WO2014166415A1

WO2014166415A1 - Image guidance method employing two-dimensional imaging

Info

Publication number: WO2014166415A1
Application number: PCT/CN2014/075126
Authority: WO
Inventors: 母治平
Original assignee: 重庆伟渡医疗设备股份有限公司
Priority date: 2013-04-12
Filing date: 2014-04-10
Publication date: 2014-10-16

Abstract

An image guidance method employing two-dimensional imaging, comprising the following steps: selection and screening of a reference characteristic area, searching for a real-time characteristic area, positioning calculation on the basis of the reference characteristic area and of the real-time characteristic area, and, shape distortion estimation and error estimation. When the two-dimensional imaging is employed for image guidance, generation of a two-dimensional DRR image on the basis of a three-dimensional imaging of a focus is required, and then an offset between a current position and a predetermined position is determined by comparing a characteristic area in a real-time image collected in real-time to a characteristic area in the two-dimensional DRR. Use of the present method obviates the need for implantation of a marker into a patient during detection, and, allows for reduced pain for the patient and for reduced treatment risks and costs for the patient.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the field of image guidance, and more particularly to an image guidance method using two-dimensional images.

Background technique

The basic purpose of image guidance is to reproduce the pre-set diseased body position during actual treatment or treatment, referred to as the preset body position. Taking radiotherapy as an example, the doctor collects the patient's 3D CT image, develops a radiotherapy plan based on the image, and selects a point in the image as a treatment center, called the isocenter. The anatomical location represented by these centers needs to be placed in the treatment device, usually at the center of the medical electron linear accelerator (also known as the isocenter, the isocenter of the device). The preset position refers to the position of the patient when the anatomical position of the isocenter representative set by the radiotherapy plan is located at the center of the device. The purpose of image-guided technology is to determine the deviation (or desired position adjustment parameter) of the patient's current position and preset position by real-time acquisition of intraoperative images (hereinafter referred to as real-time images or real-time images) before or during treatment. The doctor or radiotherapy technician can adjust the patient's position based on this deviation to improve the placement accuracy and achieve precise radiotherapy.

There is a type of guidance in image guidance that uses implanted markers for positional reference. The marker is previously implanted into the object to be measured by a doctor through surgery or puncture. The basic steps of implementing image guidance by this method may include: A1) The marker (as opposed to the preset body center) The three-dimensional position can be obtained in advance in a three-dimensional image (such as CT), Rl, R2...Rn, called a preset. Position; A2) Find the location of these markers in the acquired real-time image; the method of finding is described in references [A] and [B]; if using a two-dimensional image for guidance, images at least at two angles Find their position; A3) through the back projection operation, get their position in 3D space at this time, Sl, S2, ... Sn, called the intraoperative position; A4) finally according to the preset position and the intraoperative position It is possible to calculate the deviation of the intraoperative position from the preset position or the required position adjustment. This positional deviation includes displacement in three-dimensional space and may also include rotation about three degrees of freedom. This method is a general method, has a large amount of literature, and has been used in commercial products for many years without further elaboration. The operator adjusts the position of the treatment bed based on the deviation of the intraoperative position from the pre-set position or the desired position adjustment.

The key in this method is to find the corresponding position of each marker in the real-time two-dimensional image. Implanted markers are usually made of heavy metals such as gold, stainless steel, etc., in X-ray images It will exhibit a higher contrast to the surrounding tissue and is therefore easier to distinguish in a two-dimensional image (Figure 5). However, implanting markers is an invasive clinical tool that increases certain patient risks and costs, and its scope of application is limited. Clinically, after the marker is implanted in the body for about one week, after the implant is settled, the preparation for radiotherapy such as three-dimensional image acquisition is performed, and thus the treatment has a certain time delay. And the implanted marker is prone to positional displacement or shedding in the tissue, ie, deviating from its implantation position. The use of these markers in this case introduces large positioning errors and is not easily found. This can be avoided by using a three-dimensional image taken on the patient or by marking a feature area that reflects the patient's own anatomical features on a plurality of two-dimensional DRRs generated from the three-dimensional image.

Summary of the invention

In view of the deficiencies of the above methods, the present invention provides an image guiding method using a two-dimensional image without implanting a marker.

