CN115359467A - Identification method and device for deformable medical target - Google Patents

Abstract

The invention provides an identification method and device for a deformable medical target, which are used for detecting the deformable medical target in an image under the conditions of partial occlusion, clutter and nonlinear illumination change. An integrated method for deformable medical target detection is disclosed, which combines: a matching metric based on the normalized gradient direction of the model points, a decomposition of the model into parts, and a search method that considers all search results for all parts simultaneously. Although the target model is decomposed into sub-parts, the relevant size of the model used to search at the highest pyramid level is not reduced. Thus, the present invention does not have the speed limitations of the reduced number of pyramid stages that exist in prior art methods. Finally, the identification accuracy and speed of the medical target under the deformation condition are improved, and the accurate needle insertion point positioning of the operation is improved.

Description

A recognition method and device for deformable medical targets

technical field

The invention relates to a recognition method for deformable medical targets, belonging to the technical field of machine vision systems.

Background technique

Minimally invasive puncture surgery is a major change from traditional open surgery. It has the advantages of high operation efficiency, less postoperative complications, less trauma and quick postoperative recovery. At present, traditional surgical puncture is performed under medical imaging based on doctor's experience, and there is a problem of lesion target positioning deviation.

At the present stage of clinical practice, surgical puncture positioning relies on electromagnetic positioning, mechanical positioning, ultrasonic positioning, and optical positioning.

The mechanical positioning device can assist positioning to a certain extent, but due to the existence of the device, the operating space is reduced, and the degree of freedom and flexibility of the device is poor, which affects the implementation of the operation. At the same time, the rigid puncture path positioning , ignoring the influence of tissues and organs on the deformation of surgical instruments, which will easily lead to surgical positioning deviation.

Ultrasonic positioning technology has made great breakthroughs in accuracy and speed, and the cost is low, but the position error judgment often comes from the environment, including temperature, humidity, airflow and noise, etc. These environmental factors are not easy to control, and additional algorithms are required. corresponding compensation.

Magnetic positioning navigation is widely used in the field of minimally invasive medical surgery. This method has high measurement accuracy and good repeatability. However, magnetic positioning navigation is easily affected by the physical characteristics of the patient during surgery, including breathing, heartbeat and movement. At the same time, magnetic positioning navigation is not only costly, but also has a small range of action, which limits the scope of the doctor's operation, and the existence of some necessary magnetic devices and magnetic interference materials seriously affects the positioning accuracy.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the prior art, and provide a method and device for identifying deformable medical targets, which is robust to occlusion, clutter and nonlinear contrast changes, and improves the performance of medical targets under deformation conditions. recognition accuracy and speed.

In order to achieve the above object, the present invention is achieved by adopting the following technical solutions:

In a first aspect, the present invention provides a method for identifying a deformable medical target, comprising the following steps:

(a) acquiring an image of the medical target model;

(b) transforming the image of the target into a multi-level pyramid representation consistent with the recursive subdivision of the search space, the multi-level pyramid including at least the image of the target model;

(c) For each discretization level of the search space, generate at least one pre-calculated target model of the medical target, the pre-calculated target model includes a plurality of model points with corresponding direction vectors, the model points of the target and Orientation vectors are generated by image processing operations that return an orientation vector for each model point;

(d) subdividing the model points into multiple parts, wherein the deformed instance of the target model is represented by transforming the parts; each part is composed of multiple points, and the multiple parts form overlapping point sets ;

(e) obtain the search image;

(f) transforming the search image into a multilevel representation consistent with a recursive subdivision of the search space, said multilevel representation including at least said search image;

(g) performing an image processing operation on each transformed image of the multilevel representation that returns direction vectors for a subset of target model points within the search image corresponding to the search for the at least one The range of transformations that the precomputed target model should target;

(h) Cluster the target model points to form multiple independent parts. Each independent part is matched and calculated within a certain range to obtain the local measurement results, which are obtained by calculating the sum of the results normalized by the number of model points in each part. global matching measure;

(i) determining the target model pose for which the global matching metric exceeds a selectable threshold and whose matching metric is locally maximum, and generating said at least one precomputed target at the coarsest discretization level of the search space based on the model pose a list of instances of the model;

(j) obtaining an estimate of the deformation by obtaining the best fit for each part, computing a deformation transformation describing the local displacement of said part;

(k) tracking instances of said at least one precomputed target model at the coarsest discretization level of the search space through recursive subdivision of the search space until the finest discretization level is reached;

(l) Computing a corresponding deformation transformation at each stage and passing said deformation transformation to the next stage;

(m) Provides deformation transformations and model poses of target model instances at the finest discretization level.

