diverse signal strength range of soft tissue such as proximate skin, therefore Threshold sementation based on image greyscale Handmarking can be separated from neighbouring body surface.Owing to handmarking's signal value scope in CT imaging is fluctuation, Different sticking positions has different signal intensitys, easily produces segmentation by mistake based on cutting techniques such as threshold values and leaks showing of segmentation As.And, handmarking is close to body surface, the most very neighbouring body Endoskeleton, and handmarking has in CT image with skeleton The highest signal value, this adhesion is had easily to cause extraction mistake.In the case of handmarking's limited amount, automanual Mutual Threshold segmentation can improve positional precision and the accuracy that handmarking extracts.But, when handmarking's data are more and Sites distributed more widely general time, this mutual threshold segmentation method can bring bigger burden and time to operator Expend.

Owing to handmarking's signal value scope in CT imaging is fluctuation, different sticking positions has different letters Number intensity, easily produces segmentation by mistake and the phenomenon of leakage segmentation based on cutting techniques such as threshold values.And, handmarking be close to body surface, The most very neighbouring body Endoskeleton, and handmarking and skeleton all have the highest signal value, this adhesion in CT image Easily cause extraction mistake.In the case of handmarking's limited amount, automanual mutual Threshold segmentation can improve manually The positional precision of labelling extraction and accuracy.But, when handmarking's data are more and sites distributed more widely general time, this Plant mutual threshold segmentation method and can bring bigger burden and time consumption to operator.

In the case of handmarking's model determines, method based on Model Matching is also widely used for body surface marking Location.This kind of technology based on threedimensional model coupling needs the whole three-dimension curved surface of search spread to make computational complexity higher, and And the extraction accuracy of three-dimension curved surface can have a strong impact on the result of coupling of labelling.When label size is less and CT image quality less High or in the case of there is noise jamming, this performance based on model and SURFACES MATCHING can significantly reduce.

The body surface marking extractive technique filtered based on three-dimensional surface rebuilding and three-dimensional multi-scale, first on computational complexity The highest, the extraction of surface characteristics is difficult to accomplish to process in real time.Additionally the multi-scale filtering of three-dimension curved surface easily causes space bit The drift put, the positional precision of handmarking's point also can be by large effect.Being similar to, the reconstruction precision of three-dimensional surface is same Determine the performance of artificial locating mark points, when handmarking is less in 3-dimensional image mesoscale and three-dimensional surface rebuilding performance cannot Three-dimensional localized indentation convexity in the presence of objective statement body surface marking, this technical method is difficult to stablize, automatically extract accurately manually Labelling.

Summary of the invention

In view of this, it is necessary to for the problems referred to above, it is provided that in a kind of surgical operation, CT image body surface handmarking is automatic Extracting method, is not required to use any three-dimensional traversal coupling or three dimensions Filtering Processing, and the extraction rate of labelling point is non- The fastest.

To achieve these goals, technical scheme is as follows:

CT image body surface handmarking's extraction method in a kind of surgical operation, comprises the following steps:

S1, reading three-dimensional CT images, and CT image is carried out body surface segmentation, it is thus achieved that three-dimensional body surface curved surface；

S2, on the axle position two-dimensional silhouette image that three-dimensional body surface curved surface is corresponding, the local of the three-dimensional body surface curved surface of detection is concavo-convex Property, and automatically extract the singular point through patched two-dimensional silhouette, determine the Probability Area that handmarking puts；

S3, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, on three dimensions Form the cluster of singular point set, and determined the substantially three-dimensional space position of handmarking's candidate point by cluster centre；

S4, by the bone segmentation image of CT image, handmarking's candidate point is verified, determine what handmarking put Precise area；

S5, output handmarking's point.

As preferably, described step S1 specifically includes:

S11, the method purged body off-balancesheet impurity increased by region, determined body surface general profile, and other substantially taken turns Region CT value zero setting beyond wide correspondence position, obtains coarse segmentation data；

S12, each axle position body surface two-dimensional silhouette to body surface coarse segmentation volume data carry out repairing treatment.