When using 2D images for image guidance, it is necessary to generate a 2D reference image based on the 3D image of the lesion (usually CT), ie DRR (Digital Reconstructed Radiograph), and then compare the real-time acquired 2D image (ie real-time map) with The DRR determines the deviation of the current position from the preset position. On the other hand, when using 2D images for image guidance, it is usually necessary to use 2D images generated from multiple angles and positions to achieve accurate positioning calculations (because if only one angle of 2D images is used, the projection direction is extended). Positional deviations will be difficult to determine accurately). Accurate positioning calculations are typically achieved using two 2D images acquired close to the orthogonal angle. The DRR is a projection imaging process that simulates a two-dimensional radiological imaging system, and the object or patient anatomy represented by the three-dimensional image is generated at a specific position (or relative to the imaging system). The DRR in the above steps may be representative of a plurality of image systems and acquisition angles, and may be a plurality of sets of DRRs generated by placing a patient's three-dimensional image in a plurality of known body positions (relative to the imaging system).

The technical solution adopted by the present invention is that the steps of the image guiding method using the two-dimensional image are:

1) Selection and screening of reference feature regions, which must be determined first in the determination of the isocenter (the process of determining the isocenter is the entire image guidance or normal (non-image guided) radiotherapy process.) 3D images taken by the patient Or selecting a plurality of two-dimensional DRRs generated from the three-dimensional image with a certain grayscale, texture, shape, or color The obvious feature areas of the features are marked (the feature areas (points) marked on the three-dimensional image need to be calculated by the projection relationship on the two-dimensional DRR, and the nearby area is used as the feature area), which has obvious features. The feature area is a small area in the image centered on a certain feature point (the size is 0. 5-6 cm, the shape can be square, circular, spherical or other shape determined according to a certain template), which can be a feature region formed by the marker and a feature region formed near the anatomical structure, or only a feature region formed near the anatomical structure, and then selecting a reference feature region, wherein at least one of the reference feature regions is near the anatomical structure The characteristic area formed. The feature area is defined on the three-dimensional image. It can be selected on the 3D image or selected on the 2D DRR, and the selected area has obvious features on both the 3D and 2D images.

2) searching for the real-time feature area, searching for real-time feature areas corresponding to the reference feature areas in step 1) in the real-time image, and searching the real-time feature areas as real-time feature areas corresponding to all reference feature areas, or A part of the reference feature area corresponds to the real-time feature area of the feature area.

3) Based on the positioning calculation of the reference feature area and the real-time feature area, the position deviation as the image guidance is determined by comparing the real-time feature area on the real-time image with the position of the reference feature area on the two-dimensional DRR. This positional deviation includes displacement in three-dimensional space and may also include rotation about three degrees of freedom.

Further included is an error estimating step for the positional deviation result obtained in step 3), the error estimate including an estimate of the rotational error and the deformation error. The estimation of the rotation error is mainly estimated by calculating the possible error of the isocenter position. The possible error of the isocenter position is 1*ΑΘ, where L is the distance between the center of the reference feature region and the isocenter, and ΔΘ is the rotation error. The rotation error estimate can be determined based on the test results of the device. If the test results show that the angular error of the image guidance system is ± a degree, then A0 = a. The center of the feature area is the center point of the feature area.

The estimating step of the deformation error is: integrating the deformation degree of the tissue on the connection line between the reference feature area and the isocenter according to the position of each reference feature area in the three-dimensional image, and obtaining the reference feature area relative to The possible degree of deformation of the isocenter, and then average or weighted the degree of deformation to obtain an estimate of the degree of deformation of the reference feature region, and then based on the estimate of the degree of deformation gives an isocenter position caused by the deformation Estimate the error, and finally calculate the relative equal center position of the reference feature area according to the look-up table or through the empirical formula. Set the error.

The selection and screening of the feature regions in the above step 1) can be performed on the three-dimensional image. The user can select a feature point in the three-dimensional image for marking, the feature point is located near the anatomical structure with obvious features, and the area centered on the feature point of the mark is used as an optional feature area with obvious features; In the DRR, the projection position of the feature point is calculated according to the projection relationship and marked out; then it is judged whether the image near the position has sufficient feature degree, if it is enough to retain, if not enough, it is discarded and re-selected. The "significant feature" in this application refers to a more significant change in the gray level of the image near the feature point, rather than a flat, untextured area. These features are usually formed by bony tissue. This selection principle is very easy to understand and master for operators familiar with X-ray images, and the subsequent real-time feature area search algorithm has high fault tolerance. Therefore, this method is used to select the reference feature area to guide the image. The positional deviation obtained by the method can fully meet the requirements of image guidance.

