A 3D CT / PET image region localization method based on projection dimensionality reduction and SSIM

Abstract

The present invention discloses a three-dimensional CT / PET image region positioning method based on projection dimensionality reduction and SSIM, which belongs to the technical field of calculation, inference or counting. The present invention converts the three-dimensional multimodal image positioning problem into a two-stage two-dimensional positioning problem along the sagittal axis and the coronal axis by averaging the projection data dimensionality reduction, so that the time cost of the algorithm is effectively reduced. At the same time, the second-order convergence advantage of the DFP optimization algorithm is utilized to increase the speed of determining interval parameters during the iterative calculation of regional positioning in the local sliding window, so that the calculation time is further reduced. In the present invention, SSIM is used as a similarity measure, and the consistency of the regional structure of the same scan object is utilized to effectively reduce the computational complexity caused by the grayscale distribution difference of PET and CT different modality images, and exhibits higher positioning accuracy. In summary, the positioning algorithm of the present invention has higher accuracy and speed performance in the three-dimensional CT / PET image region positioning problem.

Description

Technical Field

[0001] The invention discloses a three-dimensional CT / PET image region positioning algorithm based on projection dimensionality reduction and SSIM, relates to computer image processing technology, and belongs to the technical field of calculation, estimation or counting. Background Art

[0002] Modern medical imaging can be divided into two categories: anatomical imaging and functional imaging. Computed tomography (CT) is an anatomical imaging technique characterized by high resolution and the ability to provide anatomical and morphological information about organs and tissues. Positron emission computed tomography (PET) is a functional imaging technique characterized by its ability to provide functional and metabolic information about organs and tissues, but with lower imaging resolution. Clinical practice often requires the registration and integration of complementary regional images from CT and PET to improve diagnostic accuracy and address key needs such as lesion monitoring, treatment verification, intraoperative navigation, and target localization. Due to differences in imaging equipment, angles, time periods, and requirements, images acquired from CT and PET modalities typically exhibit significant differences in field of view and size. Accurately calibrating the position of the imaging target in the larger field of view modality image, which is a key prerequisite for subsequent image registration and data fusion, is crucial.

[0003] Multimodal medical image localization is an extremely challenging task in the field of medical image processing. The large differences between CT images and PET images in imaging principles, grayscale distribution, resolution, and field of view size bring obvious difficulty to regional localization.

[0004] Existing multimodal medical image localization technology usually uses regional localization methods based on image grayscale and regional features to solve the positioning problem of two-dimensional slices of medical images. The positioning target is generally a specific organ or lesion. For the positioning of uncertain local areas, it is often necessary to intervene in the regional localization algorithm by manually defining the positioning range before the positioning of uncertain local areas of medical images can be completed.

[0005] The traditional parameter search method based on image grayscale and regional features, which uses similarity measurement for iterative optimization, usually has the following two limitations: (1) the difference between the target image and the image to be located is small, and there are certain requirements on the size of the initial overlapping area between the target image and the correct positioning area; (2) when it is expanded to solve the three-dimensional multimodal image positioning problem, the time cost is high and it is difficult to ensure the positioning accuracy.

[0006] In summary, the present invention aims to propose a three-dimensional CT / PET image region positioning method based on projection dimensionality reduction and SSIM to overcome the above-mentioned defects. Summary of the Invention

[0007] The present invention aims to address the shortcomings of the above-mentioned background technology and provide a 3D CT / PET image region localization method based on projection dimensionality reduction and SSIM, which can quickly and accurately calibrate the position of the local region corresponding to a target image in a small field of view modality in another large field of view modality image, thereby resolving the technical problems of high time cost and difficulty in ensuring positioning accuracy in 3D CT / PET multimodal image region localization technology.

