Brain CT and CTP image registration method based on point cloud technology

Abstract

The invention provides a brain CT and CTP image registration method based on a point cloud technology, and the method comprises the following steps: S1, carrying out the spatial resolution standardization processing of a CT image and a CTP image, carrying out the normalization processing of the gray values of the CT image and the CTP image, and adjusting the gray range to a unified scale; s2, extracting feature points from the CT image and the CTP image which are subjected to mask masking and normalization processing; s3, converting the feature points into three-dimensional point cloud data; s4, performing preliminary alignment on the three-dimensional point cloud data based on a rigid body transformation model, and estimating rotation and translation parameters of the image; s5, combining the similarity of point cloud gray information, fusing the multi-modal image space and gray features, and further optimizing a point cloud alignment result; s6, evaluating the spatial alignment precision of point cloud registration by adopting a geometric error measurement method; through the point cloud technology, the registration problem of brain CT and CTP images caused by differences of resolution, gray scale and modal characteristics is solved.

Description

Technical Field

[0001] The invention belongs to the field of image registration, and in particular relates to a brain CT and CTP image registration method based on point cloud technology. Background Art

[0002] Multimodal medical image registration is an important research direction in medical image processing and is widely used in the diagnosis and treatment of brain diseases. CT (computed tomography) and CTP (cerebral perfusion imaging) are two commonly used imaging modalities in clinical practice, providing anatomical information and cerebral hemodynamic information, respectively. However, due to the significant differences between CT and CTP images in imaging mechanism, spatial resolution, layer thickness, and grayscale range, there are many technical challenges in directly registering the two images.

[0003] Traditional multimodal registration methods mainly include global registration algorithms based on grayscale similarity and local registration algorithms based on feature points. However, these methods have the following problems: Global registration methods (such as mutual information-based registration) have difficulty capturing stable correspondences when the grayscale differences of multimodal images are significant. Local feature methods (such as SURF and ORB) perform poorly in cross-modal images, as the registration accuracy decreases due to unstable feature point matching. Although the rigid transformation model can achieve preliminary alignment, it lacks the ability to integrate multimodal grayscale characteristics and cannot meet the needs of high-precision registration. Summary of the invention

[0004] The purpose of the present invention is to provide a brain CT and CTP image registration method based on point cloud technology. The point cloud technology solves the registration problem caused by the differences in resolution, grayscale and modal characteristics of brain CT and CTP images, realizes high-precision multimodal image fusion, and provides reliable technical support for the diagnosis, treatment and postoperative evaluation of brain diseases, which has broad clinical application value.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for accurate registration of brain CT and CTP images based on point cloud technology, comprising the following steps: S1, performing spatial resolution standardization processing on the CT image and the CTP image respectively, so that images with different resolutions are sampled on a unified spatial scale; and performing normalization processing on the grayscale values ​​of the CT image and the CTP image, adjusting the grayscale range to a unified scale; Specific steps for spatial resolution standardization: Implementation method: Adjust CT and CTP images to a uniform resolution through interpolation algorithms (such as cubic spline interpolation, bilinear interpolation, etc.). For example, adjust the slice thickness of CT images from 5mm to 1.25mm to keep consistent with CTP images.

[0006] Operation goal: Make the voxel spacing of CT and CTP images consistent in the spatial coordinate system, thereby achieving spatial standardization.

[0007] Results (Effects): The differences in spatial resolution were eliminated, and there was a one-to-one correspondence between the voxels of CT and CTP images.

[0008] Gray value normalization processing specific steps: Implementation: Calculate the normalized grayscale value of each voxel, for example, using the linear normalization formula: ,in, I normI Normalized voxel value (i.e. grayscale value), and are the global maximum and minimum grayscale values ​​in the image.

[0009] Operation goal: Adjust the gray value range of CT images and CTP images to a unified standardized scale (such as 0 to 1).

[0010] Effect: The problem of inconsistent grayscale range is solved, providing consistent grayscale input for multimodal image registration.