Further, for the deformable target model, the spline function is defined by the displacement, and each set of multiple points is searched independently for the local displacement, all the way along the image to the lowest pyramid level, at the lowest pyramid level, even higher than The resolution of the original image determines the displacement, instantiated at locations with sub-pixel accuracy, where partial displacements are bounded by the magnitude of the gradient, and the target position is determined at a resolution higher than the finest discretization level.

Further, in addition to a user-selectable threshold, only selectable instances of target hypotheses satisfying the deformation requirements are also generated into the list of possible matches at the coarsest discretization level.

Further, in step h, the local measurement result of each part must exceed the user-selectable local threshold, otherwise, the part is considered to be occluded and the part is discarded without further processing.

Further, in step d, k-means clustering or normalized segmentation is used for subdivision.

Further, in step j, the calculated transformation is a perspective transformation.

Further, in step j, according to receiving as input the internal geometric parameters of the imaging device and the metric information of the model, the calculated transformation is a three-dimensional pose.

Further, in step h, the sum of the normalized dot products of the transformed target model part and the direction vector of the search image is used for the local metric result.

Further, in step h, in order to achieve invariance with respect to contrast inversion, polarity information is discarded from the local metric results, the absolute sum of the normalized dot products of the transformed target model part and the direction vector of the search image The sum of the absolute values of values or normalized dot products is used for the local metric results.

In a second aspect, the present invention provides an identification device for a deformable medical target, comprising program code means stored on a computer-readable medium, for executing the method described in the first aspect when the computer program product is run on a computer. A method for the identification of deformable medical targets.

Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

1. The present invention proposes a recognition method for accurately recognizing deformed medical targets under optical imaging. The method incorporates an invariant matching metric, decomposing the target model into parts. Although the target model is decomposed into subparts, the relative size of the model used to search at the highest pyramid level is not reduced. Thus, the present invention does not have the speed limitation of the reduced number of pyramid levels present in prior art methods. Robust to partial occlusion, clutter and nonlinear illumination changes;

2. Even when the target is transformed due to perspective deformation or deformation in a wider sense, the method can identify the target, which improves the accuracy and robustness of target identification.

3. This invention uses the optical positioning method. Positioning is performed through optical instrument navigation, the camera collects the target scene, and the image processing method is used to obtain the attitude of the object in three-dimensional space. This method is low in cost and high in accuracy. This identification method for deformable medical targets is used to detect deformable medical targets in images in the presence of partial occlusion, clutter and nonlinear illumination changes, which can improve the efficiency and accuracy of target detection.

Description of drawings

Figure 1 is a flow chart of a preferred embodiment of the present invention showing the steps of the method;

Figure 2 shows an image of a medical target and the region of interest around the target for model generation;

Figure 3A shows model generation of a target, where the target is on a planar surface;

Figure 3B shows model generation of a target, where the target is occluded by a calibration plate;

Figure 4A shows a set of target model points generated by an edge filter;

Figure 4B shows an example subdivision of a target model point into parts depicting the center of the model and the translation of those parts relative to the center of the model;

Figure 4C shows a typical deformation of the target model due to local movement of the parts in the vicinity;

Figure 5 shows the current image containing two instances of deformations of the target and two instances of target detections found for the model;

Figure 6 shows the deformation map between the rigid template and the deformable template produced by fitting the deformation function, where the example point of the deformation map is the center of the part.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to more clearly illustrate the technical solutions of the present invention, but cannot limit the protection scope of the present invention with this.

The present invention provides a method for deformable medical target recognition that is robust to occlusions, clutter, and nonlinear contrast changes.

