JP2005305048A - Apparatus and method for positioning patient - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus and a method for positioning a patient with high accuracy. <P>SOLUTION: The method for positioning the patient comprises first steps S1-S3, second steps S4-S5, third steps S6-S9, and fourth step S8 to determine the position of the patient, wherein the first steps being steps wherein images in the body of the patient are input as reference images to provide the bone edge image of a specific region in the reference images, the second steps being steps wherein images in the body of the patient under treatment are input as check images to provide the bone edge image of a specific region in the check images, the third steps being steps wherein the relative amount of shift between the bone edge image of a specific region in the reference images and the bone edge image of a specific region in the check images is referred to as a position-determining parameter, which is changed to calculate the normalized correlation between both the images, so that the correlation values between both the images are calculated against the position-determining parameters, and the fourth step being a step wherein the degree of coincidence between both the images is evaluated from the calculated correlation values to determine the optimal value of the position-determining parameters in the check images. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a patient positioning apparatus and a patient positioning method for positioning a patient during treatment in particle beam therapy.

In the particle beam therapy, highly accurate patient positioning is required because of the dose concentration of the particle beam. In a conventional patient positioning device, using a fluoroscopic image from two directions on the positive side, a reference image captured in advance and a verification image captured at the time of treatment are overlapped, the degree of coincidence is visually compared, and manually, The collation image is moved to a position where the collation image matches the reference image, and the patient is moved based on the amount of movement. For this reason, there is a drawback that the burden on the engineer is large and the positioning time becomes long. Therefore, methods for automating the determination of the degree of coincidence between the reference image and the collation image and the calculation of the amount of movement of the patient necessary for matching both images have been proposed (for example, Patent Document 1 and Non-Patent Document 1). reference.).
In the method disclosed in Patent Document 1, the body surface shape of a patient is imaged by two cameras, and a subtraction image is obtained from a reference image at each camera axis and the latest patient image, thereby causing positional deviation. Is determined as a visual difference in the subtraction image, and the positioning amount is obtained by minimizing this difference.
Further, in the method disclosed in Non-Patent Document 1, a reference image and a collation image are obtained from an X-ray fluoroscopic image, the movement amount is used as a parameter, and the cumulative amount of difference for each pixel of the reference image and the collation image is used as an index. By defining an evaluation function and minimizing this evaluation function, an optimization parameter for alignment is obtained by automatic calculation.

JP 2003-319930 A (page 5-7, FIG. 1)

“Development of automatic image alignment function in patient positioning system”, Akagi et al., Medical Physics Vol. 23 Supplement No. 2 April 2003 (pp. 59-61)

In the conventional patient positioning method as described above, in the automation, the X-ray fluoroscopic image or the camera image of the patient body surface is used, and the difference amount of the pixel value between the reference image and the collation image is used as an index of the positioning parameter. There is a problem that sufficient positioning accuracy cannot be obtained depending on the contrast of the image and the S / N ratio. That is, when an image with poor image quality is used, there is a problem that the difference amount is not an accurate positioning index due to a noise component. In addition, the number of repeated calculations is large in the optimization calculation (the calculation time is long), and there are many cases where the optimization calculation diverges, and there is a problem that it converges to a local solution instead of the optimal solution. Moreover, if the time required for patient positioning becomes long, it becomes a burden on the patient, and the probability that the patient moves during positioning increases. In this case, there is a problem that it takes more time for positioning.
Furthermore, in long-term particle beam therapy, the patient's body surface shape may change during the treatment period due to slimming, obesity, etc. When positioning using the patient's body surface shape, a reference image that serves as a reference In the first place, there is a problem that the latest collation image does not match. Therefore, it is inappropriate to use the body surface shape as a positioning index.
In addition, in the case of component positioning by image processing in the field of FA or the like, it is possible to repeatedly shoot and obtain the best image quality for positioning, but in-vivo images by fluoroscopy or the like in patient positioning When X is used, the amount of X-ray exposure increases, so retaking an image or taking a long time to improve image quality is limited as much as possible, and an image with the best image quality is not always obtained at the time of positioning. There was no problem.

The present invention has been made to solve such a problem, and an object of the present invention is to provide a patient positioning device capable of highly accurate patient positioning.
It is another object of the present invention to provide a patient positioning method that enables highly accurate patient positioning.