[0011] Spatial resolution standardization and grayscale normalization are performed through interpolation algorithms and normalization formulas respectively, which are specific technical operation steps. The effect of standardization: After performing these steps, the image achieves consistency in spatial resolution and grayscale distribution, which is the result of standardization.

[0012] S2, extracting feature points with rotation, scale and grayscale change invariance from the standardized and normalized CT images and CTP images; S3, converting the feature points into three-dimensional point cloud data to simultaneously include the spatial position and grayscale texture information of the feature points; S4, preliminarily aligning the above three-dimensional point cloud data based on a rigid body transformation model, and estimating the rotation and translation parameters of the image; S5. Based on the preliminary alignment, the similarity of the grayscale information of the point cloud is combined to fuse the multimodal image space and grayscale features to further optimize the point cloud alignment result; S6. The geometric error measurement method is used to evaluate the spatial alignment accuracy of point cloud registration, and the distribution similarity measurement method is used to verify the fusion effect of image grayscale information.

[0013] Furthermore, in the spatial resolution standardization process of step S1, the CT image I CT (x,y,z) and CTP image I CTP(x, y, z) are subjected to trilinear interpolation processing respectively to make the two consistent in resolution and layer thickness. The grayscale value of the interpolation point (x, y, z) is calculated by the following formula: For the interpolation point (x, y, z), determine its eight nearest neighboring points (x i ,y j ,z k ), (x i ,y j ,z k ) represents i neighboring stores, using interpolation weights Calculate the difference weight of the interpolation point and use the trilinear interpolation formula Calculate the gray value of the interpolation point; The trilinear interpolation formula is based on the three-dimensional linear interpolation theory, and ensures the continuity and smoothness of the image data through weighted summation, effectively eliminates the difference in spatial sampling between CT and CTP, and provides a unified spatial scale for subsequent feature point extraction.

[0014] Furthermore, the normalization processing of the grayscale values ​​of the CT image and the CTP image includes: performing a normalization mask on the image CT value obtained after the interpolation according to the following formula:

[0015] in, Indicates that according to the window width and window level The mask that masks the CT and CTP values, normalizing only the unmasked parts, Represents the gray value of the midpoint of the CT image, It is the grayscale value of the voxel in the CT image. Unifying the brightness range of CT and CTP images to the interval [0,1] reduces the cross-modality matching problem caused by the different grayscale dynamic ranges of CT and CTP images, and ensures the comparability of subsequent feature point extraction and point cloud generation in the grayscale dimension.

[0016] x, y, z represent the three-dimensional coordinates of the interpolation points at a uniform resolution (in the target coordinate system). These coordinates are usually located between the original voxel points, so their grayscale values ​​need to be calculated through interpolation.

[0017] Original coordinate system: : The two original sampling points adjacent to the interpolation point in the x direction; : The two original sampling points adjacent to the interpolation point in the y direction; : The two original sampling points adjacent to the interpolation point in the z direction. The grayscale value of the interpolation point (x, y, z) is calculated by trilinear interpolation from the grayscale values ​​of the eight original voxel points where it is located.

[0018] Through trilinear interpolation, the resolution and layer thickness of CT images and CTP images can be adjusted to a unified scale, so that the two can correspond to each other in spatial position. This processing can effectively eliminate the impact of resolution differences on subsequent registration and provide consistent input data for feature extraction and point cloud generation.

[0019] Summary: The gray value of the interpolation point (x, y, z) is calculated by the trilinear interpolation formula. The weights u, v, w in the formula represent the relative position of the interpolation point in each dimension, and the gray values ​​of the eight adjacent voxel points I000 to I111 participate in the weighted calculation. This process ensures the unity of CT and CTP images in resolution and layer thickness, laying the foundation for subsequent processing.

[0020] Further, in step S2, the feature point extraction includes: Normalized image Perform Gaussian blur convolution to construct scale space; ;in, is a Gaussian kernel of scale σ, It represents the blurred image of the image after Gaussian kernel filtering at a certain scale σ; Then, through the scale difference formula Find local extreme points and get the coordinates of feature points , so that key points that are robust to noise, rotation and scale changes can be detected at different resolutions, laying the foundation for subsequent point cloud generation.

