Image registration method and device, motion correction method and system for medical images
By selecting the image with the lowest degree of motion artifacts as the registration reference in medical images, and combining feature operators and cluster analysis, the problem of unreliable image registration results is solved, and high-quality and efficient image registration and motion correction are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI UNITED IMAGING HEALTHCARE
- Filing Date
- 2022-12-13
- Publication Date
- 2026-06-02
AI Technical Summary
In existing technologies, the selection of reference images for image registration is easily affected by motion artifacts and anatomical ghosting, leading to unreliable subsequent image registration results.
By acquiring medical image data at multiple time points, the degree of motion artifacts and equilibrium position are determined. The image with the lowest degree of motion artifacts is selected as the registration reference image. Feature operators and cluster analysis are used, combined with mean square error and threshold to filter candidate images, and a matching registration algorithm is used to perform image registration.
It improves the quality and efficiency of image registration, reduces the difficulty and cost of registration, eliminates anatomical structure position deviations caused by patient movement, and enhances the accuracy and efficiency of motion correction.
Smart Images
Figure CN115810032B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to an image registration method and apparatus, a motion correction method and system for medical images, equipment, and media. Background Technology
[0002] Currently, image registration or motion correction functions typically require first setting a registration reference image, and then registering images from other time points to the space containing the reference image. The reference image is usually set manually by the user as an image at a specific time point, or a reference frame. Alternatively, depending on the characteristics of the perfusion acquisition process, the first, middle, or last time point may be selected as the reference frame. This method of selecting a reference image is susceptible to motion artifacts. If the registration reference image has image quality issues such as motion artifacts or ghosting of anatomical structures, the subsequent image registration results will be unreliable. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to overcome the defect that the method of determining the registration reference image in the prior art makes the subsequent image registration results unreliable, and to provide an image registration method and device, a motion correction method and system for medical images, equipment, and medium.
[0004] The present invention solves the above-mentioned technical problems through the following technical solution:
[0005] Firstly, an image registration method is provided, including:
[0006] Acquire medical image data containing the target object at multiple time points;
[0007] The degree of motion artifacts in the medical image data is determined, and several candidate image data frames are determined based on the balance position of the target object in the medical image data.
[0008] The medical image data with the lowest degree of motion artifacts among the candidate images is selected as the registration reference image data.
[0009] The medical image data is registered using the registration reference image data.
[0010] Optionally, several candidate image data frames are determined, including:
[0011] Based on the feature operator, the feature values of each frame of medical image data are calculated; the feature values represent the spatial location of the target object.
[0012] Based on the aforementioned feature values, all medical image data are divided to obtain at least one time point subset; the time point subset contains at least two frames of medical image data that are spatially close.
[0013] The subset of time points containing the largest number of medical image data is determined as the candidate subset, and the medical image data in the candidate subset is determined as the candidate image data.
[0014] Optionally, several candidate image data frames are determined based on the equilibrium position of the target object in the medical image data, including:
[0015] Feature extraction is performed on each of the medical image data, and cluster analysis is performed on each of the medical image data based on the extracted features to obtain multiple clusters;
[0016] The cluster containing the most medical image data among the multiple clusters is identified as a candidate cluster, and the medical image data in the candidate cluster is identified as the candidate image data.
[0017] Optionally, several candidate image data frames are determined, including:
[0018] For each frame of medical image data, the mean square error of the medical image data and other frames of medical image data is calculated, and the mean square error is summed to obtain the mean square error corresponding to each frame of medical image data.
[0019] Based on the sum of the mean square errors, determine the error threshold;
[0020] Images with mean square error less than the error threshold in each frame of medical image data are identified as candidate image data for that frame.
[0021] Optionally, image registration of the medical image data using the registration reference image data includes:
[0022] Based on the correspondence between object type and registration algorithm, determine the target registration algorithm that matches the object type of the target object;
[0023] The medical image data is registered using the registration reference image data according to the target registration algorithm.
[0024] Secondly, a motion correction method for medical images is provided, including:
[0025] The image registration method provided in the first aspect is used to perform image registration on medical image data to obtain spatial registration relationships;
[0026] Motion correction is performed on the medical image data based on the spatial registration relationship.
[0027] Optionally, it also includes:
[0028] Output alarm information for the medical image data; wherein the alarm information includes at least one of the following: degree of motion artifacts, identification of medical image data with poor motion correction, and similarity between each frame of medical image data and the registration reference image data.