The degree of feature can be calculated in a variety of ways, such as local grayscale changes (which can be expressed by variance, standard deviation), information entropy (entropy), contrast, and so on. The specific choice may also need to match the search algorithm of the feature area.

The selection and screening of the reference feature regions can be performed on the two-dimensional image. The steps of selecting a plurality of feature regions having distinct features on the two-dimensional DRR as described in step 1) are: a) selecting one of a plurality of two-dimensional DRRs representing a unified position of the patient generated by the three-dimensional image taken by the patient as The first two-dimensional DRR selects a strong A point on the first two-dimensional DRR as the center point of the feature area with obvious features; b) marks the corresponding A point on the second two-dimensional DRR (This step can be implemented by computer software); c) selecting a strong B point from the projection line as the center point of the feature area having significant features on the second two-dimensional DRR; d) according to step a) And the two center points A and B in c), using the back projection relationship to determine the position of the reference feature area in the three-dimensional image (this step can be implemented by computer software).

The process of selecting and screening out the reference feature area in step 1) may be determined by the operator based on the visual evaluation or automatically by the device responsible for image processing. The automatic selection method may be: calculating the feature degree of each point in a certain range around the center of the three-dimensional image according to the foregoing feature degree calculation method; and selecting several points of the loca l maximum as the candidate feature area. There are two ways to complete automatic screening: one is, Calculating the feature degree of the reference feature area on the three-dimensional image, and setting a threshold value, if the feature degree is less than the threshold value, deleting the reference feature area, otherwise retaining; the other is, generating a plurality of representations representing larger body positions (such as The test DRR with axial deviation ±15mm, ±5° around the axis, etc., to simulate the possible change of the reference feature area in the real-time image when there is a large deviation between the patient's body position and the preset body position; Calculate the reference feature area according to the projection relationship The position in these test DRRs; their similarity in the two-dimensional DRR in all test DRRs and step 1) is calculated separately for each reference feature region; if the similarity in all test DRRs is higher than the preset threshold If the ratio is higher than or reaches a certain preset ratio, such as 80% or 100%, the reference feature area is retained. The above similarity is calculated in many ways, and may include correlation coefficients (correla t ion coeff ic ient), mutual information (mutua l informa t ion ), and the like. The value of the above-mentioned threshold is also related to the method of calculating the similarity. For example, the correlation coefficient can be selected to be 0.7.

Another processing method is to select a plurality of reference feature regions, and after obtaining the real-time image, according to the similarity between the regions in the real-time map and the corresponding reference feature regions, the reference feature region with high similarity is selected, and the similarity is low and difficult to use. Those reference feature areas searched in the real-time graph.

The purpose of searching for the real-time feature area in the above step 1) is to find the position of the reference feature area selected in step 1) in the real-time map. According to the reference feature center point, that is, the position of the feature point in the three-dimensional image and the projection relationship of the imaging system, the projection position of the reference feature area in the DRR can be calculated. According to this projection position, an area of 0. 5-6 cm centered on the feature point can be defined in the DRR, and the size and shape of the area (usually square, rectangular) can be selected and adjusted. Use this small area as a template in the DRR. The search for the real-time feature area is to find the location of the small area most similar to the template in the corresponding real-time map. There are various methods. For example, one method, as described in the literature [A], includes multi-value processing to filter "suspected" regions (blobs), filtering with patterns of shape, size, brightness, etc., vertical axis (super ior-infer Ior axi s) decision, layout (conf igura t ion) determination and other steps. Finally, the blobs in the layout mode giving the best position matching are determined as the corresponding feature areas in the real-time image.

Another method for searching for real-time feature regions is as described in [B], including pre-processing, correlation processing, and extracting local maximum points to form a candidate region list, and using CVA algorithm to select the best matching candidate region as the selection. Corresponding feature area in the real-time image.