[0021] Further, in step S3, when the feature points are converted into three-dimensional point cloud data, the feature point coordinates are The corresponding normalized grayscale value and local descriptor information combination is expressed as: ,in, I normI is the normalized gray value of the feature point, D It is a local texture or gradient descriptor; this point cloud form can simultaneously represent the spatial position and modal texture information of feature points in the same coordinate system, providing a unified data structure for multimodal image fusion and improving the registration efficiency.

[0022] Further, in step S4, when the three-dimensional point cloud data is preliminarily aligned, the rigid body transformation model is used. ,in, P Represents the spatial coordinate vector of a point in the point cloud, R is the rotation matrix, tThe rotation and translation parameters are adjusted in sequence to eliminate the angle and position errors caused by CT and CTP during shooting, ensuring the initial alignment consistency of the image in the three-dimensional coordinate space.

[0023] Furthermore, the rigid body transformation parameters R and t are optimized by the iterative closest point (ICP) algorithm, using the objective function ;in, and are the corresponding point pairs selected by the closest point strategy in the CT point cloud and the CTP point cloud. N is the number of corresponding point pairs; the ICP algorithm is based on the least squares principle, which can gradually reduce the distance error between corresponding points during the iteration process until it converges to a high-precision geometric registration result, laying a reliable spatial position foundation for cross-modal fusion.

[0024] Further, in step S5, during the point cloud alignment process, the following comprehensive objective function is adopted: , where E ICP is the geometric distance error, α and β is the weight coefficient, represents the grayscale similarity, I CT and I CTP The information comes from the CT image and CTP image carried by the point cloud respectively; this comprehensive goal introduces cross-modal grayscale consistency on the basis of geometric position alignment, and by considering the dual factors of space and texture intensity, CT and CTP are simultaneously fused in position and brightness distribution, thereby improving the final multimodal registration accuracy and robustness.

[0025] The present invention conducts a systematic difference analysis of brain CT (computed tomography) and CTP (cerebral perfusion imaging) in terms of resolution, grayscale range, texture features, etc., and uses multiple methods such as unified sampling, cross-modal feature detection and point cloud fusion to significantly enhance the accuracy and robustness of cross-modal image registration, which has considerable application value in clinical diagnosis and scientific research analysis. The following describes the beneficial effects brought about by the present invention in detail from multiple dimensions, and the length is relatively long to ensure sufficient explanation and logical coherence.

[0026] First, at the level of resolution and grayscale normalization, the present invention adopts two major strategies, trilinear interpolation and linear normalization, at the beginning. For brain CT images, the thickness is usually large and the pixel sampling is relatively sparse. In order to capture real-time blood flow characteristics, CTP uses a smaller layer thickness and dynamically collects the time dimension, which leads to great differences in spatial resolution and brightness distribution between the two. If the CT image and the CTP image are directly superimposed or features are extracted at the original resolution, it is likely that the corresponding position cannot be found on some faults, or because the CT brightness is concentrated in the range of bones and soft tissues, while the CTP has brightness fluctuations in high or low perfusion areas, it is impossible to achieve effective matching. The present invention carefully designs a trilinear interpolation formula so that the grayscale value of each interpolation point is obtained by the weighted sum of its adjacent coordinates, fully ensuring the smoothness and consistency of the image in three-dimensional space; coupled with linear normalization, the grayscale of CT and CTP is stretched to the same range. This "starting line consistency" approach greatly reduces the coordinate and brightness mismatch caused by different resolutions and grayscales, allowing any subsequent cross-modal feature detection and point cloud fusion to be performed under a unified metric standard. The benefit of this is that it fundamentally solves the contradiction between CT and CTP: "one large grid, one small grid; one with concentrated brightness, the other with a very large dynamic range of brightness", allowing subsequent steps to focus more on the texture difference itself, rather than dealing with the confusion of basic sampling first.