As preferably, described step S11 specifically includes:

S111, outside body surface, select seed points, increase purged body off-balancesheet impurity by region；

Result after S112, employing three-dimensional edges detection processing region growth process, determines the general profile of body surface；

S113, the CT value of general profile correspondence position is kept constant, and by beyond other general profile correspondence position Region CT value zero setting.

As preferably, described step S12 specifically includes:

S121, for present on contour line be interrupted, carry out automatic polishing；

S122, the main outline of Extracting contour set；

S123, for outside main wheel profile occur tiny branch, carry out automatic minor matters deletion；

S124, all axle positions slice image is carried out profile repairing treatment, obtain final body surface segmentation result.

As preferably, described step S2 uses multiple dimensioned profile inflection point detection method, multiple dimensioned by profile The curvature extremum value of local configuration curve determine that handmarking puts the singular point occurred on body surface profile, and then determine artificial mark The region of note.

As preferably, described multiple dimensioned profile inflection point detection method specifically includes:

It is respectively directed to contour edge discrete point set { (x_i,y_i: i=0 ..., n-1) x and y one-dimensional vector carry out respectively Multi-scale filtering, carries out the multiple dimensioned smoothing processing of profile by Gaussian filter, and the expression formula of Gaussian function is as follows:

$G\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}{e}^{-\frac{x^2}{2{\mathrm{\σ}}^2}}$

Wherein σ is the standard deviation of Gaussian function, i.e. the scale factor of multi-scale filtering device；

To profile point { (x in rectangular coordinate system_i,y_i: i=0 ..., n-1) to carry out Parameter Expression be P (u)=[x (u), y (u)], then the curvature estimation formula of contour curve is:

$k\left(u_i\right)=\frac{\left|x^{\′}{y}^{\′\′}-y^{\′}{x}^{\′\′}\right|}{{(x^{\′2}+y^{\′2})}^{3/2}}$

Wherein x ', y ', x " and y " be defined respectively as:

$x^{\′}\left(u_i\right)=\frac{x(u_i+\mathrm{\Δ}u)-x\left(u_i\right)}{\mathrm{\Δ}u},y^{\′}\left(u_i\right)=\frac{y(u_i+\mathrm{\Δ}u)-y\left(u_i\right)}{\mathrm{\Δ}u}$

$x^{\′\′}\left(u_i\right)=\frac{x(u_i+\mathrm{\Δ}u)+x(u_i-\mathrm{\Δ}u)-2x\left(u_i\right)}{{\left(\mathrm{\Δ}u\right)}^2},y^{\′\′}\left(u_i\right)=\frac{y(u_i+\mathrm{\Δ}u)+y(u_i-\mathrm{\Δ}u)-2y\left(u_i\right)}{{\left(\mathrm{\Δ}u\right)}^2}$

In formula, u is parameter of curve parameter and u ∈ [0,1], u_iFor profile point (x_i,y_iParameter of curve corresponding to) Value, △ u is the micro-side-play amount on parameter curve, takes 0.01；It is determined by the point with extreme curvature on multiple yardstick, detects profile The singular point of upper stable existence.

As preferably, described step S3 specifically includes:

S31, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, at three dimensions The cluster of upper formation candidate singular point set；

S32, by clustering procedure, singular point set is carried out cluster analysis, so that it is determined that the center of cluster.

As preferably, described step S32 use k-mean clustering procedure singular point set is carried out cluster analysis, specifically Including:

S321, from N number of candidate singular point set, randomly select K as initial cluster center；

S322, to remaining each singular point, measure its distance to each cluster centre, and it be grouped into nearest poly- Class center；

S323, recalculate the cluster centre c [i] of each class obtained=all data [j] being labeled as i it With/it is labeled as the number of i；

S324, iteration S32, S23 are until new cluster centre is equal with former cluster centre or less than specifying threshold value, algorithm is tied Bundle, obtains the three-dimensional space position of handmarking's candidate point.