[0008] The present invention adopts the following technical solutions to achieve the above-mentioned purpose:

[0009] A 3D CT / PET multimodal image region localization method based on projection dimensionality reduction combined with the SSIM similarity measure and the DFP optimization algorithm is used in medical clinical practice to localize the small-field-of-view modal image to the corresponding local area of ​​the large-field-of-view modal image when the multimodal image acquisition fields differ significantly. This allows for field-of-view adaptive automatic localization of images of the two modalities, anatomical imaging CT and functional imaging PET, which present complementary information. This method, which serves as a necessary prerequisite for subsequent precise registration and image fusion to meet diagnostic needs, includes the following eight steps.

[0010] Step 1: Take a pair of 3D volume data from a PET modality and a pair of 3D volume data from a CT modality as 3D original volume data. The 3D volume data with a larger field of view is collected as the 3D fixed image f(X) to be positioned; the 3D volume data from the other modality, i.e., the 3D volume data with a smaller field of view, is collected as the 3D positioning target floating image m(Y).

[0011] Step 2: Perform background noise reduction preprocessing on the 3D volume data of the PET modality and the CT modality respectively.

[0012] Step 3: Using the three-dimensional volume data of the 3D fixed image f(X) to be positioned as a reference, resample and interpolate the 3D positioning target floating image m(Y) so that the voxel spacing of the 3D positioning target floating image m(Y) is consistent with that of the 3D fixed image f(X) to be positioned in each dimension.

[0013] Step 4: Perform average intensity projection (AIP) along the sagittal and coronal axes of the 3D fixed image f(X) to be positioned and the 3D floating image m(Y) to be positioned, respectively, to obtain two-dimensional projections of the two modal image data along the two axes. The principle of the AIP method is described as follows:

[0014] AIP is the average grayscale value of all projection points of continuous multi-layer two-dimensional slice images along the axial ray direction, which is used as the grayscale value of the corresponding point on the projection image. The formula is as follows:

[0015]

[0016] Where AIP(x,y) is the grayscale value at the point (x,y) in the AIP image, and SN r is the number of projection layers, I k (x, y) is the grayscale value of the point (x, y) on the k-th layer of the two-dimensional slice image in the two-dimensional slice image sequence of the original modality image.

[0017] Step 5: Initialize the local sliding window size based on the size of the two-dimensional projection image m'(Y) of the 3D positioning target floating image along the sagittal axis. Set different sliding window initial seeds within the size range of the two-dimensional projection image f'(X) of the 3D fixed image to be positioned along the sagittal axis. Based on each sliding window initial seed, a partial image of the two-dimensional projection image f'(X) of the 3D fixed image to be positioned along the sagittal axis is intercepted. Calculate the structural similarity measure (SSIM) between the partial image of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis intercepted by the sliding window and the two-dimensional projection image m'(Y) of the 3D positioning target floating image along the sagittal axis. Select the position of the sliding window initial seed with the largest similarity measure value as the local sliding window initial position. Given two images x and y, the calculation formula and principle of their structural similarity SSIM are as follows:

[0018]

[0019] Among them, μ x is the average grayscale value of the pixels in image x, μ y is the average grayscale value of pixels in image y, is the pixel grayscale variance of image x, is the pixel grayscale variance of image y, σ xy is the pixel grayscale covariance of image x and image y, c1 and c2 are constants used to maintain stability, c1 = (k1L) 2 , c2=(k2L) 2 L is the dynamic range of the grayscale values ​​of pixels in image x and image y, k1 = 0.01, k2 = 0.03.

[0020] Step 6: In the optimization search module, the DFP optimization algorithm is used to optimize the objective function S(f'(X), m'(Y)) after the SSIM is negated. The parameters of the sliding window position transformation function Trans mainly include the translation parameters of the two degrees of freedom in the two-dimensional projection image. The final local sliding window positioning frame interval parameters of the two-dimensional projection image along the sagittal axis of the 3D fixed image to be positioned after the initialization position in step 5 are obtained through iterative calculation. The DFP optimization algorithm used in the present invention is expressed as:

[0021] The DFP (Davidon-Fletcher-Powell) algorithm is a type of quasi-Newton optimization algorithm. Its core is to obtain an approximate estimate H of the inverse Hessian matrix of the current parameter coordinate point of the objective function in the parameter space through iterative calculation; and when the acquired images of the CT and PET modalities present large differences in field of view and size, compared with other existing iterative optimization positioning methods, it can effectively avoid the local optimal solution from falling and improve the convergence speed; and different from other existing technologies, the improvement of the traditional DFP optimization algorithm in terms of memory cost, the present invention only involves the two-degree-of-freedom translation transformation and the parameter threshold requirements of the image field of view size for the determination of the local sliding window positioning frame of the two-dimensional projection image, and adds parameter boundary constraints in the iterative process to ensure the normal progress of the iterative calculation while constraining the transformation space, further shortening the time cost. In this algorithm, the two-dimensional translation parameters of the sliding window position transformation function Trans in the iteration are represented by (x1, x2), the local sliding window translation transformation parameter vector is recorded as x = (x1, x2), and the gradient vector is recorded as Define the parameter vector along the optimization path single step update amount as s k =x k+1 -x k , the gradient vector update amount is y k =g k+1 -g k , x k+1 、x k is the k+1th and kth iteration values ​​of the local sliding window positioning frame translation parameter vector of the 2D projection image of the 3D fixed image to be positioned along the sagittal axis, g k+1 、g k is the k+1th and kth iteration values ​​of the gradient vector, and the correction formula of the DFP algorithm iterative calculation is:

[0022]

[0023] The DFP algorithm process and formula are as follows:

[0024] ①k is initially set to 0, and the initial parameter vector x0 in the sliding window position parameter space is selected, and the preset parameters are given β′<β<1 and a symmetric positive definite starting matrix H0;

[0025] ②Calculate d k =-H k g k , x k+1 =x k +α k d k , where the step size α k Satisfies the Wolfe criterion: Initial setting step size αk =1, where x k+1 Constraints on the 3D fixed image to be positioned along the sagittal axis 2D projection size [x TL ,x LR ] interval, H k+1 、H k are the k+1th and kth iteration values ​​of the Hessian matrix of the structural similarity measurement function in the parameter space;

[0026] ③ Determine H by the correction formula calculated by the DFP algorithm iteratively above k+1 ;

[0027] ④ Let k:=k+1 and return to step ② to iterate.

[0028] The number of iterations reaches the upper limit or ||g k+1 ||≤ε, where ε is the minimum descent rate set in advance, that is, the target accuracy requirement has been achieved, the iteration ends, and the optimal solution x is returned n , which corresponds to the final translation transformation parameters of the local sliding window used for regional positioning, and combined with the sliding window size to determine the final local sliding window positioning frame interval parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis, that is, to obtain the regional positioning interval parameter results of the two-dimensional projection image of the 3D positioning target floating image along the sagittal axis.

[0029] In step 7, for the two-dimensional projection images of the 3D positioning target floating image m(Y) and the 3D fixed image to be positioned f(X) along the coronal axis, the same processing method as steps 5 and 6 is adopted to obtain the final local sliding window positioning frame interval parameters of the two-dimensional projection images of the 3D fixed image to be positioned along the coronal axis, that is, to obtain the regional positioning interval parameter results of the two-dimensional projection images of the 3D positioning target floating image along the coronal axis.

[0030] Step 8: Comprehensively calculate the regional positioning interval parameter results of the two-dimensional projection images of the 3D positioning target floating image m(Y) along the sagittal axis and the coronal axis to obtain the three-dimensional minimum closed interval, that is, the three-dimensional positioning interval parameters and the three-dimensional regional positioning results.