[0027] Next, at the level of feature extraction and point cloud generation, the present invention uses a multi-scale differential detection method to mine key points that are significant at multiple fuzzy scales, targeting the inherent differences between the CT and CTP modalities of "anatomy vs. hemodynamics". Specifically, by performing a series of Gaussian smoothing on the normalized image and then performing differentials, local extreme points are found to ensure that points that appear as significant edges of bone or soft tissue on CT and points that appear as significant differences in perfusion on CTP can be captured in multi-scale images. This multi-scale detection allows a relatively stable cross-modal correspondence to be formed at certain vascular bifurcations, ventricular edges or lesion locations, even if the texture mechanisms of CT and CTP are different. Then, these key points are converted into point clouds, and the coordinates and descriptor information are recorded together. The advantage is that there is no need to consider the synchronization of resolution or grayscale again when performing subsequent rigid body transformation and alignment in a three-dimensional coordinate system, thereby greatly improving the degree of freedom and accuracy of feature matching. Compared with the traditional method based only on mutual information or only on ordinary corner points, the present invention directly corresponds the "co-occurrence areas" of anatomical-blood flow features in three-dimensional space in the form of point clouds, making subsequent alignment more intuitive and efficient.

[0028] At the level of rigid body transformation and grayscale fusion, the present invention first uses the iterative closest point (ICP) method to minimize the geometric distance, thereby quickly correcting the deviation between CT and CTP in shooting position and angle; then, in order to ensure the correct correspondence of blood flow distribution in CT space, a cross-modal grayscale similarity term is introduced on the basis of geometric error. The purpose of this is that only geometric alignment may cause brightness mismatch in high perfusion areas, affecting lesion localization; and if only grayscale similarity is focused on, overall spatial alignment may be sacrificed. The fusion objective function proposed by the present invention incorporates the two into the same optimization framework, and under the dual conditions of "spatial accuracy" and "grayscale consistency", the anatomical structure of the CT image and the blood flow information of the CTP image are more perfectly superimposed. This move has significant significance in clinical brain diagnosis: doctors can clearly see the anatomical positions such as bones and brain parenchyma, and can also observe the real distribution of cerebral blood flow at the same time, which is conducive to discovering the location of cerebral infarction, evaluating vascular patency, and formulating personalized interventional treatment plans.

[0029] Furthermore, the objective measurement of the registration results of the present invention is also evaluated through two dimensions: geometric error and grayscale distribution similarity. Geometric error can ensure that the CT and CTP point clouds achieve high-quality overlap in three-dimensional coordinates. Once the error exceeds a certain threshold, it means that the initial alignment or subsequent grayscale fusion relying on rigid body transformation is still insufficient. Grayscale distribution similarity (or mutual information, correlation coefficient, etc.) can detect the degree of fusion of images in brightness or texture. Some doctors who use blood perfusion areas as the main observation target can quickly find out whether there is a situation where the local blood flow is too low or too high and does not match the anatomical position, so as to determine whether there are serious diseases such as vascular stenosis, insufficient blood supply or local bleeding. It can be seen that at the evaluation level, the dual evaluation of "position + grayscale" undoubtedly improves the reliability of registration, and can also better compare the advantages and disadvantages of different registration schemes in scientific research offline testing.

[0030] On this basis, the beneficial effects of the present invention can be summarized as follows: One-time solution to multiple inconsistencies: This invention takes into account three factors: resolution difference, grayscale dynamic range difference, and modal texture difference. Through S1 and S2 steps, the grid and grayscale are unified at the initial level, and then multi-scale feature point detection is used to select cross-modal comparable key points. Finally, point cloud + rigid body + grayscale fusion is used for fine matching, avoiding simple repair of single points or single links, and the system is more complete and stable.

[0031] Dual alignment of space and brightness: Compared with the traditional global algorithm based on grayscale similarity (such as mutual information), which is easy to fail when the CTP texture changes greatly, the present invention first uses point cloud ICP for geometric alignment, and then uses grayscale similarity to supplement the microscopic brightness distribution, so that the registration process can achieve a balance between "macro position alignment" and "local texture matching". CT and CTP not only overlap in the skeleton, but also achieve higher consistency in brain parenchyma and even blood flow details, which is of great help to real clinical needs.