As preferably, in described step S4, by multi-level Threshold segmentation multiple target discontinuity zone, by skeleton and Handmarking puts unified splitting, and retrieves singular point cluster and form handmarking's time in the three-dimensional data of bone segmentation Reconnaissance coordinate, determines that handmarking puts region.

Compared with prior art, the beneficial effects of the present invention is:

1, quick three-dimensional body surface extracts and have employed two-dimensional silhouette and repair, and has that speed is fast, precision is high in terms of body surface extraction Feature, various noises in terms of CT imaging and non-perceptive distortion all may cause " over-segmentation " (body surface in segmentation result of body surface Adhesion with neighbouring body inner tissue) or " less divided " (in segmentation result, cavity or discontinuous situation occurs in body surface), and the present invention Propose to improve three-dimensional body surface curved surface extraction performance by two-dimensional silhouette repairing to eliminate above-mentioned " over-segmentation " completely and " owe to divide Cut " situation；

2, it is difficult to stable and precisely detection and existence due to the three-dimensional localized indentation convexity of labelling point and body surface at three dimensions The deficiency that computational complexity is too high, there is curvature pole in the axle position two-dimensional silhouette that the present invention proposes by analyzing three-dimensional body surface curved surface The concavo-convex characteristic of value, so that it is determined that the position that handmarking is in two-dimensional silhouette；Due to several in CT of some handmarking All formed concavo-convex on the profile of individual continuous axle position, then the point with extreme curvature that continuous several axle positions Plane Location is the most close, so that it may To determine that handmarking, in three-dimensional position, and can get rid of the impact of other unartificial labelling；Therefore, the formation of noise A certain layer convex and concave feature miscalculation be difficult to have influence on the extraction of real handmarking, because noise is difficult in position the most close Several continuous axle profiles on all formed concavo-convex；

3, any three-dimensional traversal coupling or three dimensions Filtering Processing, labelling are not used due to the inventive method The extraction rate of point is the fastest；The repairing of three-dimension curved surface, is converted into the repairing of two-dimensional curve profile, and the office of three-dimensional space curved surface The concavo-convex detection in portion, is converted into local curvature's extremum extracting of two-dimensional curve profile, will significantly reduce in computing cost.

Accompanying drawing explanation

Fig. 1 is the method flow diagram of the embodiment of the present invention；

Fig. 2 is the CT image schematic diagram in the embodiment of the present invention after labelling；

Fig. 3 is coarse segmentation schematic diagram in the embodiment of the present invention；

Fig. 4 is design sketch after embodiment of the present invention profile is finely repaired；

Fig. 5 is profile multiple dimensioned inflection point detection schematic diagram in the embodiment of the present invention；

Fig. 6 is profile inflection point detection result schematic diagram in the serial section of embodiment of the present invention axis position；

Fig. 7 is singular point three dimensions cluster schematic diagram in the embodiment of the present invention；

Fig. 8 is that in the embodiment of the present invention, bone segmentation puts extraction result schematic diagram containing handmarking's image and handmarking.

Detailed description of the invention

Below in conjunction with the accompanying drawings with embodiment to CT image body surface handmarking in a kind of surgical operation of the present invention certainly Dynamic extracting method is described further.

The following is the optimal of CT image body surface handmarking's extraction method in a kind of surgical operation of the present invention Example, the most therefore limits protection scope of the present invention.

Fig. 1 shows CT image body surface handmarking's extraction method in a kind of surgical operation, to area of computer aided outside In section's surgical navigational, CT image body surface handmarking carries out full-automatic rapid extraction, and the method for the present invention comprises the following steps:

S1, reading three-dimensional CT images, and CT image is carried out body surface segmentation, it is thus achieved that three-dimensional body surface curved surface；

S2, on the axle position two-dimensional silhouette image that three-dimensional body surface curved surface is corresponding, the local of the three-dimensional body surface curved surface of detection is concavo-convex Property, and automatically extract the singular point through patched two-dimensional silhouette, determine the Probability Area that handmarking puts；

S3, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, on three dimensions Form the cluster of singular point set, and determined the substantially three-dimensional space position of handmarking's candidate point by cluster centre；

S4, by the bone segmentation image of CT image, handmarking's candidate point is verified, determine what handmarking put Precise area；

S5, output handmarking's point.

