Medical image and three-dimensional space registration method based on 3D printing technology

Abstract

The invention discloses a medical image and three-dimensional space registration method based on a 3D printing technology, and relates to the technical field of medical image processing, and the method comprises the steps: S1, data acquisition and preprocessing; s2, performing three-dimensional reconstruction and model generation; s3, making a self-adaptive registration strategy; s4, performing registration implementation and result verification; and S5, application and feedback of the 3D printing model. According to the medical image and three-dimensional space registration method based on the 3D printing technology, high-precision medical image and three-dimensional space registration is realized by combining the 3D printing technology and a self-adaptive registration strategy. The core of the self-adaptive registration strategy is to dynamically adjust the registration algorithm according to the anatomical structure and pathological change of the patient, so that the precision and robustness of processing the patient with complex anatomical structure and pathological change and registration between image data of different modalities are remarkably improved. All the steps are mutually associated and supported to jointly form a complete, efficient and reliable process system.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a medical image and three-dimensional space registration method based on 3D printing technology. Background Art

[0002] Medical imaging technologies, such as X-ray, computed tomography (CT), magnetic resonance imaging (MRI) and positron emission tomography (PET), play a vital role in modern medical diagnosis. Three-dimensional spatial alignment is the key technology to align and fuse these imaging data from different times, different devices or different modalities, so that doctors can have a more comprehensive understanding of the patient's anatomical structure and pathological status.

[0003] Existing medical imaging technologies usually use computer algorithms to capture and process detailed structures inside the human body. These images are usually presented in the form of two-dimensional images, but in recent years, with the advancement of technology, three-dimensional imaging has become more and more common. These technologies not only improve the accuracy of diagnosis, but also provide strong support for surgical planning and treatment effect evaluation. The three-dimensional space registration method involves accurately aligning image data from different sources in three-dimensional space. This is usually achieved through algorithms such as feature point matching, mutual information maximization, or rigid / non-rigid transformation. The registered image data can generate a comprehensive view to help doctors identify changes in anatomical structures, tumor growth, or the effects of treatment. Among them, the feature-based method is currently the most common registration method, which extracts feature points in the image, such as corners, edges, anatomical landmarks, etc., and then matches these feature points in the real world to achieve image registration.

[0004] However, in the prior art, the feature points in the image cannot be accurately matched to the feature points matched in the real world. For example, the feature points in the local area of ​​the image are not obvious enough, and it is impossible to manually select or automatically identify enough feature points for registration; for example, the feature points in the image that are manually selected or automatically segmented cannot accurately select the corresponding feature points when selecting in the real world. Disadvantages: Insufficient number of feature points leads to inability to register, and the feature points cannot accurately match one by one, resulting in large registration errors.

[0005] In theory, image registration requires at least three pairs of feature points to be realized, because three points can determine a plane, and a plane and the points on the plane can calculate the transformation of the image coordinate system and the spatial coordinate system. However, if the transformation needs to be calculated more accurately, more pairs of feature points are needed. Generally, 4-5 pairs are recommended, and the one-to-one corresponding point pairs are selected accurately, so that the robustness and accuracy of the registration calculation can be guaranteed. However, if the registration points are not enough or the relative error of the selected point pairs is large, registration will fail or the registration error is too large. In order to make the feature points clearer or more accurate, many practices use developing markers, such as zirconium oxide, titanium nails, etc., which can solve the problem of insufficient or unclear feature points, but the problem that follows is that the markers need to be pre-buried or fixed before shooting, and after the shooting is completed and during the registration period, it must be ensured that the markers do not move. This brings two problems. First, the pre-embedding or fixation of markers may cause trauma. For example, titanium nails need to be fixed in the bone, which requires skin incision. For example, skin adhesives are needed to fix zirconium oxide balls, which may cause skin damage when the adhesive is removed. Second, to ensure that the markers do not move, it will affect the normal life of the photographer.

[0006] In summary, the prior art has the following three problems:

[0007] 1. Not enough feature points;

[0008] 2. The position accuracy of feature points in the image and the real world is not enough;

[0009] 3. The development markers are invasive and affect normal life.

[0010] Therefore, it is urgent to improve this shortcoming. The present invention studies and improves the existing technology and its shortcomings, and provides a medical imaging and three-dimensional space registration method based on 3D printing technology. Summary of the invention

[0011] The purpose of the present invention is to provide a medical image and three-dimensional space registration method based on 3D printing technology to solve the problems raised in the above background technology.