[0032] Stronger robustness: The present invention has obvious robustness advantages in details such as multi-scale feature extraction, iterative nearest point, and fusion grayscale optimization. For example, multi-scale features can avoid feature point mismatches caused by noise and too fine textures to the greatest extent; ICP can still converge to a reasonable alignment result through iteration when the posture deviation is large; and the grayscale fusion term can further correct the matching of the spatial point cloud when the perfusion value changes extremely. These characteristics enable this method to maintain a high registration accuracy under various actual conditions (such as patient movement, differences in scanning conditions, dynamic blood flow fluctuations, etc.).

[0033] The results can be quantitatively evaluated: through the dual measurement of geometric error and grayscale distribution similarity, the present invention allows users to clearly know "Is the position accurate? Is the brightness blended well?". If the spatial alignment is good in a case, but the grayscale similarity is low, it means that there is still a certain deviation between the blood flow and the anatomical structure, and the S5 grayscale fusion parameters can be fine-tuned; if the geometric error is large, it means that the number of ICP iterations can be appropriately increased or the key point pairs can be better selected. In this way, purely subjective judgments are effectively reduced, and it is more repeatable and objective.

[0034] Suitable for both clinical application and scientific research verification: On the one hand, in actual clinical applications such as thrombolysis of stroke, blood supply assessment of brain tumors, and blood flow analysis of aneurysms, since the images fused with CT and CTP can intuitively show the correspondence between the anatomy and blood flow of the lesion site, doctors can take timely measures; on the other hand, researchers can also use the present invention for offline testing under different noise levels and different perfusion parameters, and combine dual evaluation indicators to objectively measure the performance of the registration algorithm.

[0035] It can be seen that the present invention has comprehensive beneficial effects such as "eliminating multiple modal differences at one time, taking into account macroscopic position and local texture, providing quantifiable objective evaluation, being highly robust and adapting to real clinical needs", etc., and provides a systematic and efficient technical solution for brain CT and CTP image registration in terms of resolution synchronization, grayscale normalization, cross-modal feature extraction, rigid body transformation, grayscale fusion and evaluation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0037] The differences between CT and CTP images are reflected in the differences in spatial resolution, layer thickness, and gray value distribution.

[0038] Differences in spatial resolution and slice thickness: CT images usually use a larger slice thickness, while CTP images have a smaller slice thickness, resulting in different spatial resolutions of the two images, and the spatial positions cannot be directly matched. Differences in grayscale value distribution: CT images mainly reflect anatomical structures, and their grayscale value range is concentrated and stable; CTP images reflect the dynamic perfusion state of the brain, and their grayscale values ​​are affected by blood flow parameters and vary greatly. The inconsistency of grayscale scale increases the complexity of image registration.

[0039] The core of multimodal image registration is how to accurately find the correspondence between the two modal images. The main challenges include: Inconsistent resolution and grayscale: Due to the different resolutions and grayscale ranges between CT and CTP images, large spatial and grayscale errors will occur during direct registration. It is necessary to unify the spatial resolution and grayscale scale to solve the problems of resolution mismatch and grayscale information inconsistency, and provide a consistent basis for subsequent feature extraction.

[0040] Differences in modal features: CT images mainly provide anatomical information, while CTP images reflect the hemodynamic characteristics of the brain. The two differ significantly in texture characteristics and spatial structures. It is necessary to extract robust multimodal features to ensure the accuracy of feature point matching across modal images.

[0041] Complex spatial alignment: Since the spatial positions of CT and CTP images may be different during acquisition, registration requires the simultaneous estimation of rotation and translation parameters. It is difficult to achieve high-precision alignment relying solely on spatial information, and optimization needs to be combined with the grayscale and texture characteristics of the image.

[0042] Consistency of multimodal fusion: After completing the spatial alignment, it is necessary to ensure the fusion consistency of the image in terms of grayscale information and texture information to ensure the accuracy and reliability of the registration results.