In the present embodiment, described step S1 specifically includes:

S11, the method purged body off-balancesheet impurity increased by region, determined body surface general profile, and other substantially taken turns Region CT value zero setting beyond wide correspondence position, obtains coarse segmentation data；

S12, each axle position body surface two-dimensional silhouette to body surface coarse segmentation volume data carry out repairing treatment.

In the present embodiment, described step S11 specifically includes:

S111, outside body surface, select seed points, increase purged body off-balancesheet impurity by region；

Result after S112, employing three-dimensional edges detection processing region growth process, determines the general profile of body surface；

S113, the CT value of general profile correspondence position is kept constant, and by beyond other general profile correspondence position Region CT value zero setting.

As preferably, described step S12 specifically includes:

S121, for present on contour line be interrupted, carry out automatic polishing；

S122, the main outline of Extracting contour set；

S123, for outside main wheel profile occur tiny branch, carry out automatic minor matters deletion；

S124, all axle positions slice image is carried out profile repairing treatment, obtain final body surface segmentation result, such as Fig. 2 Shown in, left hand view is the unsceptered figure of CT image, and right part of flg is CT image axle bitmap；Present invention achieves the extraction of quick three-dimensional body surface to adopt With two-dimensional silhouette repairing, there is in terms of body surface extraction the feature that speed is fast, precision is high.Various noises in terms of CT imaging and Non-perceptive distortion all may cause " over-segmentation " (body surface and the adhesion of neighbouring body inner tissue in segmentation result) of body surface or " owe to divide Cut " (in segmentation result, there is cavity or discontinuous situation in body surface), and the present invention proposes to be repaired by two-dimensional silhouette to improve three-dimensional Body surface curved surface extracts performance can eliminate above-mentioned " over-segmentation " and " less divided " situation completely.

When being close to body surface due to handmarking's body surface, body surface curved surface can exist three-dimensional localized indentation convexity, the present invention will The axle position bidimensional image that three-dimensional body surface curved surface is corresponding detects this three-dimensional localized indentation convexity, namely two through repairing Singular point is automatically extracted in dimension main outline；Making discovery from observation, the body surface profile having handmarking to put appearance is compared there is not people The extraction region of work labelling is generally protruded, and the shape of this protrusion has certain change.In the present embodiment, the present invention carries Go out a kind of multiple dimensioned profile inflection point detection method, determine this protruding point by the local extremum that profile is multiple dimensioned, from And determine the Probability Area of handmarking, described step S2 uses multiple dimensioned profile inflection point detection method, passes through profile The curvature extremum value of multiple dimensioned local configuration curve determines that handmarking puts the singular point occurred on body surface profile, and then determines The region of handmarking, as shown in Figure 3, Figure 4, in Fig. 3, left figure is coarse segmentation body surface rendering result, seven body surface markings of chest Point is protruding visible；Right figure is coarse segmentation axle position contour line, and body surface marking causes contour line to be interrupted and disappearance；Fig. 4 is that profile is fine Effect after repairing.

In the present embodiment, described multiple dimensioned profile inflection point detection method specifically includes:

It is respectively directed to contour edge discrete point set { (x_i,y_i: i=0 ..., n-1) x and y one-dimensional vector carry out respectively Multi-scale filtering, carries out the multiple dimensioned smoothing processing of profile by Gaussian filter, and the expression formula of Gaussian function is as follows:

$G\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}{e}^{-\frac{x^2}{2{\mathrm{\σ}}^2}}$

Wherein σ is the standard deviation of Gaussian function, i.e. the scale factor of multi-scale filtering device, as it is shown in figure 5, for wheel in figure Wide multiple dimensioned inflection point detection result, in figure, Sigma, i.e. σ, be the standard deviation of Gaussian function, i.e. the chi of multi-scale filtering device The degree factor, indicates in figure and represents under different scale factors, the inflection point detection result of corresponding same profile.