[0031] The present invention adopts the above technical solution and has the following beneficial effects: the multimodal medical image region localization method proposed in the present invention first performs projection dimensionality reduction processing on the three-dimensional multimodal medical image, converts the three-dimensional multimodal medical image region localization problem into a two-dimensional localization problem, reduces the algorithm complexity, takes the structural similarity measure between the local projection image of one modality three-dimensional medical image and the projection image of another modality three-dimensional medical image as the objective function, optimizes the local sliding window translation parameters through the DFP algorithm to maximize the structural similarity measure, utilizes the second-order convergence advantage of the DFP optimization algorithm to improve the speed of determining the local sliding window interval parameters, thereby shortening the calculation time of positioning iterations within the local sliding window scanning area, and can quickly and accurately calibrate the local region position corresponding to the small field of view modality target image in the other large field of view modality image. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the flow of the 3D CT / PET image region localization algorithm of the present invention;

[0033] FIG2 is a schematic diagram of the three-dimensional volume data of the PET fixed image to be positioned and the CT positioning target floating image, wherein FIG2(a) is the original volume data of the PET fixed image to be positioned, FIG2(b) is the original volume data of the positioning target CT floating image, FIG2(c), FIG2(d) and FIG2(e) are three-dimensional cross-sectional views of the original data of the PET fixed image to be positioned, and FIG2(f), FIG2(g) and FIG2(h) are three-dimensional cross-sectional views of the original data of the positioning target CT floating image, respectively.

[0034] Figure 3 shows the dimensionality reduction projections of the 3D PET fixed image to be positioned and the 3D CT floating image of the positioning target, where Figure 3(a) is the projection of the PET fixed image along the sagittal axis, Figure 3(b) is the projection of the CT floating image along the sagittal axis, Figure 3(c) is the projection of the PET fixed image along the coronal axis, and Figure 3(d) is the projection of the CT floating image along the coronal axis.

[0035] FIG4 is a schematic diagram of the regional positioning result after CT image target positioning is performed on the three-dimensional PET fixed image to be positioned, wherein FIG4(a) is the projection positioning result of the three-dimensional PET fixed image to be positioned along the sagittal axis, FIG4(b) is the projection positioning result of the three-dimensional PET fixed image to be positioned along the coronal axis, FIG4(c) is the final regional positioning three-dimensional interval frame of the three-dimensional PET fixed image to be positioned, FIG4(d) is the volume data matching result of the PET fixed image according to the regional positioning three-dimensional interval frame and the CT target image along the sagittal axis view, and FIG4(e) is the volume data matching result of the PET fixed image according to the regional positioning three-dimensional interval frame and the CT target image along the coronal axis view. DETAILED DESCRIPTION

[0036] The present invention is further illustrated below with reference to specific examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, modifications of various equivalent forms of the present invention made by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0037] Example: A three-dimensional CT / PET image region positioning algorithm based on projection dimensionality reduction and SSIM is as follows: Figure 1 As shown, the specific steps include:

[0038] Step 1: Take a pair of 3D PET and CT modality raw volume data of the area to be positioned. The PET acquisition field of view (corresponding to image size) is larger and is used as the 3D fixed image f(X) to be positioned according to needs; the CT acquisition field of view is smaller and is used as the 3D positioning target floating image m(Y).

[0039] Step 2: Perform background noise reduction preprocessing on the 3D volume data of PET and CT modalities respectively.

[0040] Step 3: Using the 3D volume data of the PET modality 3D fixed image f(X) to be positioned as a reference, the CT modality 3D positioning target floating image m(Y) is resampled and interpolated so that its voxel spacing in each dimension is consistent with that of the 3D fixed image to be positioned.

[0041] Step 4: Perform average intensity projection (AIP) along the sagittal and coronal axes on the 3D fixed image f(X) to be positioned and the 3D floating image m(Y) to be positioned, respectively, to obtain two-dimensional projections of the two modal image volume data along the two axes.

[0042] Step 5: Initialize the local sliding window size based on the size of the two-dimensional projection image m'(Y) of the 3D positioning target floating image along the sagittal axis. Set different sliding window initial position seeds within the size range of the two-dimensional projection image f'(X) of the 3D fixed image to be positioned along the sagittal axis. Use each sliding window initial seed to calculate the structural similarity measure (SSIM) between the intercepted portion of the two-dimensional projection image f'(X) of the 3D fixed image to be positioned along the sagittal axis and the two-dimensional projection image m'(Y) of the 3D positioning target floating image along the sagittal axis. Select the seed point with the maximum similarity as the local sliding window initial position.