[0043] Therefore, the present invention solves the registration problem caused by differences in resolution, grayscale and modal characteristics between brain CT and CTP images through point cloud technology, realizes high-precision multimodal image fusion, provides reliable technical support for the diagnosis, treatment and postoperative evaluation of brain diseases, and has broad clinical application value.

[0044] The present invention discloses a brain CT and CTP image registration method based on point cloud technology. By making innovative designs in six steps (S1-S6) including image preprocessing, feature extraction, three-dimensional point cloud generation, rigid transformation alignment, cross-modal grayscale fusion and result evaluation, the differences between CT images and CTP images in resolution, grayscale range and modal texture characteristics are overcome, and finally multi-modal fusion with highly consistent anatomical position and blood flow distribution is achieved.

[0045] In step S1, the present invention first solves the inconsistency problem between CT and CTP images in terms of spatial resolution and grayscale dynamic range. This is because if the resolution and brightness range are not synchronized, the subsequent feature detection and point cloud alignment will fall into the dilemma of coordinate mismatch or grayscale comparison. Specific technical means include: (1) Perform trilinear interpolation on CT images and CTP images respectively to keep them consistent in resolution and slice thickness; the interpolation formula is: , Among them, (x i ,y j ,z k ) represents the original grid coordinates adjacent to the interpolation point (x, y, z). In this embodiment, the number of adjacent points is selected as eight points around the interpolation point. w ijk The weights are based on the distance weighted sum to ensure the smoothness and continuity of the image data; (2) Use linear normalization on the interpolated image

[0046] in, Indicates that according to the window width and window level The mask is used to mask the CT value and CTP value, and the unmasked part is normalized to compress the brightness to the [0,1] interval. The meaning of this is that through these two steps, CT and CTP can be subsequently aligned on "uniform resolution + uniform grayscale" without being restricted by the original sampling and brightness differences.

[0047] Then, in step S2, considering that CT emphasizes the anatomical edges of the brain more and CTP focuses more on hemodynamic changes, the present invention adopts a multi-scale feature point detection method to extract key points that are less sensitive to cross-modal differences. The core approach is to perform a multi-scale feature point detection on the normalized image Inorm(x, y, z) at different Gaussian smoothing scales. σ Next Generation .

[0048] On the other hand, using the scale difference formula ,Find local extreme points (x, y, z) in the differential image. The reason for this is that only points that are significant under multi-scale blur are more likely to correspond to the same physical position in CT and CTP (such as the edge of brain tissue or the area of ​​blood flow mutation), thus laying a reliable feature point foundation for subsequent cross-modal matching.

[0049] After acquiring the feature points, step S3 converts these key points into three-dimensional point cloud data to simultaneously record spatial coordinates and texture information. Specifically, the present invention converts the coordinates (x, y, z), normalized grayscale I norm (x,y,z) and gradient descriptors D (such as gradient direction histogram) is represented as In this way, both the position and the modality grayscale characteristics can be saved in a unified data structure. The reason why both CT and CTP are converted into point clouds is that it is more natural to make feature point correspondence and geometric alignment in the three-dimensional coordinate space, without having to worry about the resolution alignment or grayscale difference of the image matrix; in addition, the point cloud format can also significantly reduce useless background information, focus on the truly matchable feature points, and improve computing efficiency.

[0050] However, in the multimodal registration process of brain CT and CTP images, due to the differences in shooting angles and positions, there are usually significant angle deviations and position errors between the two images in the spatial coordinate system. To solve this problem, step S4 uses a rigid body transformation model to perform preliminary alignment of the 3D point cloud data by adjusting the rotation matrix. R and translation vectors t Eliminate this type of error. The core formula of rigid body transformation is P'=R.P+t, where P Represents the spatial coordinate vector of a point in the point cloud, R is a 3×3 orthogonal matrix used to adjust the direction of the point cloud. t Represents the translation vector, which is used to correct the position of the point cloud. In order to make the initially aligned point cloud as close as possible to the target point cloud, the rigid body transformation model minimizes the distance error objective function between the matching points. The process first finds matching point pairs between point clouds through nearest neighbor matching. P i and Q i , then use singular value decomposition (SVD) to optimize the rotation matrix R , and optimize the translation vector based on the centroid difference t , and finally obtain the preliminary aligned point cloud data.