[0012] To achieve the above object, the present invention provides the following technical solution: a medical image and three-dimensional space registration method based on 3D printing technology, comprising the following steps:

[0013] S1. Data acquisition and preprocessing:

[0014] Receive patient image data transmitted by various medical imaging devices, and perform denoising, enhancement and segmentation processing on the image data to improve the quality of the image data and the accuracy of subsequent analysis, reduce the interference of noise and artifacts, and highlight the region of interest (ROI);

[0015] S2. 3D reconstruction and model generation:

[0016] Use reconstruction algorithms (such as iterative reconstruction, filtered back projection, etc.) to convert two-dimensional image data into a three-dimensional model, and use 3D printing technology to generate a three-dimensional digital model and guide plate of the patient based on the three-dimensional reconstruction results;

[0017] S3. Adaptive registration strategy formulation:

[0018] Extract key information such as anatomical feature points, surfaces or volumes from the 3D digital model and guide, select the registration algorithm based on the feature extraction results and the modality difference of the image data, and adjust the algorithm parameters according to real-time feedback during the registration process to optimize the registration results;

[0019] S4. Registration implementation and result verification:

[0020] Align image data of different modalities or at different times according to the adaptive registration strategy, and use quantitative indicators to evaluate the accuracy and robustness of the registration results, ensure the accuracy of the guide plate in the actual position of the patient, and ensure that it is consistent with the designed 3D model. If necessary, the practicality of the registration can also be verified through the doctor's subjective evaluation and clinical needs;

[0021] S5. Application and feedback of 3D printing models:

[0022] The registered 3D model and guide are used in practical scenarios such as surgical planning, teaching and training, or patient communication. For example, the 3D printed guide can be adapted to the surface fixation of soft tissue, hard tissue, or a mixture of soft and hard tissue. If the fixation is not stable enough, it can be fixed with bone screws or other adhesives. The evaluation and feedback of doctors, patients, or teaching staff on the 3D printed model are collected to optimize subsequent steps and processes.

[0023] Further, in step S1, the medical imaging equipment includes but is not limited to CT (computed tomography), MRI (magnetic resonance imaging) and PET (positron emission tomography), and the imaging data formation process is as follows: the patient is placed in the corresponding imaging equipment according to clinical needs, and the equipment performs scanning according to preset scanning parameters (such as layer thickness, scanning time, radiation dose, etc.). After the scanning is completed, the equipment generates original imaging data and stores it in DICOM (digital imaging and communications) format;

[0024] The CT scans the human body at multiple angles using X-rays, and then reconstructs a three-dimensional image using a computer algorithm. CT images are particularly sensitive to bones and high-density tissues.

[0025] The MRI uses strong magnetic fields and radio frequency waves to image the human body, with high resolution for soft tissues (such as the brain, muscles, internal organs, etc.) and without the need for ionizing radiation;

[0026] The PET generates images by injecting radioactive tracers into the patient's body and then measuring the distribution of the tracers in the body. It is mainly used to evaluate tissue function, metabolic activity and blood flow.

[0027] Furthermore, in step S1, the image data is processed as follows:

[0028] Denoising: Use filtering techniques (such as Gaussian filtering, median filtering, etc.) to reduce random noise in the image, and use specific denoising algorithms to remove metal artifacts and motion artifacts;

[0029] Metal artifacts (artifacts caused by metal objects such as implants and dental fillings in CT scans) are removed by using iterative reconstruction algorithms, sinogram correction or linear interpolation. These methods restore the original information of the image by adjusting or replacing the pixel values ​​affected by the metal artifacts.

[0030] Motion artifacts (artifacts caused by physiological activities such as patient movement or breathing during scanning) are removed by using motion correction algorithms, respiratory gating technology or image registration methods, which restore image clarity and accuracy by tracking and compensating for patient movement;

[0031] Enhancement: Adjust image contrast to make anatomical structures and pathological changes clearer, and use histogram equalization, adaptive contrast enhancement, and linear / nonlinear stretching techniques to improve the overall visual effect of the image;

[0032] The histogram equalization adjusts the histogram distribution of the image to make the image contrast more uniform;

[0033] The adaptive contrast enhancement dynamically adjusts the contrast according to the local features of the image;

[0034] The linear / non-linear stretching adjusts the grayscale value range of the image through linear or non-linear transformation;

[0035] Segmentation: Segmentation is performed using machine learning methods. The segmentation results can be used for subsequent quantitative analysis, 3D reconstruction, and registration steps to identify and extract regions of interest (ROI, i.e., anatomical structures or pathological change areas in medical images that are related to specific diagnosis or treatment. Identifying and extracting ROI can reduce the amount of data for subsequent processing and improve analysis efficiency), such as tumors, blood vessels, organs, etc.

[0036] Other preprocessing: Correct geometric distortion or registration errors in the image, and resize and optimize the image resolution to meet the needs of subsequent analysis.

[0037] Furthermore, in step S2, different reconstruction algorithms are used for different imaging principles and characteristics of medical imaging modalities (CT, MRI and PET) to ensure the accuracy and reliability of the three-dimensional model, as follows:

[0038] For CT image data, the reconstruction algorithm uses filtered back projection and iterative reconstruction. The filtered back projection filters the projection data and then back projects it into the spatial domain to reconstruct a three-dimensional image. The iterative reconstruction optimizes the reconstruction results through continuous iterations, and is particularly suitable for processing low-quality data (such as low-dose scanning) to reduce the generation of artifacts. Iterative reconstruction can be customized and optimized according to specific clinical needs, such as improving the contrast of specific tissues or reducing noise levels.