[0043] Step 6: In the optimization search module, the DFP optimization algorithm is used to optimize the objective function S(f'(X), m'(X)) after the SSIM is negated. The sliding window position transformation parameters mainly include the translation parameters of the two degrees of freedom in the two-dimensional projection image. The final local sliding window positioning frame interval parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis after the position is initialized in step 5 are obtained through iterative calculation, that is, the positioning interval parameter results of the two-dimensional projection image area of ​​the 3D positioning target floating image along the sagittal axis are obtained.

[0044] In step 7, for the two-dimensional projection images of the 3D positioning target floating image m(Y) and the 3D fixed image to be positioned f(X) along the coronal axis, the same processing method as steps 5 and 6 is adopted to obtain the final local sliding window positioning frame interval parameters of the two-dimensional projection images of the 3D fixed image to be positioned along the coronal axis, that is, to obtain the regional positioning interval parameter results of the two-dimensional projection images of the 3D positioning target floating image along the coronal axis.

[0045] Step 8: Comprehensively calculate the regional positioning interval parameter results of the two-dimensional projection images of the 3D positioning target floating image m(Y) along the sagittal axis and the coronal axis to obtain the three-dimensional minimum closed interval, that is, the three-dimensional positioning interval parameters and the three-dimensional regional positioning results.

[0046] Effect evaluation

[0047] In order to verify the effectiveness of the three-dimensional CT / PET image region localization algorithm disclosed in the present invention, an image pair dataset with a large difference in the acquisition field of view of the two modalities of human three-dimensional PET-CT is used to verify the certain advantages of the method disclosed in the present invention in algorithm performance such as positioning accuracy and speed. Figure 2 is a dimensionality reduction projection diagram of the 3D PET fixed image to be positioned and the 3D CT positioning target floating image. Figure 2(a) shows the original volume data of the PET fixed image to be positioned, and Figure 2(b) shows the original volume data of the positioning target CT floating image. Figures 2(c), 2(d) and 2(e) show the 3D cross-sectional diagrams of the original data of the PET fixed image to be positioned, and Figures 2(f), 2(g) and 2(h) show the 3D cross-sectional diagrams of the original data of the positioning target CT floating image. The 3D multimodal image positioning problem is converted into a 2D positioning problem by the average projection method. The dimensionality reduction projection diagrams of the 3D PET fixed image to be positioned and the 3D CT positioning target floating image are shown in Figure 3. Figure 3(a) shows the projection diagram of the PET fixed image along the sagittal axis, Figure 3(b) shows the projection diagram of the CT floating image along the sagittal axis, Figure 3(c) shows the projection diagram of the PET fixed image along the coronal axis, and Figure 3(d) shows the projection diagram of the CT floating image along the coronal axis. From the perspective of the complexity of the data dimension, the positioning algorithm time is shortened. The cost is effectively reduced. At the same time, the second-order convergence advantage of the DFP optimization algorithm in the present invention greatly improves the speed of determining interval parameters during the iterative calculation of regional interval positioning using a local sliding window, further reducing the calculation time. The present invention adopts SSIM as a similarity measure and effectively reduces the computational complexity caused by the grayscale distribution differences between PET and CT images of different modalities by utilizing the consistency of the regional structure of the same scan object. FIG4(a) shows the projection positioning result of the 3D PET fixed image to be positioned along the sagittal axis, FIG4(b) shows the projection positioning result of the 3D PET fixed image to be positioned along the coronal axis, FIG4(c) shows the final regional positioning 3D interval frame of the 3D PET fixed image to be positioned, and FIG4(d) shows the volume data matching result of the PET fixed image and the CT target image based on the regional positioning 3D interval frame along the sagittal axis. The final regional positioning and the volume data matching results within the interval after CT image target positioning on the 3D PET fixed image to be positioned in FIG4 clearly demonstrate the high accuracy of the positioning algorithm of the present invention.