To profile point { (x in rectangular coordinate system_i,y_i: i=0 ..., n-1) to carry out Parameter Expression be P (u)=[x (u), y (u)], then the curvature estimation formula of contour curve is:

$k\left(u_i\right)=\frac{\left|x^{\′}{y}^{\′\′}-y^{\′}{x}^{\′\′}\right|}{{(x^{\′2}+y^{\′2})}^{3/2}}$

Wherein x ', y ', x " and y " be defined respectively as:

$x^{\′}\left(u_i\right)=\frac{x(u_i+\mathrm{\Δ}u)-x\left(u_i\right)}{\mathrm{\Δ}u},y^{\′}\left(u_i\right)=\frac{y(u_i+\mathrm{\Δ}u)-y\left(u_i\right)}{\mathrm{\Δ}u}$

$x^{\′\′}\left(u_i\right)=\frac{x(u_i+\mathrm{\Δ}u)+x(u_i-\mathrm{\Δ}u)-2x\left(u_i\right)}{{\left(\mathrm{\Δ}u\right)}^2},y^{\′\′}\left(u_i\right)=\frac{y(u_i+\mathrm{\Δ}u)+y(u_i-\mathrm{\Δ}u)-2y\left(u_i\right)}{{\left(\mathrm{\Δ}u\right)}^2}$

In formula, u is parameter of curve parameter and u ∈ [0,1], u_iFor profile point (x_i,y_iParameter of curve corresponding to) Value, △ u is the micro-side-play amount on parameter curve, takes 0.01；It is determined by the point with extreme curvature on multiple yardstick, detects profile The singular point of upper stable existence.It is determined by the point with extreme curvature on multiple yardstick, detects the unusual of stable existence on profile Point；Fig. 6 is profile inflection point detection result in the serial section of axle position, Slice represents the Slice Sequence of volume data axle position aspect, Such as: Slice#83, i.e. the 83rd layer, axle position section, Slice#84, i.e. the 84th layer, axle position section, figure shows in continuous sequence Axle position aspect section on, corresponding profile inflection point detection result.

Owing to all forming differ in size concavo-convex on some handmarking several continuous axle profiles in CT, then even The point with extreme curvature that continuous several axle positions Plane Location is the most close, will correspond to handmarking's point.The present invention is by right In the two-dimensional silhouette of all axle positions, the profile singular point of multiple scale detecting, carries out three-dimensional cluster analysis, then same person Work labelling must form the little cluster of singular point set on three dimensions.The most both people can substantially be determined by cluster centre The three-dimensional space position of work labelling, can get rid of the singular point set of some unartificial labellings simultaneously.In the present embodiment, described Step S3 specifically includes:

S31, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, at three dimensions The cluster of upper formation candidate singular point set；

S32, by clustering procedure, singular point set is carried out cluster analysis, so that it is determined that the center of cluster, such as Fig. 7 institute Show, figure is (4) singular point three dimensions cluster.

As preferably, described step S32 use k-mean clustering procedure singular point set is carried out cluster analysis, specifically Including:

S321, from N number of candidate singular point set, randomly select K as initial cluster center；

S322, to remaining each singular point, measure its distance to each cluster centre, and it be grouped into nearest poly- Class center；

S323, recalculate the cluster centre c [i] of each class obtained=all data [j] being labeled as i it With/it is labeled as the number of i；

S324, iteration S32, S23 are until new cluster centre is equal with former cluster centre or less than specifying threshold value, algorithm is tied Bundle, obtains the three-dimensional space position of handmarking's candidate point.