[0039] For MRI image data, the reconstruction algorithm uses Fourier transform reconstruction and iterative reconstruction. The Fourier transform reconstruction obtains image data by performing inverse Fourier transform on the acquired signal. The iterative reconstruction is used to process low-quality data or perform customized optimization, similar to CT. The application of iterative reconstruction in MRI must take into account factors such as the phase information of the signal and artifacts in the acquisition process.

[0040] For PET image data, the reconstruction algorithm adopts ordered subset expectation maximization (OSEM), which is an accelerated version of maximum likelihood expectation maximization (MLEM, which estimates the image by maximizing the likelihood function of the observed data, can handle Poisson noise in PET data, and generate high-quality reconstructed images, but has high computational complexity and requires a long reconstruction time). By dividing the data set into multiple subsets and iteratively updating them in sequence, the reconstruction speed is accelerated. The OSEM algorithm significantly improves the reconstruction efficiency while maintaining the advantages of the MLEM algorithm.

[0041] Furthermore, in step S2, the key operation process of the guide plate is as follows:

[0042] 1) Generate matching guides

[0043] Selecting an area: In the model after 3D reconstruction, the operator can manually select, or the algorithm can automatically select, a specific area that contains the target site for surgery or treatment;

[0044] Generate guides: Generate surgical guides that accurately match the target tissue through morphological processing and Boolean operations or other applicable algorithms;

[0045] Select registration points: When generating the guide surface, the operator can manually select, or use the algorithm to automatically select 4 to 5 key registration points, and record the 3D coordinate values ​​of these registration points in the standard DICOM image coordinate system. The registration points exist on the guide surface in a form that facilitates the subsequent precise acquisition of physical coordinates, such as ball pits, sharp pits, etc.;

[0046] Guide design verification: Verify the design of the guide to ensure that it meets the needs of the surgery or treatment, can accurately cover the selected area, and contains all the necessary registration points;

[0047] 2) Guide plate data export

[0048] Export file: Export the designed guide data containing registration point information to a file format suitable for 3D printing, such as STL or OBJ;

[0049] File check: Check the exported file to ensure that it is complete and the geometry and registration point information of the guide are correct;

[0050] 3) Guide plate fixing and coordinate acquisition

[0051] Guide plate fixation: fix the guide plate to the actual soft tissue, hard tissue or the combination of soft and hard tissue of the patient to ensure that the position of the guide plate is consistent with the selected area;

[0052] Collecting coordinates: Using a three-dimensional measurement device (such as an optical tracking system or a laser scanner, etc.) to collect the three-dimensional coordinates of the recorded registration point (ball pit) in the patient's actual coordinate system.

[0053] Furthermore, in step S3, the three-dimensional coordinates of the collected registration points in the patient's actual coordinate system are registered with the three-dimensional coordinates of the recorded registration points in the image coordinate system by using a registration algorithm, and a transformation matrix between the image coordinate system and the space coordinate system is calculated by using the registration algorithm;

[0054] The registration algorithms include rigid registration, affine registration, non-rigid registration, and multimodal registration, as follows:

[0055] The rigid body registration is applicable to the case where the transformation between two images only involves rotation and translation, such as the alignment of bone structures. The implementation method is: using the Iterative Closest Point (ICP) algorithm to optimize the transformation parameters through continuous iterations to minimize the distance between the two sets of feature points.

[0056] The affine registration is applicable to transformations between images, including not only rotation and translation, but also linear transformations such as scaling and shearing, and is implemented by optimizing the parameters of the affine transformation matrix to achieve the best match between the two sets of images in terms of overall morphology;

[0057] The non-rigid registration is applicable to images with complex local deformations, such as deformations of soft tissues or organs, and is implemented by estimating and simulating deformation fields using methods based on physical models, statistical models or machine learning;

[0058] The multimodal registration is suitable for registering images of different modalities (such as CT and MRI), and is implemented by using similarity metrics such as mutual information and cross entropy as optimization targets, and performing registration in combination with appropriate transformation models and optimization algorithms.

[0059] Furthermore, during the rigid body registration process, the matching errors between feature points are monitored in real time, and parameters such as the number of iterations and convergence conditions of the ICP algorithm are adjusted accordingly to achieve a better registration effect.

[0060] Furthermore, the affine registration adjusts the parameter search range, step size, etc. of the affine transformation according to the error metric (such as mean square error MSE) in the registration process to accelerate convergence and improve the registration accuracy.

[0061] Furthermore, during the registration process, the non-rigid registration adjusts the parameters of the deformation model, such as the elastic modulus, the regularization term weight, etc., according to the smoothness and continuity of the deformation field and the matching degree of the feature points, so as to ensure the rationality of the deformation field and the accuracy of the registration.