Figure 1 gives an overview of the steps of the method. A method for identifying deformable medical targets is provided, comprising the steps of:

Step 1: Obtain the image of the medical target model through the camera;

Step 2: Transform the image of the target obtained above into a multi-level pyramid representation consistent with the recursive subdivision of the search space, wherein the multi-level pyramid representation includes at least an image of the target model;

Step 3: According to each discretization level of the search space, generate at least one pre-calculated target model of the medical target, the pre-calculated target model includes a plurality of model points with corresponding direction vectors, the model points of the target and Orientation vectors are generated by image processing operations that return an orientation vector for each model point;

Step 4: Generate a subdivision that subdivides the plurality of model points into a plurality of parts, wherein the deformed instance of the target model is represented by transforming the parts; each part is composed of a plurality of points, wherein the Subdivision produces overlapping sets of points.

Step 5: Obtain the search image through the camera;

Step 6: transforming the search image into a multi-level pyramid representation consistent with the recursive subdivision of the search space, the multi-level pyramid representation including at least the search image;

Step 7: Perform an image processing operation on each transformed image of the multi-level representation that returns direction vectors for a subset of target model points within the search image corresponding to the search for the at least one The range of transformations that the precomputed target model should target;

Step 8: Clustering the target model points, each independent part is matched within a certain range to obtain the local metric results, and the global matching metric is obtained by calculating the sum of the results normalized by the number of model points in each part, where, For local metrics, parts of the model are searched within a restricted range of transformations close to the precomputed target model, and the best fit for each part is considered as that part's contribution to the global matching metric;

Step 9: Determining target model poses for which the global matching metric exceeds a selectable threshold and whose matching metric is locally maximum, and generating said at least one precomputed target at the coarsest discretization level of the search space from these model poses a list of instances of the model;

Step 10: Obtained the estimate to deformation by obtaining the best match of each part, calculate the deformation transformation describing the local displacement of said part;

Step 11: Track instances of said at least one pre-computed target model at the coarsest discretization level of the search space, until the finest discretization level is reached, by recursive subdivision of the search space;

Step 12: Computing the corresponding deformation transformation at each stage, and transmitting said deformation transformation to the next stage;

Step 13: Provide the deformation transformation and model pose of the target model instance at the finest discretization level.

The target model to be recognized includes a plurality of points with corresponding direction vectors, which can be obtained by standard image processing algorithms, for example by edge detection methods or line detection methods. When generating the target model, the point set is divided into multiple parts. These parts can be moved relative to their original positions during the search, allowing the model to change its shape flexibly. In a preferred embodiment, each part of the target model includes only one model point. In another preferred embodiment, each section comprises a number of adjacent points which are rigid relative to each other.

During the search, the original target model is instantiated eg for a generalized affine pose range. At each location, an instance of the model is deformed by transforming each part independently with a close-range transformation. For each part, a matching metric is computed at each transformation within this restricted range. In a preferred embodiment, the matching metric is the normalized dot product of the direction vector of the portion and the direction vector of the preprocessed search image. The matching metric for the entire target model is the normalized sum of the best-fit parts at the deformation transformations of the best-fit parts. In a preferred embodiment, those parts with a score below a threshold on the matching metric are assumed to be occluded and thus discarded in further processing. The transformation that matches the part with the largest metric determines the deformation of that part relative to the original position. This displacement is used to calculate the pre-selected deformation model. In a preferred embodiment, the model of nonlinear deformation is a perspective transformation. In another embodiment, it is, for example, a spline function or another method known in the art for interpolating or approximating a set of points. Once this transformation function has been computed, the deformation of the found image region can be inverted to produce a rectified image.

The method is divided into an offline phase of generating a target model and an online phase of finding said target model in a search image. The input to the model generation is an example image showing the target in an undistorted manner. In FIG. 2 , an example image 202 of a target shape is shown. A region of interest 201 constrains the position of the target in the image. Typically, this region is defined by the user during the offline training phase. If the user of the target recognition system is only interested in subsequently correcting the target in the search image, sometimes only a part of the target needs to be selected as the region of interest 203. It defines the location and size of this region for the model that must then be unwarped.