  The patient positioning device according to the present invention is an image input unit that inputs an image inside a patient's body, uses an image inside the body that is a reference of the patient as a reference image, obtains a bone edge image of a predetermined region of the reference image, and the patient An image processing unit for obtaining a bone edge image of the collation image using the in-vivo image at the time of treatment of the collation, and a relative movement between the bone edge image of the predetermined region of the reference image and the bone edge image of the collation image The amount is used as a positioning parameter, and while the positioning parameter is changed, a normalized correlation calculation is performed between both images, and a correlation value of both images with respect to each positioning parameter is calculated. An optimum value calculation unit that evaluates the degree of coincidence and determines the optimum value of the patient positioning parameter in the collation image is provided.

  In the patient positioning method according to the present invention, a first step of inputting an in-vivo image serving as a reference of a patient as a reference image and obtaining a bone edge image of a predetermined region of the reference image; A second step of inputting an image as a collation image and obtaining a bone edge image of the collation image, and using a relative movement amount between a bone edge image of a predetermined region of the reference image and a bone edge image of the collation image as a positioning parameter The third step of performing normalized correlation calculation between the two images while changing the positioning parameter and calculating the correlation value of the two images with respect to the positioning parameter, and the degree of coincidence of the two images based on the calculated correlation value And a fourth step of determining an optimum value of the patient positioning parameter in the collation image.

  In the patient positioning device and the patient positioning method according to the present invention, an edge image of a bone part is extracted from an image inside a patient's body, and positioning is performed using bone contour information useful for patient positioning. This has the effect of reducing the influence of contrast differences. In addition, in the bone edge extraction process, while narrowing the edge width and using a constraint condition that considers the continuity of the edge, the bone contour information is emphasized, and information on an area unnecessary for patient positioning is suppressed. High-precision positioning is possible. In addition, since the normalized correlation value between the edge image of the predetermined region of the reference image and the edge image of the collation image is calculated as an evaluation index for obtaining the positioning parameters such as the translation amount and the rotation amount, the edge The correlation value can be obtained stably without being affected by the difference in absolute intensity. Thereby, it becomes possible to perform highly accurate positioning rapidly.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a patient positioning device according to Embodiment 1 of the present invention. The patient positioning apparatus according to the present embodiment uses an image input unit 1 that inputs an image inside a patient's body captured by an X-ray fluoroscopic camera or the like, and an image inside the body that serves as a reference for the patient as a reference image, and a predetermined reference image An image processing unit 2 that obtains a bone edge image of a region and obtains a bone edge image of a collation image using a body image at the time of treatment of a patient as a collation image, and a bone edge image of the predetermined region of the reference image and the collation image The relative amount of movement with the bone edge image is used as a positioning parameter, and the normalized correlation calculation is performed between both images while changing the positioning parameter, and the correlation value of both images for each positioning parameter is calculated. An optimal value calculation unit 3 that evaluates the degree of coincidence of the two images based on the correlation value and determines an optimal value of the patient positioning parameter in the collation image, and an image display unit 4 It is obtain things.
The image processing unit 2 includes a smoothing processing unit 21 that smoothes an image inside the patient's body, an edge extraction processing unit 22 that extracts a bone edge image from the smoothed image, and a predetermined region in the reference image. An ROI (Region of Interest) region setting unit 23 that is set as a region of interest and an image storage unit 24 that stores the obtained bone edge image are provided.

Next, the operation of the patient positioning device according to the present embodiment will be described. FIG. 2 shows a processing procedure and an example of the processing result in the patient positioning apparatus according to Embodiment 1 of the present invention.
In step S <b> 1 (reference image input process), an in-vivo image serving as a reference for a patient captured by an X-ray fluoroscopic camera is input to the image input unit 1 as a reference image 100.
In step S2 (smoothing processing and edge extraction processing), the reference image 100 input by the smoothing processing unit 21 is smoothed, an edge image is extracted from the image smoothed by the edge extraction processing unit 22, and an edge in the edge image is further extracted. A bone edge image 101 of the reference image 100 is obtained by thinning the width and extracting a continuous curve from the edge image.
In step S3 (ROI region setting process), the ROI region setting unit 23 sets a predetermined region in the reference image 100 as a reference mask region, and the bone edge in the reference mask region is obtained from the bone edge image 101 obtained in step S2. The image 102 is extracted, and the extracted bone edge image 101 is stored in the image storage unit 24.
In step S <b> 4 (collation image input process), an image inside the patient's body at the time of treatment captured by the X-ray fluoroscopic camera is input to the image input unit 1 as a collation image 200.
In step S5 (smoothing processing and edge extraction processing), the collation image 200 input by the smoothing processing unit 21 is smoothed, an edge image is extracted from the image smoothed by the edge extraction processing unit 22, and an edge in the edge image is further extracted. While narrowing the width and extracting a continuous curve from the edge image, a bone edge image 201 of the collation image 200 is obtained, and the obtained bone edge image 201 is stored in the image storage unit 24.