[0051] Although the two point clouds are initially aligned in space through rigid body transformation, it is difficult to fully achieve the consistency of multimodal registration using only spatial information, because CT images mainly reflect anatomical structures, while CTP images reflect brain perfusion status. Therefore, step S5 further optimizes the alignment result by fusing the spatial and grayscale features of multimodal images based on the initial alignment and the similarity of the grayscale information of the point clouds. The optimization process is performed through the objective function Conduct, among which w i Represents the grayscale similarity weight of the matching point. The weight calculation formula is: , h ( P i )and h ( Q i ) are the grayscale histograms of the matching points. This step makes the point cloud alignment result more accurate in space and more consistent in texture and grayscale features through dynamic weight adjustment of grayscale information similarity.

[0052] After the point cloud data is accurately aligned, the registration results need to be verified to ensure that they meet the clinical needs of multimodal image fusion. Step S6 evaluates the registration results from two aspects: spatial alignment accuracy and grayscale consistency, using the geometric error metric method and the distribution similarity metric method. The geometric error metric uses the mean square error (MSE) as the evaluation index, and the calculation formula is: ,in P i and Q i MSE is used to measure the alignment error of two point clouds in spatial position. Gray consistency verification calculates the distribution similarity of gray histograms through Jensen-Shannon divergence (JSD), and the formula is: is the Kullback-Leibler divergence, which indicates the difference between two grayscale distributions. Through the above geometric and grayscale evaluation indicators, when the MSE is less than the set threshold and the JSD meets the consistency standard, the registration result is considered successful.

[0053] Through the above steps, the rigid body transformation model realizes the correction of spatial errors in the preliminary alignment stage, and the grayscale similarity optimization integrates the spatial and texture information of multimodal images in the precise registration stage. Finally, the consistency of the registration results in spatial position and texture characteristics is ensured through geometric and grayscale verification, providing accurate and reliable technical support for the multimodal fusion of CT and CTP images. Example Description

[0054] Example 1 (clinical stroke diagnosis): In the hospital radiology department, S1-S5 is used to unify the resolution and grayscale processing of CT and CTP images of stroke patients, and then a good registration is obtained through rigid body transformation and grayscale fusion. At S6, if the geometric error Egeo <1 and the grayscale correlation coefficient Ssimilarity>0.90, the doctor can directly use the fused image to determine the location of the lesion and the blood flow defect area, and the treatment plan is more accurate and efficient.

[0055] Example 2 (scientific research offline algorithm evaluation): The researcher obtains simulated brain CT and CTP from public databases (such as BrainWeb) and adds noise or rotation deviation. After unified processing in S1-S2, key points are extracted and point clouds are generated, and then alignment and fusion are performed through S3-S5. S6 uses geometric error and similarity to quantify the convergence performance of the method of the present invention under different noise levels or initial offsets. Compared with traditional mutual information (MI) or ICP-only solutions, it can often be found that the method still has better robustness under high noise or large offset conditions.

[0056] Example 3 (Cloud Telemedicine): Deploy the above-mentioned entire process on the cloud platform and perform parallel registration on batches of CT / CTP uploaded by multiple hospitals. If the geometric or grayscale fusion effect is detected to be up to standard at S6, the final fused image can be returned to each hospital for sharing, significantly improving the efficiency of multi-hospital collaborative diagnosis and achieving the goal of cross-regional expert joint diagnosis of brain lesions.

[0057] Through the above six steps and multiple application scenarios, it can be seen that the present invention particularly highlights the significance of eliminating resolution and grayscale differences (S1), capturing key points of cross-modal adaptation (S2-S3), first performing posture alignment and then grayscale fusion (S4-S5), and dual measurement (S6); it respectively presents how to achieve consistency basis through interpolation and normalization, how to use multi-scale differences to overcome modal texture differences, how to efficiently perform rigid body transformation in three-dimensional point clouds and finally fuse intensity distribution, etc. Without violating the above ideas and technical effects, equivalent replacement or local improvement of each formula or implementation details still falls within the scope of protection of the present invention.