The method described in [B] was developed for implanted markers. Clinically implanted markers At the time, the implantation site is required to be around the lesion guided by the image. Thus, these markers are typically located in the central region of the image (eg, the implant marker in Figure 5) in the reference DRR at the predetermined treatment location. However, when image guidance is performed using the anatomical feature regions of the non-implanted markers, these feature regions may be located in the edge regions of the DRR (Fig. 1, Fig. 2, feature area 0). This difference affects the computational association (association, formula 25) steps in the CVA algorithm. In this step it is assumed that the same feature area (implanted marker) has almost the same coordinates on the common axis (X-axis) in the two real-time images. This 4 set is basically applicable to the feature area located in the central area, but a large error is generated for the characteristic area located in the edge area. To solve this problem, the degree of correlation can be calculated according to the principle of the Epipolar line. Assume that the central coordinate of a candidate region φ in real-time graph A is (x _A , yJ, and the central coordinate of a candidate region 实时 in real-time graph B is ( , y _B ). When calculating the degree of association between the two, find the real-time first. In Figure B, the Epipolar line corresponding to φ, find the x coordinate x _P at the line y=y _B , and then calculate the correlation according to x _B and x _P :

(|x _B -Xp|)) , where f (X), β are defined as [Β].

The method of calculating the positioning based on the reference feature area and the real-time feature area in the above step 3) may be various. Assume that the image guidance system uses a real-time image taken at one imaging angle to achieve guidance. One is the real-time map of the i-th (i=l,..N), firstly calculating the central position of each real-time feature area searched on the real-time map, and Pu represents the j-th real-time feature area (j=l, ..n) the position in the i-th real-time map; recalculate and the positional deviation of the center of each real-time feature area in the corresponding DRR, dij; and take the average value of each real-time feature area, di; Finally, according to N real-time The deviation obtained from the current patient position and the preset position is calculated in the figure. For example, [C] gives a method for calculating the three-dimensional spatial position deviation based on άΑ xl, yl ) and d ₂ = (x2, y2) in the stereo imaging system (N=2) (Equation 2).

The second method is to calculate the coordinates of the real-time feature regions obtained from the real-time map in the above-mentioned step A3 in the three-dimensional coordinates S1, .., Sn, and calculate the rest of the corresponding reference feature region preset positions R1, ..., Rn. The position difference, di = Si-Ri; then calculate the average value of the position difference d = ∑di / n; d is the position difference between the current patient position and the preset position.

The third method is to use an optimization algorithm to find a spatial transformation (including translation and rotation) T, so that each reference feature region undergoes a new position after transform T, R, =TR, and the distance between S and S is minimized. This method is described in detail in [D].

Image guidance using this method eliminates the need for implantable markers, making the test non-invasive, reducing patient suffering and reducing patient care costs and costs. At the same time, because no need to wait If the implant is to be placed in the body, the method can also shorten the waiting period for treatment, and the patient can start treatment as soon as possible.

DRAWINGS

Figure 1 is a schematic diagram of the selected reference feature area. The left side of the "+" mark in the figure is the reference feature area on the right side of the reference feature area.

2 is a schematic diagram showing that the information of the reference feature area is weak. The left side of the "+" mark in the figure is the reference feature area on the right side of the reference feature area.

3 is a schematic diagram of a process of selecting a reference feature region from a DRR image; S1 and S2 represent the position of the radiation source; D1 and D2 represent imaging detectors corresponding to the two sources; FIG. 4 is not suitable for use in the three-dimensional image. An example of a reference feature area; the "+" mark in the figure is selected by a two-dimensional image, and has strong features in both DRRs, but in a three-dimensional image in a region where the grayscale is gentle, there is no obvious feature, Shown in the three views of CT, the left picture is the cross section; the upper right is the coronal plane; the lower right is the sagittal plane;

Figure 5 is a schematic diagram of a reference feature area and a real-time feature area of the implanted marker. There are two DRRs on the left side; the real-time map corresponding to the right side, and the implanted markers in the box on the right. detailed description

In the method, the positional deviation is determined by searching some real-time feature regions in the real-time map and comparing the positions of the real-time map with the reference feature region in the DRR in the real-time feature region. These reference feature regions are small regions in a three-dimensional image whose projection in a two-dimensional image (including a real-time map and a two-dimensional DRR) is a small region having features different from the surrounding region. The specific feature description depends on the image mode and image alignment (also called registration or fusion) algorithm. For example, in the X-ray 2D image, the reference feature area can be a large grayscale, unique change. Small area, as shown in Figure 1. The reference feature region selected here is the reference feature region of the non-artificial implant marker. The use of implanted metal or other markers to achieve image guidance in radiotherapy has long been reported and has been widely used clinically. These implants exhibit strong features in 2D and 3D images. The reference feature region in the present invention refers to a feature region formed by the body's own anatomical structure in addition to these implantable markers, usually formed near the bone structure. At least one such reference feature area is selected at the time of selection, and the size and shape of the reference feature area are adjustable.