[0029] Thirdly, an image registration apparatus is provided, comprising:
[0030] The acquisition module is used to acquire medical image data containing multiple time points of the target object;
[0031] The determination module is used to determine the degree of motion artifacts in the medical image data and to determine several candidate image data frames based on the balance position of the target object in the medical image data.
[0032] The selection module is used to select the medical image data with the lowest degree of motion artifacts among the candidate image data of the plurality of frames as the registration reference image data.
[0033] The registration module is used to perform image registration on the medical image data using the registration reference image data.
[0034] Fourthly, a motion correction system for medical images is provided, comprising:
[0035] The third aspect provides an image registration device for performing image registration on medical image data to obtain spatial registration relationships;
[0036] A correction device is used to perform motion correction on the medical image data according to the spatial registration relationship.
[0037] Optionally, it also includes:
[0038] An alarm module is used to output alarm information for the medical image data; wherein the alarm information includes at least one of the following: degree of motion artifacts, identification of medical image data with poor motion correction, and similarity between each frame of medical image data and the registration reference image data.
[0039] Fourthly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in any of the preceding claims.
[0040] Fifthly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.
[0041] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of the present invention.
[0042] The positive and progressive effects of this invention are as follows: This invention determines the registration reference image based on two dimensions: the degree of motion artifacts and the balance position. This avoids selecting images with poor image quality as registration reference images, and at the same time reduces the difficulty and cost of registration, thereby improving the quality and efficiency of image registration. Attached Figure Description
[0043] Figure 1 A flowchart of an image registration method provided as an exemplary embodiment of the present invention;
[0044] Figure 2 A flowchart of a motion correction method for medical images provided as an exemplary embodiment of the present invention;
[0045] Figure 3 A schematic diagram of a module for an image registration device provided as an exemplary embodiment of the present invention;
[0046] Figure 4 A schematic diagram of a motion correction system for medical images provided as an exemplary embodiment of the present invention;
[0047] Figure 5 This is a schematic diagram of the structure of an electronic device shown in an example embodiment of the present invention. Detailed Implementation
[0048] The present invention will be further illustrated by way of embodiments below, but the present invention is not limited to the scope of the embodiments described herein.
[0049] Figure 1 A flowchart of an image registration method provided as an exemplary embodiment of the present invention, the image registration method comprising the following steps:
[0050] Step 101: Obtain medical image data containing multiple time points of the target object.
[0051] One medical image data point at a time is also called one frame of medical image data, that is, one time point corresponds to one frame of medical image data, and multiple time points correspond to multiple frames of medical image data.
[0052] In one embodiment, a medical scan, such as a perfusion scan, is performed on the target object to obtain medical image data. Perfusion imaging is an increasingly widely used clinical imaging method. It is based on hemodynamics but differs from vascular imaging; it reflects the microcirculation or capillary network opening in tissues and observes the microscopic movement of molecules within the tissue. In perfusion imaging, taking CT (Computed Tomography) perfusion imaging as an example, multiple rapid perfusion scans are performed on the slice of interest of the target object after intravenous injection of contrast agent to obtain CT medical image data of the region of interest. Taking MR (Magnetic Resonance) perfusion imaging as an example, a rapid scanning sequence is used for continuous multi-slice multiple imaging to obtain MR medical image data.
[0053] In addition to CT and MR medical image data, the embodiments of the present invention are also applicable to image registration of other types of medical image data such as PET (Positron Emission Tomography) / CT medical image data, PET / MR medical image data, PET medical image data, and ultrasound medical image data.
[0054] In one embodiment, a dynamic contrast-enhanced scan of the target object is performed to obtain medical image data. Specifically, a 4D CT dynamic contrast-enhanced scan is performed on the target object.
[0055] The medical image data mentioned above may contain three-dimensional or two-dimensional images, and the embodiments of the present invention do not impose any particular limitation on this.
[0056] Step 102: Determine the degree of motion artifacts in the medical image data, and determine several candidate image data frames based on the balance position of the target object in the medical image data.
[0057] The candidate image data consists of medical image data showing the target object in an equilibrium position (average position). Several candidate image frames are medical image data from multiple time points where the target object is in an equilibrium position.