The ROI 201 in the image only specifies the location and size of the target in the image. To determine the metric pose of the target, the internal geometric parameters of the imaging device must be provided. The internal geometric parameters of the imaging device 300 (see FIG. 3A ) are typically described by the following aspects: its focal length, the position of the principal point in the image, the size of the pixel elements in the row and column directions, and the simulated pillow caused by the lens. Distortion coefficients for shape distortion or barrel distortion. The internal and external parameters of the camera can be determined in advance by methods known in the art (see e.g. MVTec Software GmbH, HALCON8.0 Documentation, Solution Guide II-F, 3d Machine Vision, 2007).

Once these parameters are determined, the relative pose of the region of interest 203 of the target model 301 in the camera coordinate system is needed to estimate the relative pose of the target (see Figures 3 and 4). Since usually no a priori metric information of the imaged target is available, it cannot be said whether the target is e.g. small and close to the camera or large and far away from the camera. Both cases here will result in the same image. A typical method of providing this metrology information is, in a preferred embodiment, to overlay a measured planar calibration plate 303 on a medical target, and to acquire an image 302 showing the calibration plate (see Figure 3B). In this schematic illustration, the calibration plate 303 contains dark circular defined points. Because the dimensions of the calibration plate and the precise metric positions of the points are known, the relative pose of the calibration plane can be determined in the camera coordinate system. The calibration plate is then removed from the target and a second image is acquired showing the target at the same position as the calibration plate. Since the pose of the calibration plate and the pose used for target model generation are the same in the world and in the image, the corresponding pose of the target is automatically determined. The area of the calibration plate 303 is used directly in conjunction with the image of the target for model generation.

Medical target recognition methods transform model-generated images into recursive subdivisions that contain smoothed and subsampled versions of the original images. In the following expressions, recursive subdivision, multi-level representation and image pyramid have the same meaning when used. In a preferred embodiment, the recursive subdivision is a mean image pyramid. In another preferred embodiment, a Gaussian image pyramid is applied. The same multilevel representation is generated from regions of interest defining the position of the target model. For each multi-level representation, a target model is generated to extract edge points from said region of the image. The edge detection results are shown in Figure 4A. Among them, edge detection not only extracts the position, but also extracts the direction of strong contrast changes. The edge detection used is for example a Sobel filter or a Canny edge detection filter, or any other filter known in the art for extracting directional feature points from an image. The present invention is not limited to edge features, and those skilled in the art can easily extend it to line features or interest point features. For clarity, we restrict ourselves to edge points in further discussion. Small arrows 400 in Figure 4A indicate the location and orientation of the edge points. For each model point, the extracted edge points are transformed into a model coordinate frame (represented by circle 401) and saved into memory. Thus, the recognition method yields a geometric description of the imaged target.

A model coordinate frame 401 defining the origin of the target model is typically computed by taking the center of gravity of the point set. The orientation of the coordinate frame is the same as the target. Therefore, the transformation that maps the model coordinate frame to the template image coordinate frame is a simple translation. Different instances of the model can be projected into the image by applying a generalized affine transformation mapping from the model coordinate frame to the image coordinate frame.

To account for successive nonlinear target model deformations, the plurality of edge points are organized into groups of sub-multiple points. By locally transforming the sub-multiple points of the groups, the spatial relationship of the groups to each other is changed, resulting in a non-linear shape change of the entire target. Here, the local transformations applied to each group are sufficiently small affine transformations, or a subset thereof, such as rigid transformations or translations. The model subdivision of the targets is shown in Figure 4B. The input to the partial generation is a set of edge points 400 previously generated by feature extraction.