[0062] Furthermore, during the registration process, the multimodal registration dynamically adjusts the parameters of the optimization algorithm (such as learning rate, number of iterations, etc.), as well as the complexity and regularization strength of the transformation model according to changes in the similarity metric to balance the registration accuracy and computational efficiency.

[0063] Furthermore, in step S4, the evaluation indicators and the evaluation process of the quantitative indicator evaluation are as follows:

[0064] Mean square error (MSE, which measures the difference in pixel values ​​between image data after registration): Calculate the average of the sum of squares of the differences in corresponding pixel values ​​between image data before and after registration to obtain the MSE value. The smaller the MSE value, the higher the registration accuracy. The MSE value is compared with the set threshold to determine whether the registration meets the accuracy requirements.

[0065] Mutual Information (MI, which measures the statistical correlation between two image data sets): Calculate the relative entropy between the joint probability distribution of the two image data sets and the product of their respective probability distributions to obtain the MI value. The larger the MI value, the better the registration effect. Similarly, compare it with the set threshold to determine whether the registration is successful.

[0066] Feature point matching error: For the extracted anatomical feature points, the Euclidean distance or other distance measurement between the feature points before and after registration is calculated. The smaller the feature point matching error, the higher the registration accuracy. The error threshold is set according to clinical needs.

[0067] Smoothness and continuity of deformation field: For non-rigid registration, the geometric properties of the deformation field, such as gradient and curvature, are calculated to evaluate the smoothness and continuity of the deformation field. The deformation field should be as smooth and continuous as possible to avoid unreasonable deformation. This can be evaluated by visual inspection or by calculating the geometric properties of the deformation field.

[0068] The present invention provides a medical image and three-dimensional space registration method based on 3D printing technology, which has the following beneficial effects:

[0069] 1. The present invention realizes high-precision medical imaging and three-dimensional space registration by combining 3D printing technology and adaptive registration strategy. The core of the adaptive registration strategy is to dynamically adjust the registration algorithm according to the patient's anatomical structure and pathological changes, thereby significantly improving the accuracy and robustness of processing patients with complex anatomical structures and pathological changes, as well as the registration between different modal image data. The various steps are interrelated and mutually supportive, and together constitute a complete, efficient and reliable process system. Not only is the registration process short in time, but the entire process can be done without human participation, avoiding operational complexity and the risk of human error, and significantly improving the efficiency and accuracy of the clinical workflow.

[0070] 2. The present invention does not require any advance processing during image capture, as long as the image data is standard, it can be registered. And the feature points on the 3D printed guide can be customized according to the tip of the device that collects the registration points in the real world, which can ensure that the points are collected more accurately. In addition, the 3D printed guide can be adapted to the surface fixation of soft tissue, hard tissue or mixed soft and hard tissue. If the fixation is not stable enough, it can also be fixed by bone screws or other adhesives. Because the registration area is the surgical area, neither the bone screws nor the adhesive will cause secondary damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A schematic diagram of the steps of a medical image and three-dimensional space registration method based on 3D printing technology of the present invention;

[0072] Figure 2This is a schematic diagram of a reconstruction algorithm of a medical image and a three-dimensional space registration method based on 3D printing technology in the present invention. DETAILED DESCRIPTION

[0073] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0074] like Figure 1-Figure 2 As shown, a medical image and three-dimensional space registration method based on 3D printing technology includes the following steps:

[0075] S1. Data acquisition and preprocessing:

[0076] Receive patient image data transmitted by various medical imaging devices, and perform denoising, enhancement and segmentation processing on the image data to improve the quality of image data and the accuracy of subsequent analysis, reduce the interference of noise and artifacts, and highlight the region of interest (ROI). Among them, medical imaging devices include but are not limited to CT (computed tomography), MRI (magnetic resonance imaging) and PET (positron emission tomography), and the image data formation process: the patient is placed in the corresponding imaging device according to clinical needs, and the device scans according to preset scanning parameters (such as layer thickness, scanning time, radiation dose, etc.). After the scan is completed, the device generates original image data and stores it in DICOM (digital imaging and communications) format. The principles of CT, MR and PET are as follows:

[0077] CT scans the human body at multiple angles using X-rays, and then reconstructs a three-dimensional image using computer algorithms. CT images are particularly sensitive to bones and high-density tissues.

[0078] MRI uses strong magnetic fields and radio frequency waves to image the human body, with high resolution for soft tissues (such as the brain, muscles, and internal organs) and without the need for ionizing radiation;

[0079] PET generates images by injecting radioactive tracers into the patient's body and then measuring the distribution of the tracers in the body. It is mainly used to evaluate tissue function, metabolic activity, and blood flow.

[0080] In this embodiment, the steps of processing the image data are as follows:

[0081] Denoising: Use filtering techniques (such as Gaussian filtering, median filtering, etc.) to reduce random noise in the image, and use specific denoising algorithms to remove metal artifacts and motion artifacts;

[0082] Metal artifacts (artifacts caused by metal objects such as implants and dental fillings in CT scans) are removed by using iterative reconstruction algorithms, sinogram correction or linear interpolation. These methods restore the original information of the image by adjusting or replacing the pixel values ​​affected by the metal artifacts.