The selection of these reference feature zones can be done manually. It can be selected by the operator in a three-dimensional image (such as CT) used to generate the DRR, or directly on the DRR. When the reference feature area is selected in the three-dimensional image, the operator selects several feature points in the three-dimensional space as the center of several reference feature areas (which can be implemented by software assistance). According to the projection relationship, the position of these points in the two-dimensional DRR can be calculated, and the selected reference feature regions are marked on the two-dimensional DRR, and then the operator can visually judge whether the feature information of these regions is strong enough, and the selection is retained one by one. Or delete. This may be the case: A reference feature area is reluctant in one of the projected DRRs but weaker in the other DRR. As shown in FIG. 2, the DRR of another angle corresponding to FIG. 1 is shown, wherein the feature information of No. 1 reference feature area is stronger in FIG. 1, but the gray level change is not obvious in FIG. 2, and the feature is weak. In this case, a feature area like the reference feature area No. 1 should be deleted in principle.

This process can also be done directly on the 2D DRR, as shown in Figure 3. The specific steps are as follows: Select a strong feature point on a DRR as the reference feature area center point P1; then mark the corresponding projection line corresponding to the point P1 on another DRR (as an auxiliary means, can be implemented by software) , that is, Epipolar l ine, L2; the operator selects a point with strong features from the projection line L2 (as the center point of the reference feature region selected first in the DRR), P2; The position of these two points determines the center point position P of the feature point in the three-dimensional image with reference to the feature area.

One of the above DRRs corresponds to one of the other DRRs. The information of a point in the DRR (e.g., P1) represents the integral effect of the medium on the line between the source S1 and the detector D1 pixel that collects the point information. This connection is called the projection line of P1. Projection line S1_P1 is projected on D2 in another imaging system consisting of S2 and D2. It is a line formed by the projection of each point on S1_P1 on D2, called the Epipolar line corresponding to P1. Using the basic concepts in computer vision technology, it can be easily calculated.

The back projection relationship, which determines the back projection line based on the projection points P1 and P2 on the two DRRs, and then calculates the spatial intersection point P is a back projection process and is the basic technique in computer vision.

Referring to the screening of the feature area, the reference feature area selected according to the above steps may be further selected to meet the requirements of the feature area. Because the feature area selected by the two-dimensional image may have obvious features on the two two-dimensional images, but there are no significant features on the three-dimensional image, as shown in the "+" mark in Figure 4; some features may be caused by The difference between the postoperative position and the preset position (especially when there is a large angular difference) and a large change in the real-time graph, This makes it difficult to search for such feature areas in real-time graphs. Such feature areas do not meet the requirements. The screening methods for the reference feature area are:

1. Calculate the feature degree of the feature area on the three-dimensional image, and the feature degree may be a change of its three-dimensional neighborhood (variation of gray value) or a measure of information amount. If the characteristic degree does not reach a certain threshold, delete it.

2. Generate a test DRR representing the difference between the postural position and the preset body position (ie, the larger body position difference, including the larger angle and the larger displacement); according to the projection relationship, calculate the position of the feature point in these DRRs, that is, the reference feature area The position in these test DRRs; for each of the reference feature regions, their similarities in the test DRR and in the DRR representing the preset body position are calculated. The operator can decide whether to retain or delete the reference feature area based on the degree of similarity. If the similarity in all test DRRs is high, it can be retained.

The above screening process can also be done automatically by the image processing device. In the above steps, the test DRR is generated, the position of the feature point in the test DRR is calculated, and the similarity of the reference feature area is calculated automatically. When determining whether to retain, you can set a threshold; if a feature area is more similar to the preset threshold in all test DRRs or reaches a preset ratio, it is retained.

When the selected reference feature area and the real-time feature area are non-implanted feature areas, since the projection of the three-dimensional structure on the two-dimensional imaging plane changes with the position and direction of the imaged body, this represents the DRR of the preset body position. The reference feature area may decrease in similarity with the corresponding real-time feature area in the real-time map. This change is gradual and is primarily affected by directional deviations. To solve this problem, a series of DRRs can be generated, representing the expected X-ray image when the imaged body and the preset body position have a certain positional deviation, and then the feature areas in which the projections in the DRRs are not changed greatly are selected as the reference feature areas.