Once the edge points are extracted, part of the generative task is to group these points into a spatially correlated structure 403 . Here, the invention assumes that the structure of the spatial correlation remains the same even after transformation. One aspect of the invention is to perform clustering manually. Here, the user will keep similar partial selections as a group. Another embodiment of the present invention performs clustering by an automatic method. A straightforward approach is to set a fixed subdivision of the model and assign points within a subdivision unit to a part. Another approach is to compute a neighborhood graph of the model points and pick a fixed number of closest points within a section. It is important to point out that the invention is not limited to the case where the different sets of said sub-multiple points are separate sets. In a preferred embodiment, a set of sub-multiple points is generated for each point and its nearest neighbors. Regardless of the subdivision method used, the model points are divided into n parts, each part containing _ki model points. In order to speed up subsequent calculations, a data structure is used which contains, for each part, the indices n _ij of the model points it contains. Here, the index i ranges from 1 to n and defines which part is selected; j ranges from 1 to _ki and defines the point of the part. For example each part has the same number of model points, then a matrix representation is used where each row defines a part and each column defines an index into that part.

After defining such a subdivision, the center 402 of each part 403 is calculated, for example by taking the center of gravity of the corresponding point set. The transformation 404 between the center of the part and the origin of the target model 401 is saved in the model. Thus, the relative position of the center of the part is transformed into a transformation that changes the coordinate frame of the model into the coordinate frame of the part, such as a Euclidean transformation. These transformations allow the transformation of the position and orientation of model points from the part's coordinate frame to the model's coordinate frame and vice versa. Changing the relative transformation 404 between the target model and the part, for example by a small movement along the x and y axes or a rotation about the center of the part, allows instantiating a deformed version of the target model. Some example deformations due to small translations in the x and y directions are illustrated in Figure 4C.

One aspect of the present invention is to extend known methods for detecting medical targets in images in the presence of partial occlusions, clutter and non-linear illumination variations.

Compare the directed point set of the target model with the dense gradient direction field of the search image. Even though the non-linear illuminance delivered to the gradient magnitude varies considerably, the gradient direction remains the same. Furthermore, hysteresis thresholding or non-maximum suppression is completely avoided in the search image, enabling true invariance with respect to arbitrary illumination changes. Partial occlusions, noise, and clutter lead to random gradient directions in the search image. These effects lower the maximum value of the score on this metric, but do not change its position. The semantics of the score value is the score of the matching model points.

The idea behind efficient search is that target recognition is only globally instantiated for the generalized affine transformation or a subset of it. The search implicitly evaluates higher order non-linear transformations by allowing local movement of parts and taking the largest response as the best fit. This is illustrated in Figure 5, which shows a search image with two instances of the deformed target model. A transmission transform instance 500 of the model is shown on the left. A more complex arbitrary deformation 501 is depicted on the right. As shown, the locally adapted part 403 approximates the target in the search image. The local transformation between the rigid position of the changing part and the locally fitted position allows representing a wide variety of target model appearances.

An important observation is that by transforming the image into a pyramid representation, only small deformations need to be compensated at each level. For example, even though an object has complex deformations at the lowest pyramid level, the appearance at the highest pyramid level does not change much. On the other hand, if the target has a large deformation, it can be compensated at the highest level. In the present invention, deformations are propagated along the pyramid in a recursive manner. If all higher-level deformations are compensated at higher pyramid levels, the appearance of the target changes relatively little at each level.

Therefore, by splitting the search metric into a global part s _g and a local part s _l . For clarity, we only give the formula for translation, which means that the score is only calculated for each row r and column c. It can be directly extended for generalized affine parameters. As mentioned above, the target model is divided into n parts, each part contains _ki model points.

The global metric is defined as:

Meaning: it is the combination of score values of partial matches calculated for each part defined by index i.

The local matching metric is defined as:

Here, the ij pairs define an index indicating which model point in the target is in which part, where each part has _ki model points. r _ij and c _ij are the row and column displacements of the corresponding model points in the model coordinate system. The local transformation _Tl is used to change the shape of the target model. Typically these are Euclidean transforms with small effects such as 1 pixel translation in each direction. The superscripts m and s define d to be the direction vector of the target model or the corresponding position in the search image.

At each possible pose position, each part has a separate score value, for which the measure is evaluated in a range around its original affine position. The maximum score within the local neighborhood is taken as the best fit for that part. The global metric is obtained by summing the results of the local metrics normalized by the number of model points in each part. A variant of the invention is that a threshold can be set for each section that the section must exceed. Otherwise, the part is considered occluded, so it is discarded without further processing.