In step S2, if the S / N ratio of the reference image 100 is good, that is, if the S / N ratio is larger than a preset value, the smoothing process may be omitted. Similarly, in step S5, if the S / N ratio of the collation image 200 is good, the smoothing process may be omitted.
In step S3, the bone edge image 102 in the reference mask region is extracted from the bone edge image 101 obtained in step S2. However, before performing step S2, the ROI region setting unit 23 performs ROI region setting processing. Then, a predetermined area in the reference image 100 is set as a reference mask area, the reference mask area is extracted from the reference image 100, and the same as that performed in step S2 on the reference image of the extracted reference mask area. The bone edge image 102 may be extracted by performing smoothing processing and edge extraction processing.

  In step S2 and step S5, when extracting the bone edge image, as described above, the edge width is thinned and a continuous curve is extracted from the edge image. As a specific method, The edge width is narrow and the constraint condition considering the continuity of the edge is used. For example, the method described in the literature (J. Canny “A Computational Approach to Edge Detection”: IEEE Transaction on Pattern Analysis and Machine Intelligence, vol. 8, no. 6, pp. 679-6, pp. 679-6, pp. 679-6) is used.

  FIG. 3A is a normal edge image 300 that is not subjected to processing by the Canny method, and FIG. 3B is a bone edge image according to the first embodiment, and is a bone edge image 301 that is processed by the Canny method. is there. The image shown in FIG. 3A is a bone edge image adjusted to a level at which the skull contour can be extracted. However, the bone edge has a width and a large number of isolated point noises exist. On the other hand, FIG. 3B shows a bone edge image 301 in a case where a thinning or continuity constraint condition is added. In this image, isolated points often seen in an X-ray image having a poor S / N ratio are shown. Such noise is suppressed, and thinned edges are extracted so that the outline of the skull can be sufficiently grasped.

  Next, in steps S6 to S9, while using the relative movement amount between the bone edge image 102 obtained in step S3 and the bone edge image 201 obtained in step S5 as a positioning parameter, Normalized correlation calculation is performed between both images, and a correlation value of both images for each positioning parameter is calculated. That is, in step S6, initial values of translation amounts and rotation amounts as positioning parameters are set, and in step S7, normalized correlation is performed on the bone edge image 102 in the reference mask region and the bone edge image 201 of the collation image. Perform the calculation process. In step S8, the calculated correlation value is determined, and the degree of coincidence between both images is determined. If the correlation value is less than or equal to the predetermined criterion, the positioning parameter is changed in step S9, and the normalized correlation value is calculated again in step S7. If this correlation value exceeds the criterion or the convergence condition such as the number of repeated calculations specified in advance is satisfied (step S10), the positioning parameter at that time is determined as the optimum value of the patient positioning parameter. By moving the treatment table or the like based on the determined patient positioning parameter, the patient can be accurately positioned. In step S10, the image display unit 4 displays a reference image and a collation image, or displays a superimposed image 400 of the reference image and the collation image in the reference mask area as shown in FIG. By doing so, visual confirmation can also be performed.

  In the normalized correlation calculation executed in step S7, an image vector F having as components the pixel values (edge pixels: 1, non-edge pixels: 0) of the bone edge image (f) 102 in the reference mask region, An image vector G having each pixel value (edge pixel: 1, non-edge pixel: 0) of the bone edge image (g) 201 of the collation image as a component is generated and translated into an evaluation function defined as The normalized correlation value is obtained by taking the inner product value of two normalized image vectors F and G by inputting the amount and the rotation amount as positioning parameters.

  In Steps S8 and S9, the positioning parameter is changed, and the translation amount and the rotation amount that maximize the normalized correlation value obtained by the normalized correlation calculation according to Equation 1 are obtained. Specifically, as shown in Equation 1, the translation vector between the bone edge image 102 of the reference mask region and the bone edge image 201 of the collation image is t, the rotation matrix is R, and the pixel position vector in the reference mask region is x, a normalized correlation value φ (R, t) between a pixel value f (x) for the position x of the image f and a pixel value g (Rx + t) for the position Rx + t of the image g is obtained, and the normalized correlation value φ (R , T) find the maximum t and R.

  FIG. 4 is a diagram showing a correlation value map for positioning parameters in the patient positioning apparatus according to Embodiment 1 of the present invention. In FIG. 4, the translation amount is used as a positioning parameter, and the rotation amount is not changed. The correlation value changes greatly according to the change of the positioning parameter (translation amount), and it can be seen that highly accurate positioning can be realized by the method according to the present embodiment.