[0083] Motion artifacts (artifacts caused by physiological activities such as patient movement or breathing during scanning) are removed by using motion correction algorithms, respiratory gating technology or image registration methods, which restore image clarity and accuracy by tracking and compensating for patient movement;

[0084] Enhancement: Adjust image contrast to make anatomical structures and pathological changes clearer, and use histogram equalization, adaptive contrast enhancement, and linear / nonlinear stretching techniques to improve the overall visual effect of the image;

[0085] Histogram equalization makes the image contrast more uniform by adjusting the histogram distribution of the image;

[0086] Adaptive contrast enhancement dynamically adjusts the contrast based on the local features of the image;

[0087] Linear / nonlinear stretching adjusts the grayscale value range of an image through linear or nonlinear transformation;

[0088] Segmentation: Segmentation is performed using machine learning methods. The segmentation results can be used for subsequent quantitative analysis, 3D reconstruction, and registration steps to identify and extract regions of interest (ROI, i.e., anatomical structures or pathological change areas in medical images that are related to specific diagnosis or treatment. Identifying and extracting ROI can reduce the amount of data for subsequent processing and improve analysis efficiency), such as tumors, blood vessels, organs, etc.

[0089] Other preprocessing: Correct geometric distortion or registration errors in the image, and resize and optimize the image resolution to meet the needs of subsequent analysis;

[0090] Image data preprocessing is the basis for ensuring the smooth progress of subsequent registration and analysis steps. Through denoising and enhancement, the signal-to-noise ratio and contrast of image data can be improved, making the anatomical structure and pathological changes clearer. The segmentation step helps to distinguish the region of interest from the background, providing more accurate information for subsequent registration.

[0091] S2. 3D reconstruction and model generation:

[0092] Reconstruction algorithms (such as iterative reconstruction, filtered back projection, etc.) are used to convert two-dimensional image data into a three-dimensional model, and based on the three-dimensional reconstruction results, 3D printing technology is used to generate a three-dimensional digital model (such as an STL file) and a guide of the patient.

[0093] In this embodiment, different reconstruction algorithms are used for different imaging principles and characteristics of medical imaging modalities (CT, MRI and PET) to ensure the accuracy and reliability of the three-dimensional model, as follows:

[0094] For CT image data, the reconstruction algorithm uses filtered back projection and iterative reconstruction. Filtered back projection filters the projection data and then back projects it into the spatial domain to reconstruct a three-dimensional image. Iterative reconstruction optimizes the reconstruction results through continuous iterations. It is particularly suitable for processing low-quality data (such as low-dose scanning) and reducing the generation of artifacts. Iterative reconstruction can be customized and optimized according to specific clinical needs, such as improving the contrast of specific tissues or reducing noise levels.

[0095] For MRI image data, the reconstruction algorithm uses Fourier transform reconstruction and iterative reconstruction. Fourier transform reconstruction obtains image data by performing inverse Fourier transform on the acquired signal. Iterative reconstruction is used to process low-quality data or perform customized optimization, similar to CT. The application of iterative reconstruction in MRI must take into account factors such as the phase information of the signal and artifacts during the acquisition process.

[0096] For PET image data, the reconstruction algorithm uses ordered subset expectation maximization (OSEM), which is an accelerated version of maximum likelihood expectation maximization (MLEM, which estimates the image by maximizing the likelihood function of the observed data. It can handle Poisson noise in PET data and generate high-quality reconstructed images, but has high computational complexity and requires a long reconstruction time). By dividing the data set into multiple subsets and iteratively updating them in sequence, the reconstruction speed is accelerated. The OSEM algorithm significantly improves the reconstruction efficiency while maintaining the advantages of the MLEM algorithm.

[0097] Three-dimensional reconstruction is a key step in converting two-dimensional image data into three-dimensional spatial representation. By selecting the appropriate reconstruction algorithm, an accurate three-dimensional model reflecting the patient's anatomical structure and pathological changes can be generated. 3D printing technology can present these models in physical form, making it easier for doctors to intuitively understand and operate.

[0098] In this embodiment, the key operation process of the guide plate is as follows:

[0099] 1) Generate matching guides

[0100] Selecting an area: In the model after 3D reconstruction, the operator can manually select, or the algorithm can automatically select, a specific area that contains the target area for surgery or treatment;

[0101] Generate guides: Generate surgical guides that accurately match the target tissue through morphological processing and Boolean operations or other applicable algorithms;

[0102] Select registration points: When generating the guide surface, the operator can manually select, or use the algorithm to automatically select 4 to 5 key registration points, and record the 3D coordinate values ​​of these registration points in the standard DICOM image coordinate system. These registration points exist on the guide surface in a form that facilitates the subsequent precise acquisition of physical coordinates, such as ball pits, sharp pits, etc.