With this method, an estimate of the error that may exist in the resulting positional deviation result can be given. The magnitude of the error and the center of the reference feature zone are related to the distance of the isocenter. The positioning error calculated based on the position of the feature area, especially the rotation error. The rotation error is the error of the displacement part result caused by the possible error of the rotation part of the position deviation, which is amplified by the above distance:

The farther away from the center, the greater the possible error of the isocenter position (the distance is L, the rotation error is ΔΘ, and the possible error of the isocenter position is L*A0). The estimation of the rotation error can be based on experimental data. The rotation error ΔΘ can be estimated experimentally, usually in multiple Know the root mean square of the error between the test result and the known position at the position. In addition to the absolute error, the relative error can be determined experimentally in a similar manner. If the relative error of the rotation result is P% by experiments at a plurality of positions, and the rotation in the positional deviation at this time is estimated to be Θ, then Δθ=θχρ%.

On the other hand, the greater the distance, the more soft tissue is centered between the center of the reference feature and the isocenter, and the greater the possible deformation, the greater the uncertainty of the isocenter position (i.e., the error) determined from the position of the reference feature region. The soft tissue, lung tissue, and bone have different gray values in CT and X-ray images, and can be divided accordingly, and the corresponding degree of deformation can be set (the degree of deformation can be set according to the elastic coefficient of the tissue). According to the position of each feature point (ie, the center point of the feature area) in the three-dimensional CT, the deformation degree of the tissue on the line between the point and the isocenter can be integrated to obtain a possible deformation of the point relative to the isocenter. Degree measure. Then, the deformation degree of all the feature points is averaged or weighted averaged to obtain an estimate of the deformation degree of the currently selected feature region. This deformation estimate can be provided directly to the operator as a reference to the isocenter position error, or an estimate of the isocenter position error caused by the deformation can be given based on this estimate. The final step of estimating the isocenter position error from the deformation degree estimation may be a look-up table or an empirical formula calculation. For example, the measurement of the elastic coefficient of a tissue has a large amount of literature available, such as [Ε]. Both the table and the empirical formula can be determined experimentally.

This kind of error is also related to the specific anatomical position, and the possibility that the error is greater in the part affected by the breathing and peristalsis is greater. The error caused by the motion here is mainly related to the organization between the equal center and the set feature area. This can be achieved by assigning different coefficients of motion to the various anatomical parts, for example, imparting a large coefficient of motion to tissues and organs near the diaphragm, such as the liver, and giving small movements to areas with less motion, such as the intracranial coefficient. This coefficient can directly prompt the user to estimate the motion error. This factor can also be converted to an error value based on an empirical formula.

In order to verify the feasibility of the method of the present invention, the inventors conducted a control experiment using a humanoid motif. This head and neck phantom is made of a material that mimics the head and neck tissue of the human body. Figure 4 is a three-section schematic view of its CT scan, and Figures 1 and 2 are DRR maps based on its CT reconstruction. Soft tissue and bone tissue can be clearly distinguished in these images. A plurality of gold markers (go ld markers) are placed in the phantom. In order to test the accuracy of the image guidance algorithm of this patent application, the following experiment is performed: 1) Fixing the phantom on a precise mobile platform (accuracy is better than 0. 02) Mm); 2) based on a certain position and record the result (dO) given by the image guidance algorithm at that position; 3) the control platform moves to a known position ci (ie the relative displacement with the reference position is known) On, the recorded image directs the result given at these locations (di); 4) calculates the offset of di and dO and compares it with the known position, resulting in the error of the algorithm at that position, ei = (di- dO)-ci; 5) Move the platform to different positions, repeat 3, 4 steps N times, calculate the root mean square of the error, ^ E= sqrt (∑ei ² /N) ₀

The above experiment was carried out by selecting the reference feature region as the implant gold mark and the anatomical feature region proposed by the present invention, and the results obtained were very similar, and the error was better than 0.5 mm in all directions. This experiment was carried out by a domestic authoritative certification testing agency (Shenyang Medical Device Quality Supervision and Inspection Center of the State Food and Drug Administration) and issued a test report to meet clinical use requirements.

references

[A] C. B. Saw, et al, "Implementation of fiducial based image registration in the Cyberknif e robotic system", Med. Dosimetry, Vol 33, No. 2, pp. 156-160, 2008.

[B] Z. Mu, et al., IEEE TMI Vol. 27, No. 9, pp. 1288-1300, 2008.

[C] D. Fu, et al. , "A fast, accurate, and automatic 2D-3D image registration for image-guided cranial radiosurgery", Med.