其中

in

Another preferred embodiment is when the parts are of different sizes. At this point, one weights the impact of each part by the number of target model points it contains.

A set of global score values for generalized affine transformations also allows determining the approximate position of the target when the exact deformation is not known. Another variant is to discard polarity information from local metric results in order to achieve invariance against contrast inversion. This is done by using the absolute value or the sum of absolute values of the sum of the normalized dot products of the direction vectors of the model and image points in the local metric.

By obtaining the best match for each part, not only the score but also an estimate of the deformation is obtained. These are the local transformations T _l defining the maximum local score. After having the local displacements of each segment, the corresponding nonlinear target model is fitted. Smooth deformations can be computed even for locations where there are no model points. Figure 6 shows an example variant. The center of portion 402 is moved to nearby position 603 . A nonlinear transformation is fitted to these points, transforming the original rigid space (schematically depicted as grid 601 ) into deformable space 602 . This is a well known problem in the art and various solutions based on functional interpolation and approximation have been proposed. Here, an aspect of the present invention is to use only the local displacement of each part as a function point and to fit, for example, a perspective transformation for each point. For deformable target models, the spline functions are defined by displacements. This spline function is, for example, a B-spline function or a thin-plate spline function. The coefficients of these functions are calculated by direct methods. However, if eg thin plate splines are used, a very large liner system must be inverted to obtain the twist coefficients. Therefore, in another preferred embodiment a harmonic interpolation method defined by the deformation of the model points is used. At this point, the displacements of the model points are interpolated into the two images describing the distortion along the row and column directions. The deformation is then patched for areas without model points by a method called blend patching. To smooth out the distortion, the deformation is passed back to the original area of the model point. Thus, not only an interpolation function but also an approximation function is obtained. The advantage of this method is that the running time is only linearly dependent on the size of the target and not, for example, cubically dependent on the number of anchor points as in thin plate splines.

Often, especially for severe deformations, it is impossible to extract the deformations in one step. When a deformation map is given, all model points and corresponding orientations are transformed. Using this transformed target model, each set of submultiple points of the model is now again independently searched for local displacements. This gives a loop of determining small displacements and fitting the model being evaluated until convergence is reached. Typically, convergence is checked by checking whether the displacement becomes smaller than a predetermined threshold. For the defined range of global instances that exceed a threshold and are locally maximal, target hypotheses with position, score, and deformation information are put into a list so that they can be further checked at lower pyramid levels. In a preferred embodiment, not only the threshold on the global score value is set, but also the maximum number of hypotheses generated from the highest pyramid level. At this point, all hypotheses are sorted according to their score values, and only a fixed number of best matching candidates are put into the list of hypotheses for further processing.

Once the exact position and deformation of the target model at a particular pyramid level has been determined, that deformation must be passed along the pyramid to the next pyramid level. This is important so that at lower levels only local deformations of a small search range have to be evaluated. In a preferred embodiment, the original affine models from lower levels are transformed into higher pyramid levels by recursive subdivision. The higher level deformations that have been extracted are applied to the model, while the now transformed model from the lower level is transformed back to its original pyramid level. The search at this level starts with an instance of the model transformed according to the deformation of the higher pyramid level.

Hypotheses are tracked along the image pyramid until the lowest pyramid level is reached. At the lowest pyramid level, the displacement is determined at an even higher resolution than the original image. Accordingly, sections are instantiated at sub-pixel precision locations and the corresponding maximum edge magnitudes in the image are determined. At this point, the partial displacement is no longer limited by the direction of the gradient, but by the magnitude of the gradient. Following the above method, the deformation function is fitted with high accuracy using small displacements. Once the object is found at the lowest level, the position, pose and deformation functions are returned. Additionally, the value of the global scoring function is returned to provide the user with a better measure of how well the target was found.