[0048] It should be noted that the above embodiments are merely examples of specific implementations of the present invention and are not intended to limit the scope of protection of the present invention. Equivalent replacements or substitutions made on the basis of the above technical solutions all fall within the scope of protection of the present invention.

Claims

1. A method for locating regions in three-dimensional CT / PET images based on projection dimensionality reduction and SSIM, characterized in that: The steps include: Step 1: Acquire original PET modality 3D volume data and original CT modality 3D volume data, use the 3D volume data with a large field of view as the 3D fixed image to be positioned, and use the 3D volume data with a small field of view as the 3D floating image of the positioning target; Step 2: performing background noise reduction preprocessing on the 3D volume data of the 3D fixed image to be positioned and the 3D volume data of the 3D positioning target floating image respectively; Step 3: Using the 3D volume data of the 3D fixed image to be positioned as a reference, a voxel spacing uniformity preprocessing is performed on the 3D positioning target floating image by resampling and interpolation; Step 4: Performing projection dimensionality reduction processing on the 3D fixed image to be positioned and the 3D positioning target floating image after voxel spacing uniformization preprocessing, respectively, to obtain two-dimensional projection images of the 3D fixed image to be positioned along the sagittal axis and the coronal axis, and to obtain two-dimensional projection images of the 3D positioning target floating image after voxel spacing uniformization preprocessing along the sagittal axis and the coronal axis; Step 5, intercepting a partial image of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis through a local sliding window, and taking the maximum structural similarity measure between the partial image of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis intercepted by the local sliding window and the two-dimensional projection image of the 3D positioning target floating image along the sagittal axis after the voxel spacing uniformization preprocessing as the goal, optimizing the local sliding window positioning frame translation parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis, and obtaining the regional positioning interval parameter result of the two-dimensional projection image of the 3D positioning target floating image along the sagittal axis after the voxel spacing uniformization preprocessing according to the local sliding window positioning frame interval parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis; Step 6, obtaining the regional positioning interval parameter result of the two-dimensional projection image of the 3D positioning target floating image along the coronal axis after the voxel spacing uniformization preprocessing; Step 7: Combining the regional positioning interval parameter results of the two-dimensional projection image of the 3D positioning target floating image along the sagittal axis after the voxel spacing uniformization preprocessing and the regional positioning interval parameter results of the two-dimensional projection image of the 3D positioning target floating image along the coronal axis after the voxel spacing uniformization preprocessing, obtain the three-dimensional minimum closed interval, and obtain the three-dimensional positioning interval parameters and the three-dimensional regional positioning results from the three-dimensional minimum closed interval.

2. The method for locating a region in a 3D CT / PET image based on projection dimensionality reduction and SSIM according to claim 1, characterized in that: The step 4 uses the average density projection method to perform projection dimensionality reduction processing on the 3D fixed image to be positioned and the 3D positioning target floating image after voxel spacing uniformity preprocessing.

3. The method for locating a region in a 3D CT / PET image based on projection dimensionality reduction and SSIM according to claim 1, wherein: In step 5, the DFP algorithm is used to optimize the local sliding window positioning frame translation parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis.

4. The method for locating a region in a 3D CT / PET image based on projection dimensionality reduction and SSIM according to claim 1, wherein: In step 5, the local sliding window positioning frame interval parameters of the 3D fixed image to be positioned along the sagittal axis two-dimensional projection image are obtained based on the size of the local sliding window of the 3D fixed image to be positioned along the sagittal axis two-dimensional projection image, the initial position of the local sliding window of the 3D fixed image to be positioned along the sagittal axis two-dimensional projection image, and the optimization result of the local sliding window positioning frame translation parameters of the 3D fixed image to be positioned along the sagittal axis two-dimensional projection image.