Phys., Vol. 35, No. 5, pp. 2180-2194, 2008.

[D] M.J. Murphy, "Fiducial-based targeting accuracy for external-beam radiotherapy", Med. Phys., Vol.29, No.3, pp.

334-344, 2002.

[E] E.J. Chen, et al., IEEE E UFFC, Vol. 43, No. 1, pp. 191-194, 1996.

Claims

Rights request

1. An image guidance method using two-dimensional images, comprising the following steps:

1) referring to the selection and screening of the feature region, selecting a plurality of feature regions with distinct features on the three-dimensional image taken by the patient whose isocenter has been determined or on the plurality of two-dimensional DRRs generated from the three-dimensional image, and marking The characteristic regions having obvious features include a feature region formed by implanting the marker and a feature region formed near the anatomical structure, or only a feature region formed near the anatomical structure, and then screening out the reference feature region, the reference feature region At least one of them is a feature area formed near the anatomical structure;

2) searching for the real-time feature area, searching for a real-time feature area corresponding to the reference feature area in step 1) in the real-time image, and searching the real-time feature area as a real-time feature area corresponding to all the reference feature areas, or Corresponding to a real-time feature area of the feature area corresponding to a part of all reference feature areas;

3) Based on the positioning calculation of the reference feature area and the real-time feature area, the positional deviation as the image guidance is determined by comparing the real-time feature area on the real-time image with the position of the reference feature area on the two-dimensional DRR.

2. The image guiding method using two-dimensional images according to claim 1, further comprising: an error estimating step of the positional deviation result obtained in the step 3), wherein the error estimation comprises a rotation error and a deformation error estimate.

3. The image guiding method using two-dimensional images according to claim 2, wherein: the estimation of the rotation error is estimated by calculating a possible error of the isocenter position, and the possible error of the isocenter position is L*Ae. Where L is the distance between the center of the reference feature region and the isocenter, and ΔΘ is the rotation error.

4. The image guiding method using two-dimensional images according to claim 2, wherein: the estimating step of the deformation error is: according to a position of each reference feature region in the three-dimensional image, the reference Integrating the deformation degree of the tissue on the line between the center of the feature area and the isocenter, obtaining the possible deformation degree of the reference feature area with respect to the isocenter, and then averaging or weighting the deformation degree to obtain a reference to the reference The estimation of the deformation degree of the feature region, and then an estimation of the isocenter position error caused by the deformation is given according to the estimation of the deformation degree, and finally the error of the reference feature region relative to the isocenter position is calculated according to the look-up table or by an empirical formula. .

5. The image guiding method using two-dimensional images according to claim 1, wherein: the step of selecting a plurality of characteristic regions having distinct features on the two-dimensional DRR in step 1) is: a) Any one of a plurality of two-dimensional DRRs generated from a three-dimensional image taken by the patient as the first two-dimensional DRR, in the first Select a strong A point on the two-dimensional DRR as the center point of the feature area with obvious features; b) Mark the projection line corresponding to point A on the second two-dimensional DRR; c) Select a feature from the projection line Strong point B, as the center point of the feature area with distinct features on the second two-dimensional DRR; d) based on the two center points A and B in steps a) and c), using the back projection relationship to determine The position of the reference feature area in the three-dimensional image.

6. The image guiding method using two-dimensional images according to claim 1, wherein: the process of selecting and filtering out the reference feature regions in step 1) is automatically performed by a device responsible for image processing.

7. The image guiding method using two-dimensional images according to claim 6, wherein: the device responsible for image processing automatically performs image filtering in two ways: one is to calculate a reference feature region in three dimensions. The feature degree on the image, and set a threshold. If the feature degree is less than the threshold, the reference feature area is deleted, otherwise it is retained; the other is to generate a number of test DRRs representing large body position differences, and calculate reference feature areas in these Test the position in the DRR, and calculate their similarity in the two-dimensional DRR in all the test DRRs and in step 1) for each reference feature area, if the similarity in all the tested DRRs is higher than the preset threshold Or to reach a preset ratio, the reference feature area is retained.

8. The image guiding method using two-dimensional images according to claim 1, wherein: the size and shape of the reference feature area and the real-time feature area are adjustable.

9. The image guiding method using two-dimensional images according to claim 1, wherein: the plurality of two-dimensional DRRs are generated from a plurality of angles and positions according to the three-dimensional image.