[0058] Candidate image data is determined through equilibrium position analysis of medical image data. Equilibrium position analysis involves statistically analyzing the feature data of medical image data at all time points and calculating the sequence numbers of several time points with the highest frequency of equilibrium positions of anatomical structures in the images based on the feature data. Motion correction of multi-time point images is a task of spatial registration and alignment of a set of images. Each time point image is a sample. Feature operators representing spatial positions are designed for all samples, and the corresponding feature values are calculated. Feature filtering conditions are set, and several time points with similar spatial positions are selected as the "optional range" of equilibrium positions. For all time point images, there may be multiple subsets of several time points with similar spatial positions. The subset with the larger number of such subsets is selected as the "optional range" of equilibrium positions. Several time point images within the "optional range" appear frequently throughout the perfusion scan and are close to the equilibrium position. The medical image data in this cluster are determined as candidate image data. Then, registration reference image data is selected from the candidate image data to reduce the difficulty of registering medical image data from other time points to the registration reference image data.
[0059] In this embodiment of the invention, in order to avoid selecting image data with poor image quality as the registration reference image, and from the perspective of registration difficulty and cost, the registration reference image data is determined according to two dimensions: the degree of motion artifacts and the balance position. Selecting the registration reference image data from these candidate image data frames can reduce the registration difficulty and cost.
[0060] In one embodiment, an artifact detection algorithm quantifies the degree of motion artifacts in each frame of medical image data. Motion artifacts refer to the phenomenon where the edges of anatomical structures (target objects) appear blurred and ghosted in the image due to unexpected movements of the patient during the scanning process, such as head tremors, body movements, and breathing. In this embodiment, the following artifact detection algorithms may be used, but are not limited to: Laplacian variance algorithm, fast Fourier transform algorithm, various pixel grayscale-based gradient functions, feature-based artifact detection algorithms, and machine learning-based artifact detection algorithms.
[0061] The following example, using the Laplacian algorithm in a brain perfusion application scenario, further illustrates how to determine the degree of motion artifacts in medical image data. During brain perfusion image acquisition, sudden head movements can cause image blurring, necessitating the identification of blur artifacts in medical image data at various time points. The Laplacian variance algorithm is used to calculate the amount of blur in medical image data, quantifying the sharpness of the image using variance values. Specifically, the Laplacian operator is used to convolve the input medical image data, followed by calculating the variance of the response map. The Laplacian operator measures the second derivative of the medical image data, highlighting regions with rapid gradient changes. If the variance of a frame of medical image data is higher than the variance threshold (i.e., high variance), it indicates broad response, representing a normal, focused medical image. However, if the variance of a frame is lower than the variance threshold (i.e., very low variance), there will be minimal response diffusion, indicating almost no edges in the image. The more blurred the medical image data, the fewer the edges, so variance can be used to detect whether medical image data is blurred.
[0062] The setting of the variance threshold is related to the medical image data used and is determined by factors such as the type of medical image data and its intended use. This is because different types of medical image data (including modality, grayscale range, contrast, etc.) and different intended uses all affect the appearance of the image, resulting in varying amounts of boundary information and different variance values calculated using the Laplacian operator. Therefore, it is necessary to set a variance threshold that matches the type of medical image data and its intended use. In perfusion scanning, motion artifacts and blurred or ghosted anatomical structures are often found only at a small number of time points in a multi-time-point medical image dataset. This can be considered an outlier detection problem.
[0063] In one embodiment, several candidate image data frames in medical image data are determined by: extracting features from each frame of medical image data, performing cluster analysis on each frame of medical image data based on the extracted features to obtain multiple clusters, determining the cluster containing the most medical image data in the multiple clusters as candidate clusters, and determining the medical image data in the candidate clusters as candidate image data.
[0064] In one embodiment, several candidate image data frames in medical image data are determined by: calculating feature values for each frame of medical image data according to feature operators, the feature values representing the spatial location of the target object; dividing all medical image data according to the feature values to obtain at least one time point subset; the time point subset contains at least two frames of medical image data with similar spatial locations; determining the time point subset containing the largest number of medical image data as the candidate subset, and determining the medical image data in the candidate subset as candidate image data.
[0065] For example, motion correction of multi-time-point images involves performing spatial registration and alignment on a set of images. Each time-point medical image data is a sample. Feature operators characterizing spatial location are designed for all samples, and corresponding feature values are calculated. Feature filtering conditions are set, and several frames of medical image data with similar spatial locations are selected. These frames are then grouped into the same time-point subset as the "optional range" for the equilibrium position. For all time-point medical image data, there may be multiple spatially similar time-point subsets; the subset with the larger number of subsets is selected as the "optional range" for the equilibrium position.