5. The method for 3D CT / PET image region localization based on projection dimensionality reduction and SSIM according to claim 1, characterized in that: Step 6 obtains the regional positioning interval parameter results of the 3D positioning target floating image along the two-dimensional projection image of the coronal axis after voxel spacing uniformization preprocessing based on the local sliding window positioning frame interval parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the coronal axis. The method for obtaining the local sliding window positioning frame interval parameters of the 3D fixed image to be positioned along the two-dimensional projection image of the coronal axis is the same as the method for obtaining the local sliding window positioning frame interval parameters of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis.

6. The method for locating a region in a 3D CT / PET image based on projection dimensionality reduction and SSIM according to claim 2, wherein: The expression for projecting dimension reduction processing on the 3D fixed image to be positioned and the 3D positioning target floating image after voxel spacing uniformity preprocessing using the average density projection method in step 4 is: Among them, AIP(x,y) is the grayscale value at the midpoint (x,y) of the average density projection image, SN r is the number of projection layers, I k (x, y) is the grayscale value of the point (x, y) on the k-th layer of the two-dimensional slice image in the two-dimensional slice image sequence of the original modality image.

7. The method for locating regions in three-dimensional CT / PET images based on projection dimensionality reduction and SSIM according to claim 3, characterized in that: The specific method for optimizing the translation parameters of the local sliding window positioning frame of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis using the DFP algorithm in step 5 is: Step 5-1, let the local sliding window positioning frame translation parameter vector of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis be x, where x = (x1, x2), where x1 and x2 are the two-dimensional translation parameters of the local sliding window positioning frame of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis; Step 5-2, let the gradient vector be g, Trans(f'(X), x) represents a partial image of the 3D two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis obtained by intercepting the two-dimensional projection image f'(X) of the 3D fixed image to be positioned along the sagittal axis according to the local sliding window positioning frame translation parameter vector x of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis, and S(Trans(f'(X), x), m'(Y)) represents the negative of the structural similarity measure function between the partial image of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis and the two-dimensional projection image m'(Y) of the 3D positioning target floating image along the sagittal axis; Step 5-3, follow the DFP iterative calculation formula Update the Hessian matrix of the structural similarity measure function in the parameter space, H k+1 、H k is the k+1th and kth iteration values ​​of the Hessian matrix of the structural similarity measurement function in the parameter space, s k is the k-th step update of the local sliding window positioning frame translation parameter vector of the 2D projection image of the 3D fixed image to be positioned along the sagittal axis along the optimization path, s k =x k+1 -x k , x k+1 、x k is the k+1th and kth iteration values ​​of the local sliding window positioning frame translation parameter vector of the 2D projection image of the 3D fixed image to be positioned along the sagittal axis, x k+1 The value of y is constrained to be within the size range of the 2D projection image along the sagittal axis of the 3D fixed image to be positioned. k is the kth iteration value of the gradient vector, y k =g k+1 -g k , g k+1 、g k is the k+1th and kth iteration value of the gradient vector.

8. The method for locating a region in a 3D CT / PET image based on projection dimensionality reduction and SSIM according to claim 4, characterized in that: The size of the local sliding window of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis is initialized according to the size of the two-dimensional projection image of the 3D positioning target floating image along the sagittal axis.

9. The method for locating a region in a 3D CT / PET image based on projection dimensionality reduction and SSIM according to claim 4, characterized in that: The method for determining the initial position of the local sliding window of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis is as follows: setting sliding window initial seeds at different positions within the size range of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis, intercepting a partial image of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis according to each sliding window initial seed, calculating a structural similarity measurement value between the partial image of the two-dimensional projection image of each 3D image to be positioned along the sagittal axis and the two-dimensional projection image of the 3D positioning target floating image along the sagittal axis, and selecting the position of the sliding window initial seed that maximizes the structural similarity measurement value as the initial position of the local sliding window of the two-dimensional projection image of the 3D fixed image to be positioned along the sagittal axis.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to claim 1 is implemented.

Images