Owing to the three-dimensional localized indentation convexity of labelling point and body surface is difficult to stable at three dimensions and precisely detects and there is fortune Calculating the deficiency that complexity is too high, there is curvature extremum value in the axle position two-dimensional silhouette that the present invention proposes by analyzing three-dimensional body surface curved surface Concavo-convex characteristic, so that it is determined that the position that handmarking is in two-dimensional silhouette.Due to several in CT of some handmarking All formed concavo-convex on the profile of continuous axle position, then the point with extreme curvature that continuous several axle positions Plane Location is the most close, it is possible to Determine that handmarking, in three-dimensional position, and can get rid of the impact of other unartificial labelling.Therefore, the formation of noise The miscalculation of a certain layer convex and concave feature is difficult to have influence on the extraction of real handmarking, because noise is difficult in position the most close All formed concavo-convex on several continuous axle profiles.

Accurately extracted by body surface and the detection of two-dimensional silhouette singular point combine three-dimensional cluster analysis, may be used To determine the candidate point of handmarking's point.But, by handmarking's three-dimensional coordinate determined by body surface profile and cluster centre Position inaccuracy, be not the center of handmarking accurately.The present invention proposes bone segmentation image based on CT image Obtain the three-dimensional segmentation result of handmarking further, cluster the handmarking's candidate point obtained in conjunction with three dimensions singular point, Thus accurately extract handmarking cut zone and using its center of gravity as handmarking point three-dimensional space position.Due to manually Labelling and skeleton are respectively provided with higher signal strength values in CT image, then do not connected by multi-level Threshold segmentation multiple target Continuous region, it is possible to by skeleton and unified the splitting of handmarking.Now, retrieve in the three-dimensional data of bone segmentation Singular point cluster forms handmarking's candidate point coordinate, then handmarking puts region and is assured that, therefore, at the present embodiment In, by multi-level Threshold segmentation multiple target discontinuity zone, skeleton and handmarking are put unified splitting, at bone The three-dimensional data of bone segmentation is retrieved singular point cluster and forms handmarking's candidate point coordinate, determine that handmarking puts region, As shown in Figure 8, figure puts extraction result for bone segmentation containing handmarking's image and handmarking.

Embodiment described above only have expressed the several embodiments of the present invention, and it describes more concrete and detailed, but also Therefore the restriction to the scope of the claims of the present invention can not be interpreted as.It should be pointed out that, for those of ordinary skill in the art For, without departing from the inventive concept of the premise, it is also possible to make some deformation and improvement, these broadly fall into the guarantor of the present invention Protect scope.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (9)

1. CT image body surface handmarking's extraction method in a surgical operation, it is characterised in that comprise the following steps:

S1, reading three-dimensional CT images, and CT image is carried out body surface segmentation, it is thus achieved that three-dimensional body surface curved surface；

S2, on the axle position two-dimensional silhouette image that three-dimensional body surface curved surface is corresponding the localized indentation convexity of the three-dimensional body surface curved surface of detection, and Automatically extract the singular point through patched two-dimensional silhouette, determine the Probability Area that handmarking puts；

S3, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, on three dimensions formed The cluster of singular point set, and the substantially three-dimensional space position of handmarking's candidate point is determined by cluster centre；

S4, by the bone segmentation image of CT image, handmarking's candidate point is verified, determine that handmarking puts accurate Region；

S5, output handmarking's point.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 1, it is characterised in that Described step S1 specifically includes:

S11, the method purged body off-balancesheet impurity increased by region, determine body surface general profile, and by other general profile pair Answer the region CT value zero setting beyond position, obtain coarse segmentation data；

S12, each axle position body surface two-dimensional silhouette to body surface coarse segmentation volume data carry out repairing treatment.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 2, it is characterised in that Described step S11 specifically includes:

S111, outside body surface, select seed points, increase purged body off-balancesheet impurity by region；

Result after S112, employing three-dimensional edges detection processing region growth process, determines the general profile of body surface；

S113, the CT value of general profile correspondence position is kept constant, and by the region beyond other general profile correspondence position CT value zero setting.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 2, it is characterised in that Described step S12 specifically includes:

S121, for present on contour line be interrupted, carry out automatic polishing；

S122, the main outline of Extracting contour set；

S123, for outside main wheel profile occur tiny branch, carry out automatic minor matters deletion；