[0066] In one embodiment, several candidate image frames in the medical image data are determined as follows: For each frame of medical image data, the mean square error (MSE) of that medical image data and other frames of medical image data is calculated, and the MSEs are summed to obtain a sum of MSEs; an error threshold is determined based on the sum of MSEs, and medical image data frames whose sum of MSEs is less than the error threshold are identified as candidate image frames. Preferably, the error threshold is determined by performing quartile analysis on the sum of MSEs.
[0067] Features extracted from medical image data can characterize equilibrium positions. For example, the mean square error (MSE) of the gray values of the region of interest (mask region) of medical image data at each time point and the gray values of the regions of interest of medical image data at other time points can be calculated and summed to obtain the total mean square error. This total mean square error can be used as a feature of that time point.
[0068] Taking medical image data at five time points as an example, namely image data a1 to image data a5, the mean square error of image data a1 and image data a2 to image data a5 is calculated separately, resulting in four mean square error calculation results. These four mean square errors are then summed to obtain the mean square error sum corresponding to image data a1. Similarly, the mean square error of image data a2 and image data a1, and image data a3 to image data a5 are calculated separately, resulting in four mean square error calculation results. These four mean square errors are then summed to obtain the mean square error sum corresponding to image data a2. This process is repeated until the sum of mean squared errors corresponding to image data a3, a4, and a5 is obtained. The sum of mean squared errors for image data a1 represents a characteristic of image data a1, a2 represents a characteristic of image data a2, a3 represents a characteristic of image data a3, a4 represents a characteristic of image data a4, and a5 represents a characteristic of image data a5. Quartile analysis is performed on the five sums of mean squared errors to determine an error threshold. Medical image data with sums of mean squared errors less than the error threshold are identified as candidate frame image data.
[0069] Furthermore, taking a CT brain perfusion scenario as an example, during the contrast agent injection process, the grayscale values of brain tissue and blood vessels change. This change is independent of skull movement; therefore, the alignment of the skull in images at different time points can be evaluated by observing the spatial positions of internal or external landmarks, such as bones, which are unaffected by the contrast agent. In this embodiment, the alignment effect of the skull can be used as an evaluation method for image registration. Skeletal regions are extracted from medical image data to generate corresponding mask images. The bone mask is generated using bone segmentation methods, which include thresholding, region growing, and deep learning segmentation algorithms. The mean squared error (MSE) of each mask image and the mask images at other time points is calculated and summed to obtain the total MSE. All the obtained total MSE are sorted, and the threshold corresponding to the first quartile (Q1) is found according to the quartile method. Images with MSE below the Q1 threshold are determined as candidate image data and denoted as the "selectable range". Images within the "selectable range" are those that appear frequently throughout the perfusion scan and are close to the equilibrium position. Using such medical image data as registration reference image data can reduce the difficulty of registering other image data to the registration reference image data.
[0070] Step 103: Select the medical image data with the lowest degree of motion artifacts from several candidate image data frames as the registration reference image data.
[0071] In one embodiment, the degree of motion artifacts in medical image data at all time points is calculated, and the medical image data with the lowest degree of motion artifacts is selected as the registration reference image data from the candidate image data.
[0072] In one embodiment, the motion artifact level of several candidate image data frames is calculated, and the medical image data with the lowest motion artifact level is selected as the registration reference image data.
[0073] Candidate image data, close to the equilibrium position, is selected from several frames of candidate image data as registration reference image data, which can reduce the difficulty of registering other image data to the registration reference image data. Furthermore, the medical image data with the lowest degree of motion artifacts among several frames of candidate image data is the clearest medical image data, and using it as the registration reference image data can improve the quality and efficiency of image registration.
[0074] Step 104: Perform image registration on the medical image data using the registration reference image data.
[0075] In this embodiment of the invention, the registration reference image data is determined based on two dimensions: the degree of motion artifacts and the equilibrium position. This avoids selecting image data with poor image quality as the registration reference image data, while reducing the difficulty and cost of registration, thereby improving the quality and efficiency of image registration. The image registration method of this invention is applicable to motion correction in perfusion imaging, dynamic contrast-enhanced scanning, and dynamic fast (CINE) scanning, and features automation and quality control.