[0103] Guide design verification: Verify the design of the guide to ensure that it meets the needs of the surgery or treatment, can accurately cover the selected area, and contains all the necessary registration points;

[0104] 2) Guide plate data export

[0105] Export file: Export the designed guide data containing registration point information to a file format suitable for 3D printing, such as STL or OBJ;

[0106] File check: Check the exported file to ensure that it is complete and the geometry and registration point information of the guide are correct;

[0107] 3) Guide plate fixing and coordinate acquisition

[0108] Guide plate fixation: fix the guide plate to the actual soft tissue, hard tissue or the combination of soft and hard tissue of the patient to ensure that the position of the guide plate is consistent with the selected area;

[0109] Collecting coordinates: Using a three-dimensional measurement device (such as an optical tracking system or a laser scanner, etc.) to collect the three-dimensional coordinates of the recorded registration point (ball pit) in the patient's actual coordinate system.

[0110] S3. Adaptive registration strategy formulation:

[0111] Key information such as anatomical feature points, surfaces or volumes are extracted from the three-dimensional digital model and guide. The registration algorithm is selected based on the feature extraction results and the modal differences of the image data. During the registration process, the algorithm parameters are adjusted according to real-time feedback to optimize the registration results.

[0112] In this embodiment, the three-dimensional coordinates of the collected registration points in the actual patient coordinate system are registered with the three-dimensional coordinates of the recorded registration points in the image coordinate system through the registration algorithm, and the transformation matrix between the image coordinate system and the space coordinate system is calculated using the registration algorithm. Among them, the registration algorithm includes rigid registration, affine registration, non-rigid registration, and multi-modal registration, which are as follows:

[0113] Rigid body registration is applicable to the case where the transformation between two images only involves rotation and translation, such as the alignment of bone structures. It is implemented by using the Iterative Closest Point (ICP) algorithm to optimize the transformation parameters through continuous iterations to minimize the distance between the two sets of feature points. During the registration process, the matching error between feature points is monitored in real time, and the number of iterations, convergence conditions and other parameters of the ICP algorithm are adjusted accordingly to achieve a better registration effect.

[0114] Affine registration is applicable to transformations between images, including not only rotation and translation, but also linear transformations such as scaling and shearing. It is achieved by optimizing the parameters of the affine transformation matrix to achieve the best match between the two sets of images in terms of overall morphology. Affine registration adjusts the parameter search range and step size of the affine transformation based on the error metric (such as mean square error MSE) during the registration process to accelerate convergence and improve registration accuracy.

[0115] Non-rigid registration is applicable to complex local deformations between images, such as deformation of soft tissues or organs. It is implemented by estimating and simulating deformation fields using methods based on physical models, statistical models, or machine learning. During the registration process, non-rigid registration adjusts the parameters of the deformation model, such as elastic modulus, regularization term weight, etc., based on the smoothness, continuity, and matching degree of feature points of the deformation field to ensure the rationality of the deformation field and the accuracy of registration.

[0116] Multimodal registration is suitable for registering images of different modalities (such as CT and MRI). It is implemented by using similarity metrics such as mutual information and cross entropy as optimization targets, combined with appropriate transformation models and optimization algorithms for registration. During the registration process, multimodal registration dynamically adjusts the parameters of the optimization algorithm (such as learning rate, number of iterations, etc.), as well as the complexity and regularization strength of the transformation model according to the changes in the similarity metrics, in order to balance the registration accuracy and computational efficiency.

[0117] The core of the adaptive registration strategy is to dynamically adjust the registration algorithm according to the patient's anatomical structure and pathological changes. The feature extraction step provides key information for registration, while the algorithm selection ensures the accuracy and efficiency of registration. The adaptive adjustment step can further improve the robustness of registration, especially when dealing with complex anatomical structures and pathological changes.

[0118] S4. Registration implementation and result verification:

[0119] The imaging data of different modalities or at different times are aligned according to the adaptive registration strategy, and the accuracy and robustness of the registration results are evaluated using quantitative indicators to ensure the accuracy of the guide in the actual position of the patient to ensure that it is consistent with the three-dimensional model during design. If necessary, the practicality of the registration can also be verified through the doctor's subjective evaluation and clinical needs.

[0120] In this embodiment, the evaluation indicators and the evaluation process of the quantitative indicator evaluation are as follows:

[0121] Mean square error (MSE, which measures the difference in pixel values ​​between image data after registration): Calculate the average of the sum of squares of the differences in corresponding pixel values ​​between image data before and after registration to obtain the MSE value. The smaller the MSE value, the higher the registration accuracy. The MSE value is compared with the set threshold to determine whether the registration meets the accuracy requirements.

[0122] Mutual Information (MI, which measures the statistical correlation between two image data sets): Calculate the relative entropy between the joint probability distribution of the two image data sets and the product of their respective probability distributions to obtain the MI value. The larger the MI value, the better the registration effect. Similarly, compare it with the set threshold to determine whether the registration is successful.