S124, all axle positions slice image is carried out profile repairing treatment, obtain final body surface segmentation result.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 1, it is characterised in that Described step S2 uses multiple dimensioned profile inflection point detection method, by the local configuration song curvature of a curve that profile is multiple dimensioned Extreme value determines that handmarking puts the singular point occurred on body surface profile, and then determines the region of handmarking.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 5, it is characterised in that Described multiple dimensioned profile inflection point detection method specifically includes:

It is respectively directed to contour edge discrete point set { (x_i,y_i: i=0 ..., n-1) x and y one-dimensional vector carry out many chis respectively Degree filtering, carries out the multiple dimensioned smoothing processing of profile by Gaussian filter, and the expression formula of Gaussian function is as follows:

$G\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}{e}^{-\frac{x^2}{2{\mathrm{\σ}}^2}}$

Wherein σ is the standard deviation of Gaussian function, i.e. the scale factor of multi-scale filtering device；

To profile point { (x in rectangular coordinate system_i,y_i: i=0 ..., n-1) to carry out Parameter Expression be P (u)=[x (u), y (u)], then the curvature estimation formula of contour curve is:

$k\left(u_i\right)=\frac{\left|x^{\′}{y}^{\′\′}-y^{\′}{x}^{\′\′}\right|}{{(x^{\′2}+y^{\′2})}^{3/2}}$

Wherein x ', y ', x " and y " be defined respectively as:

$x^{\′}\left(u_i\right)=\frac{x(u_i+\mathrm{\Δ}u)-x\left(u_i\right)}{\mathrm{\Δ}u},y^{\′}\left(u_i\right)=\frac{y(u_i+\mathrm{\Δ}u)-y\left(u_i\right)}{\mathrm{\Δ}u}$

$x^{\′\′}\left(u_i\right)=\frac{x(u_i+\mathrm{\Δ}u)+x(u_i-\mathrm{\Δ}u)-2x\left(u_i\right)}{{\left(\mathrm{\Δ}u\right)}^2},y^{\′\′}\left(u_i\right)=\frac{y(u_i+\mathrm{\Δ}u)+y(u_i-\mathrm{\Δ}u)-2y\left(u_i\right)}{{\left(\mathrm{\Δ}u\right)}^2}$

In formula, u is parameter of curve parameter and u ∈ [0,1], u_iFor profile point (x_i,y_iParameter of curve value corresponding to), △ u For the micro-side-play amount on parameter curve, take 0.01；It is determined by the point with extreme curvature on multiple yardstick, detects on profile stable The singular point existed.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 1, it is characterised in that Described step S3 specifically includes:

S31, to profile singular point in the two-dimensional silhouette of all axle positions, carry out three-dimensional cluster analysis, shape on three dimensions Become the cluster of a candidate singular point set；

S32, by clustering procedure, singular point set is carried out cluster analysis, so that it is determined that the center of cluster.

8. CT image body surface handmarking's extraction method in the surgical operation described in claim 7, it is characterised in that described Step S32 uses k-mean clustering procedure singular point set is carried out cluster analysis, specifically includes:

S321, from N number of candidate singular point set, randomly select K as initial cluster center；

S322, to remaining each singular point, measure its distance to each cluster centre, and it be grouped in nearest cluster The heart；

S323, recalculate cluster centre c [i]={ all data [j] sums being labeled as i } of each class obtained/ It is labeled as the number of i；

S324, iteration S32, S23 are until new cluster centre is equal with former cluster centre or is less than appointment threshold value, and algorithm terminates, Obtain the three-dimensional space position of handmarking's candidate point.

CT image body surface handmarking's extraction method in surgical operation the most according to claim 1, it is characterised in that In described step S4, by multi-level Threshold segmentation multiple target discontinuity zone, skeleton and handmarking are put unified minute Cut out, the three-dimensional data of bone segmentation is retrieved singular point cluster and forms handmarking's candidate point coordinate, determine artificial Labelling point region.