Embodiment two:

This embodiment provides an identification device for a deformable medical target, which includes a program code device stored on a computer-readable medium, and is used to execute the method for deformable medical targets described in the first embodiment when the computer program product is run on a computer. Identification methods for deformable medical targets.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each process and/or block in the flowchart and/or block diagrams, and a combination of processes and/or blocks in the flowchart and/or block diagrams can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, whereby the The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (10)

1. A method for identifying a deformable medical target, the method comprising the following steps: (a) acquiring an image of the medical target model; (b) transforming the image of the target into a multi-level pyramid representation consistent with the recursive subdivision of the search space, the multi-level pyramid including at least an image of the target model; (c) For each discretization level of the search space, generate at least one pre-calculated target model of the medical target, the pre-calculated target model includes a plurality of model points with corresponding direction vectors, the model points of the target and Orientation vectors are generated by image processing operations that return an orientation vector for each model point; (d) subdividing the model points into multiple parts, wherein the deformed instance of the target model is represented by transforming the parts; each part is composed of multiple points, and the multiple parts form overlapping point sets ; (e) obtain the search image; (f) transforming the search image into a multilevel representation consistent with a recursive subdivision of the search space, said multilevel representation comprising at least said search image; (g) performing an image processing operation on each transformed image of the multilevel representation that returns direction vectors for a subset of target model points within the search image corresponding to the search for the at least one The range of transformations that the precomputed target model should target; (h) Cluster the target model points to form multiple independent parts. Each independent part is matched and calculated within a certain range to obtain the local measurement results, which are obtained by calculating the sum of the results normalized by the number of model points in each part. global matching measure; (i) determining the target model pose for which the global matching metric exceeds a selectable threshold and whose matching metric is locally maximum, and generating said at least one precomputed target at the coarsest discretization level of the search space based on the model pose a list of instances of the model; (j) an estimate of the deformation is obtained by obtaining the best fit for each part, computing a deformation transformation describing the local displacement of said part; (k) tracking instances of said at least one precomputed target model at the coarsest discretization level of the search space through recursive subdivision of the search space until the finest discretization level is reached; (l) Computing a corresponding deformation transformation at each stage and passing said deformation transformation to the next stage; (m) Provides deformation transformations and model poses of target model instances at the finest discretization level. 2. the identification method for deformable medical target according to claim 1, is characterized in that, for deformable target model, spline function is defined by displacement, searches for each group of multiple points independently for local displacement, Proceed along the image all the way down to the lowest pyramid level, where the displacement is determined at even higher resolution than the original image, instantiated at subpixel-accurate locations, where the partial displacement is bounded by the magnitude of the gradient, at resolutions higher than the finest The discretization level of resolution to determine the position of the target. 3. The identification method for deformable medical targets according to claim 2, characterized in that, in addition to user-selectable thresholds, only selectable hypothetical instances of targets satisfying deformation requirements are also generated to the coarsest discrete list of possible matches at the optimization level. 4. The method for identifying deformable medical targets according to claim 3, wherein in step h, the local measurement result of each part must exceed a user-selectable local threshold, otherwise, the part is considered to be blocked and This part was discarded without further processing. 5 . The method for identifying deformable medical targets according to claim 4 , wherein in step d, k-means clustering or normalized segmentation is used for subdivision. 6. The recognition method for deformable medical targets according to claim 4, characterized in that, in step j, the calculated transformation is a perspective transformation. 7. The recognition method for deformable medical targets according to claim 4, characterized in that, in step j, according to receiving the internal geometric parameters of the imaging device and the metric information of the model as input, the calculated transformation is a three-dimensional pose . 8. The method for identifying deformable medical targets according to claim 4, wherein in step h, the sum of the normalized dot products of the transformed target model part and the direction vector of the search image is used for the local Measurement results. 9. The method for identifying deformable medical targets according to claim 4, wherein in step h, in order to achieve invariance with respect to contrast inversion, the polarity information is discarded from the local measurement results, and the transformed The absolute value of the sum of the normalized dot products or the sum of the absolute values of the normalized dot products of the direction vectors of the target model parts and the search image is used for the local metric result. 10. An identification device for a deformable medical target, characterized in that it comprises program code means stored on a computer-readable medium, for executing the method according to claims 1-9 when the computer program product is run on a computer. The described identification method for deformable medical targets.

Images