[0123] Feature point matching error: For the extracted anatomical feature points, the Euclidean distance or other distance measurement between the feature points before and after registration is calculated. The smaller the feature point matching error, the higher the registration accuracy. The error threshold is set according to clinical needs.

[0124] Smoothness and continuity of deformation field: For non-rigid registration, the geometric properties of the deformation field, such as gradient and curvature, are calculated to evaluate the smoothness and continuity of the deformation field. The deformation field should be as smooth and continuous as possible to avoid unreasonable deformation. This can be evaluated by visual inspection or by calculating the geometric properties of the deformation field.

[0125] Registration implementation is the step of applying the adaptive registration strategy to the actual image data. By selecting appropriate quantitative indicators and subjective evaluation methods, the pros and cons of the registration results can be comprehensively evaluated. This helps to ensure the accuracy and reliability of the registration method in clinical applications.

[0126] S5. Application and feedback of 3D printing models:

[0127] The registered 3D model and guide are used in actual scenarios such as surgical planning, teaching training or patient communication. The evaluation and feedback of doctors, patients or teaching staff on the 3D printed model are collected to optimize subsequent steps and processes.

[0128] The application of 3D printed models is one of the important outputs of this method. By presenting the registered three-dimensional model in physical form, the patient's anatomical structure and pathological changes can be intuitively displayed, providing strong support for surgical planning, teaching training and patient communication. At the same time, the 3D printed guide can be adapted to the surface fixation of soft tissue, hard tissue or mixed soft and hard tissue. If the fixation is not stable enough, it can also be fixed by bone screws or other adhesives. Because the registration area is the surgical area, neither bone screws nor adhesives will cause secondary damage. In addition, collecting feedback helps to continuously improve and optimize this method and improve its practicality and satisfaction in clinical applications.

[0129] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

Claims

1. A medical image and three-dimensional space registration method based on 3D printing technology, characterized in that: The following steps are involved: S1. Data acquisition and preprocessing: receiving patient image data transmitted by various medical imaging devices, and performing denoising, enhancement and segmentation processing on the image data; S2. 3D reconstruction and model generation: Use reconstruction algorithms to convert 2D image data into 3D models, and use 3D printing technology to generate the patient's 3D digital model and guide plate based on the 3D reconstruction results; S3, Adaptive registration strategy formulation: Extract key information from the 3D digital model and guide plate, select the registration algorithm based on the feature extraction results and the modality difference of the image data, and adjust the algorithm parameters according to the real-time feedback during the registration process to optimize the registration results; S4. Registration implementation and result verification: Align the image data of different modalities or at different times according to the adaptive registration strategy, and use quantitative indicators to evaluate the accuracy and robustness of the registration results to ensure the accuracy of the guide in the actual position of the patient to ensure that it is consistent with the designed 3D model; S5. Application and feedback of 3D printed models: Use the registered 3D models and guides in actual scenarios, and collect evaluations and feedback to optimize subsequent steps and processes.

2. A medical image and three-dimensional space registration method based on 3D printing technology according to claim 1, characterized in that: In step S1, the medical imaging equipment includes but is not limited to CT, MRI and PET, and the imaging data formation process is as follows: the patient is placed in the corresponding imaging equipment according to clinical needs, the equipment performs scanning according to preset scanning parameters, and after the scanning is completed, the equipment generates original imaging data and stores it in DICOM format; The processing steps of the image data are as follows: Denoising: Use filtering technology to reduce random noise in the image, and use specific denoising algorithms to remove metal artifacts and motion artifacts; For metal artifacts, iterative reconstruction algorithms, sinusoidal correction or linear interpolation methods are used to remove them; Motion artifacts can be removed using motion correction algorithms, respiratory gating technology, or image registration methods; Enhancement: Adjust image contrast to make anatomical structures and pathological changes clearer, and use histogram equalization, adaptive contrast enhancement, and linear / nonlinear stretching techniques to improve the overall visual effect of the image; The histogram equalization adjusts the histogram distribution of the image to make the image contrast more uniform; The adaptive contrast enhancement dynamically adjusts the contrast according to the local features of the image; The linear / non-linear stretching adjusts the grayscale value range of the image through linear or non-linear transformation; Segmentation: Use machine learning methods to segment, identify and extract regions of interest; Other preprocessing: Correct geometric distortion or registration errors in the image, and resize and optimize the resolution of the image to meet the needs of subsequent analysis.

3. A medical image and three-dimensional space registration method based on 3D printing technology according to claim 2, characterized in that: In step S2, different reconstruction algorithms are used for different imaging principles and characteristics of medical imaging modalities, as follows: For CT image data, the reconstruction algorithm uses filtered back projection and iterative reconstruction. The filtered back projection filters the projection data and then back projects it into the spatial domain to reconstruct a three-dimensional image. The iterative reconstruction optimizes the reconstruction results through continuous iterations, which is suitable for processing low-quality data and reducing the generation of artifacts. The iterative reconstruction can be customized and optimized according to specific clinical needs. For MRI image data, the reconstruction algorithm uses Fourier transform reconstruction and iterative reconstruction. The Fourier transform reconstruction obtains image data by performing inverse Fourier transform on the acquired signal. The iterative reconstruction is used to process low-quality data or perform customized optimization. The application of iterative reconstruction in MRI must take into account the phase information of the signal and artifacts in the acquisition process. For PET image data, the reconstruction algorithm adopts ordered subset expectation maximization, which speeds up the reconstruction by dividing the data set into multiple subsets and iteratively updating them in sequence.

4. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 1, characterized in that: In step S2, the key operation process of the guide plate is as follows: 1) Generate matching guides Select area: Select the target area that needs surgery or treatment in the model after 3D reconstruction; Generate guides: Generate surgical guides that accurately match the target tissue through morphological processing and Boolean operations or other applicable algorithms; Select registration points: When generating the guide surface, select 4 to 5 key registration points and record the 3D coordinate values ​​of these registration points in the standard DICOM image coordinate system; Guide design verification: Verify the design of the guide to ensure that it meets the needs of the surgery or treatment, can accurately cover the selected area, and contains all the necessary registration points; 2) Guide plate data export Export file: export the designed guide data containing registration point information into a file format suitable for 3D printing; File check: Check the exported file to ensure that it is complete and the geometry and registration point information of the guide are correct; 3) Guide plate fixing and coordinate acquisition Guide plate fixation: fix the guide plate to the actual soft tissue, hard tissue or the combination of soft and hard tissue of the patient to ensure that the position of the guide plate is consistent with the selected area; Collect coordinates: Use a three-dimensional measurement device to collect the three-dimensional coordinates of the recorded registration points in the patient's actual coordinate system.

5. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 4, characterized in that: In step S3, the three-dimensional coordinates of the collected registration points in the patient's actual coordinate system are registered with the three-dimensional coordinates of the recorded registration points in the image coordinate system by using a registration algorithm, and the transformation matrix between the image coordinate system and the space coordinate system is calculated by using the registration algorithm; The registration algorithms include rigid registration, affine registration, non-rigid registration, and multimodal registration, as follows: The rigid body registration is applicable to the case where the transformation between two images only involves rotation and translation, and is implemented by using an iterative closest point algorithm to optimize the transformation parameters through continuous iteration so that the distance between the two sets of feature points is minimized; The affine registration is applicable to the transformation between images, which includes not only rotation and translation, but also linear transformations such as scaling and shearing, and is implemented by optimizing the parameters of the affine transformation matrix so that the two sets of images can achieve the best match in overall morphology; The non-rigid registration is applicable to images with complex local deformations, and is implemented by estimating and simulating deformation fields using methods based on physical models, statistical models or machine learning; The multimodal registration is suitable for registering images of different modalities, and is implemented by using a similarity metric as an optimization target and combining an appropriate transformation model and an optimization algorithm for registration.

6. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 5, characterized in that: During the rigid body registration process, the matching errors between feature points are monitored in real time, and the number of iterations and convergence condition parameters of the ICP algorithm are adjusted accordingly to achieve a better registration effect.

7. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 5, characterized in that: The affine registration adjusts the parameter search range and step size of the affine transformation according to the error metric in the registration process to accelerate convergence and improve the registration accuracy.

8. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 5, characterized in that: During the registration process, the non-rigid registration adjusts the parameters of the deformation model according to the smoothness and continuity of the deformation field and the matching degree of the feature points to ensure the rationality of the deformation field and the accuracy of the registration.

9. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 5, characterized in that: During the registration process, the multimodal registration dynamically adjusts the parameters of the optimization algorithm, as well as the complexity and regularization strength of the transformation model according to the change of the similarity metric, so as to balance the registration accuracy and the computational efficiency.

10. The medical image and three-dimensional space registration method based on 3D printing technology according to claim 1, characterized in that: In step S4, the evaluation indicators and the evaluation process of the quantitative indicator evaluation are as follows: Mean square error: Calculate the average of the sum of squares of the differences in the corresponding pixel values ​​of the image data before and after registration to obtain the MSE value. The smaller the MSE value, the higher the registration accuracy. It is compared with the set threshold to determine whether the registration meets the accuracy requirements. Mutual Information: Calculate the relative entropy between the joint probability distribution of the two image data sets and the product of their respective probability distributions to obtain the MI value. The larger the MI value, the better the registration effect. It is compared with the set threshold to determine whether the registration is successful. Feature point matching error: For the extracted anatomical feature points, the Euclidean distance or other distance measurement between the feature points before and after registration is calculated. The smaller the feature point matching error, the higher the registration accuracy. The error threshold is set according to clinical needs. Smoothness and continuity of the deformation field: For non-rigid registration, the geometric properties of the deformation field, such as gradient and curvature, are calculated to assess the smoothness and continuity of the deformation field. This can be done either by visual inspection or by calculating the geometric properties of the deformation field.

Images