As described above, when positioning is performed using an in-vivo TV image captured by an X-ray fluoroscopic camera, the in-vivo image is generally a smooth image and it is difficult to extract feature information. By extracting the edge image from the image effectively, the bone contour information of the bone region having a high X-ray absorption rate can be extracted. Therefore, this bone contour information can be used as a positioning marker, and the image intensity of the reference image and the collation image can be reduced. The patient positioning amount can be accurately obtained without being limited by the difference in absolute values and suppressing unnecessary information.
In the present embodiment, when the correlation value of two edge images is obtained, each image is processed as a normalized vector having a vector component value of 0 or 1, and therefore the absolute intensity level of the edge It is not affected by the difference.
Further, conventionally, the correlation value is obtained based on the accumulated amount of the difference between the pixel values of the two images. However, in this embodiment, the inner product of the edge images is used instead of the square sum of the density difference values of the images. Since the value is taken, it is possible to obtain a highly reliable correlation value without being influenced by some noise factors. That is, even if there is noise in one edge image, the vector component of the other edge image is almost zero, and the inner product is zero, making it difficult to add.
In general, the calculation amount of the correlation value is enormous. However, in this embodiment, since the calculation is performed using the edge image in which most components are 0 values, the calculation amount can be suppressed. Rapid processing is possible.

Embodiment 2. FIG.
In Embodiment 1, when a bone edge image is extracted in step S2 and step S5 in FIG. 2, a process that combines morphological expansion processing and contraction processing of edge pixels on the image plane is further performed. Further, an image having a narrower width and a continuous curve as an edge can be acquired.
The expansion process is a process of setting a pixel to 1 if there is even one in the vicinity of a certain pixel (for example, surrounding 8 pixels). The contraction process is a process in which even one 0 is present in the vicinity of a certain pixel. If there is, it is a process of setting the pixel to 0. When working with expansion → shrinkage, or shrinkage → expansion, the resulting image is fattened by expansion and thinned by contraction, resulting in almost no change, but when operated with expansion → shrinkage, adjacent discontinuous edges expand. Sometimes continuous. On the other hand, when the contraction → expansion is applied, the isolated noise is removed at the time of contraction. As a specific method of such processing, for example, a processing method described in a document (Rafael C. Gonzalez “Digital Image Processing”, Addison Wesley, pp. 518-524) may be used.

  Therefore, by performing the expansion process and the reduction process, an edge image in which only a strong edge component such as a bone contour is further emphasized can be acquired, and an edge image in which bone contour information that is useful information for patient positioning is enhanced is obtained. be able to.

Embodiment 3 FIG.
In the first embodiment, the positioning parameter is changed in step S9 in FIG. 2, and when the normalized correlation calculation is performed using the evaluation function shown in Equation 1, the newly set positioning parameter is input. When the positioning parameter is a translation parameter and a rotation parameter, instead of changing both at the same time, the correlation value is calculated by changing only one parameter, and after determining the optimal value of the patient positioning parameter for one parameter, the other The correlation value may be calculated by changing one of the parameters, and the optimum value of the patient positioning parameter for the other parameter may be determined. For example, if only the translation amount is changed first to obtain the optimum value of the translation amount, and then the rotation amount alone is changed based on the obtained optimum value of the translation amount, the time cost can be reduced. Such rotation amount calculation can be suppressed, and the optimization calculation time can be shortened.

Embodiment 4 FIG.
When patient positioning is performed on images from a plurality of directions such as a normal direction image and a side direction image, the optimum value of the positioning parameter may be determined by the method shown in FIG. 2 in each direction. When setting the initial value of the positioning parameter in each direction in step S6, for example, the optimal value of the patient positioning parameter obtained for the collation image in one of the positive direction and the side direction is used as the collation image in the other direction. It may be used as an initial value when determining the optimum value of the patient positioning parameter for.
FIG. 5 is a diagram for explaining a patient positioning method according to Embodiment 4 of the present invention. FIG. 5A illustrates the positioning of the forward image, and FIG. 5B illustrates the positioning of the side image. According to the method shown in the first embodiment, the positioning parameters acquired from the forward image are the translation amount x _{0 in} the x direction, the translation amount y _{0 in} the y direction, and the coordinates of the rotation center are (x _c , y _c ). In the case of θ, the translation amount y ₀ in the y direction is set to the side direction when determining the optimum value of the translation amount in the y direction of the side direction image among the positioning parameters of the side direction image shown in FIG. The initial value y ′ _{0 in} the y direction of the image is given.
By doing so, the search range for the optimum parameter is narrowed, and the calculation time can be shortened.
When y ′ ₀ is the amount of rotation θ in the forward direction image cannot be neglected, the positioning translation amount y ₀ of the forward direction image and the longitudinal direction of the forward direction image are considered in consideration of the variation due to the amount of rotation θ in the forward direction image. Using the horizontal length W of the forward image, the positioning rotation center (x _c , y _c ) of the forward image, and the rotation amount θ, and using y ′ ₀ calculated by Equation 2, the lateral image is more accurately represented. An initial value of the positioning translation amount in the y direction can be set.