[0058] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A brain CT and CTP image registration method based on point cloud technology, characterized in that: The following steps are involved: S1, performing spatial resolution standardization processing on the CT image and the CTP image respectively, normalizing the grayscale values ​​of the CT image and the CTP image, and adjusting the grayscale range to a uniform scale; S2, extracting feature points with rotation, scale and grayscale change invariance from CT images and CTP images after spatial resolution standardization and normalization; S3, converting the feature points into three-dimensional point cloud data to simultaneously include the spatial position and grayscale texture information of the feature points; S4, performing preliminary alignment on the three-dimensional point cloud data based on a rigid body transformation model, and estimating rotation and translation parameters of the image; S5. Based on the preliminary alignment, the similarity of the grayscale information of the point cloud is combined to fuse the multimodal image space and grayscale features to further optimize the point cloud alignment result; S6. The geometric error measurement method is used to evaluate the spatial alignment accuracy of point cloud registration, and the distribution similarity measurement method is used to verify the fusion effect of image grayscale information.

2. The brain CT and CTP image registration method based on point cloud technology according to claim 1, characterized in that: In the spatial resolution standardization process of step S1, the CT image I CT (x,y,z) and CTP image I CTP (x, y, z) are subjected to trilinear interpolation processing respectively to make the two consistent in resolution and layer thickness. The grayscale value of the interpolation point (x, y, z) is calculated by the following formula: Select the neighboring points of the interpolation point according to Calculate the difference weight of the interpolation point based on Calculate the grayscale value of the interpolation point; wherein, w ijk Indicates neighboring points I ( x i , y j , z k ) is the difference weight.

3. The brain CT and CTP image registration method based on point cloud technology according to claim 2, characterized in that: Normalizing the grayscale values ​​of the CT image and the CTP image includes: The grayscale value of the interpolated image is extracted and the mask is normalized according to the following formula: ； in, Indicates that according to the window width and window level The mask that masks the CT and CTP values, normalizing only the unmasked parts, Represents the gray value of the midpoint of the CT image, It is the gray value of the voxel in the CT image.

4. The method for brain CT and CTP image registration based on point cloud technology according to claim 1, characterized in that: In step S2, the extraction of the feature points includes: Normalized image Perform Gaussian blur convolution to construct scale space; ;in, is a Gaussian kernel of scale σ, It represents the blurred image of the image after Gaussian kernel filtering at a certain scale σ; By using the scale difference formula Find local extreme points and get the coordinates of feature points .

5. The method for accurately determining brain CT and CTP images based on point cloud technology according to claim 1, characterized in that: Converting feature points into 3D point cloud data includes: The feature point coordinates The corresponding normalized grayscale value and local descriptor information combination is expressed as: ,in, I normI is the normalized gray value of the feature point, D For local texture.

6. The method for accurate registration of brain CT and CTP images based on point cloud technology according to claim 1, characterized in that: When performing preliminary alignment of 3D point cloud data, the rigid body transformation model is used , in, P Represents the spatial coordinate vector of a point in the point cloud, R is the rotation matrix, t is the translation vector.

7. The brain CT and CTP image registration method based on point cloud technology according to claim 6, characterized in that: Optimization via iterative closest point algorithm R and t , the objective function during optimization is ; in, and are the corresponding point pairs selected by the closest point strategy in the CT point cloud and the CTP point cloud. N is the number of corresponding point pairs.

8. The method for brain CT and CTP image registration based on point cloud technology according to claim 5 or 7, characterized in that: In step S5, during the point cloud alignment process, the following comprehensive objective function is adopted , in, E ICP is the geometric distance error, α and β For different weight coefficients, Represents grayscale similarity, I CT with I CTP They come from the information of CT images and CTP images carried by the point cloud respectively.

Images