  Similarly, the same effect can be obtained even when the positioning parameter acquired from the side direction image is given as the initial value of the input parameter to the evaluation function at the time of positioning in the forward direction image.

It is a block block diagram which shows the patient positioning device by Embodiment 1 of this invention. It is a figure which shows the process sequence in the patient positioning device by Embodiment 1 of this invention, and the example of the process result. It is a figure which shows the bone edge image with the case where it does not give with the case where the edge extraction process concerning Embodiment 1 of this invention is given. It is a figure which shows the correlation value map with respect to the positioning parameter in the patient positioning device by Embodiment 1 of this invention. It is a figure explaining the patient positioning method by Embodiment 4 of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Image input part, 2 Image processing part, 3 Optimization calculation part, 4 Image display part, 21 Smoothing process part, 22 Edge extraction process part, 23 ROI area | region setting part, 24 Image storage part

Claims (7)

An image input unit that inputs an image inside the patient's body, and an image inside the body that serves as a reference for the patient is used as a reference image, and a bone edge image of a predetermined region of the reference image is obtained and the image inside the body at the time of treatment of the patient is collated An image processing unit that obtains a bone edge image of the collation image, and a relative movement amount between the bone edge image of a predetermined region of the reference image and the bone edge image of the collation image as a positioning parameter. The correlation between the two images is calculated for each positioning parameter, and the degree of coincidence between the two images is evaluated from the calculated correlation value. A patient positioning apparatus comprising an optimum value calculation unit for determining an optimum value of a patient positioning parameter in the patient. The image processing unit includes an ROI region setting unit that sets a predetermined region in the reference image, extracts an edge image from at least the image of the predetermined region in the reference image, thins the edge width in the edge image, and Extracting a continuous curve from the image to obtain a bone edge image of a predetermined region of the reference image, extracting an edge image from the collation image, thinning the edge width in the edge image, and a continuous curve from the edge image The patient positioning apparatus according to claim 1, wherein the bone edge image of the collation image is obtained by extraction. 3. The patient positioning apparatus according to claim 2, wherein when the bone edge image is obtained in the image processing unit, a process combining a morphological expansion process and a contraction process of the edge image is performed on the image plane. A first step of inputting an in-vivo image serving as a reference of a patient as a reference image and obtaining a bone edge image of a predetermined region of the reference image, an in-vivo image during treatment of the patient being input as a verification image, and the verification image A second step of obtaining a bone edge image of the reference image, a relative movement amount between the bone edge image of the predetermined region of the reference image and the bone edge image of the collation image as a positioning parameter, and changing both the positioning parameters A third step of calculating a normalized correlation between the images and calculating a correlation value of both images with respect to each positioning parameter; and evaluating a degree of coincidence between the images based on the calculated correlation value, and determining a patient positioning parameter in the verification image A patient positioning method comprising a fourth step of determining an optimal value of the patient. In the first step, a predetermined region in the reference image is set, an edge image is extracted from at least the image in the predetermined region of the reference image, the edge width in the edge image is thinned, and a continuous curve is extracted from the edge image. In the second step, the edge image is extracted from the collation image, the edge width in the edge image is thinned, and a continuous curve is extracted from the edge image. 5. The patient positioning method according to claim 4, wherein a bone edge image of the collation image is obtained. The positioning parameter is composed of a translation parameter and a rotation parameter. After calculating the correlation value by changing only one of the translation parameter and the rotation parameter, the optimal value of the patient positioning parameter for the one parameter is determined. 6. The patient positioning method according to claim 4, wherein a correlation value is calculated by changing the other parameter to determine an optimum value of the patient positioning parameter with respect to the other parameter. A method for positioning a patient in each of a positive direction and a lateral direction, wherein an optimum value of a patient positioning parameter obtained for one of the positive direction and the side direction is set to a positive direction or a side direction 7. The patient positioning method according to claim 4, wherein the patient positioning parameter is used as an initial value of the positioning parameter when determining the optimum value of the patient positioning parameter for the other collation image.

Images