[0076] The target objects in dynamic scanning are diverse. Dynamic scanning includes perfusion imaging, dynamic contrast-enhanced scanning, and dynamic fast (CINE) scanning. Target objects can be, but are not limited to, organs such as the liver, heart, and head. Different organs have different motion characteristics. For example, in brain perfusion, the head's movement is rigid; in liver perfusion, the liver's position usually changes with respiration, and the liver can move 2 to 3 centimeters during respiration, which is non-rigid motion; in heart perfusion, heartbeat and respiration both affect the heart's position, and the heart's position often differs in the same slice scan. Therefore, different types of target objects have different motion characteristics and different image modal grayscale distributions, requiring different registration algorithms to be matched according to the type of target object.
[0077] In one embodiment, step 104 includes: determining a target registration algorithm that matches the object type of the target object based on the correspondence between object type and registration algorithm; and performing image registration on the medical image data using registration reference image data according to the target registration algorithm.
[0078] The image registration algorithm included in the correspondence includes at least one of the following: a medical image registration algorithm based on pixel grayscale information or image features, or a registration algorithm based on deep learning. The similarity measure in the medical image registration algorithm based on pixel grayscale information includes mutual information, cross-correlation coefficient, or mean squared error.
[0079] For example, a suitable spatial transformation model is selected based on the object type of the target object; depending on the spatial transformation model of the image registration algorithm, the output of the image registration algorithm can be divided into two categories: rigid transformation matrix and dense deformation field (also known as displacement field).
[0080] Figure 2 A flowchart of a motion correction method for medical images, provided as an exemplary embodiment of the present invention, is included in the following steps:
[0081] Step 201: Perform image registration on the medical image data using the image registration method provided in any of the above embodiments to obtain the spatial registration relationship.
[0082] The spatial registration relationship can be characterized, but is not limited to, through deformation fields or rigid transformation matrices.
[0083] Step 202: Perform motion correction on the medical image data based on the spatial registration relationship.
[0084] In this embodiment of the invention, during image registration, the selected registration reference image data is medical image data with good image quality and low registration difficulty and cost. Based on this registration reference image data, the efficiency and quality of image registration can be improved, anatomical structure position deviations caused by patient movement can be eliminated, thereby improving the efficiency and accuracy of motion correction. This embodiment of the invention proposes to use an optimized motion correction algorithm to compensate for common patient movements, applicable to motion correction in subsequent image processing such as perfusion imaging, dynamic contrast-enhanced scanning, and dynamic fast (CINE) scanning. It can eliminate medical image data with severe motion artifacts and poor scan quality, further improving the accuracy and processing speed of subsequent image processing.
[0085] In one embodiment, after step 202, alarm information for the medical image data is also output; wherein, the alarm information includes at least one of the following: the degree of motion artifacts, the identifier of the medical image data with poor motion correction, and the similarity between each frame of medical image data and the registered reference image data. The identifier of the medical image data may be, but is not limited to, represented by the time point of the medical image data; the similarity is the similarity between the motion-corrected medical image data and the registered reference image data.
[0086] The alarm information may, but is not limited to, take the form of a list, indicating the degree of motion artifacts in each frame of medical image, identifying medical image data with poor motion correction, and the similarity between each frame of medical image data and the registration reference image data.
[0087] In one embodiment, an alarm message for the medical image data is output only when the image parameters of the medical image data do not meet preset conditions. The image parameters include at least one of the following: the degree of motion artifacts, an identifier for medical image data with poor motion correction, and the similarity between each frame of medical image data and the registered reference image data. The corresponding preset conditions include a motion artifact degree less than an artifact threshold and / or a similarity greater than a similarity threshold.
[0088] The degree of motion artifacts in each registered image can be determined by, but is not limited to, the following artifact detection algorithms: Laplacian variance algorithm, fast Fourier transform algorithm, various gradient functions based on pixel gray level, feature-based artifact detection algorithms, machine learning-based artifact detection algorithms, etc. The embodiments of the present invention do not impose any particular limitation on the artifact detection algorithm.
[0089] Taking the Laplacian variance algorithm as an example, the Laplacian operator is first used to convolve the input registered image data, and then the variance of the response map is calculated. If the variance is lower than a predefined threshold, the registered image data is considered to be blurry and the degree of motion artifacts is high, and an alarm message is output.
[0090] In one embodiment, similarity analysis is performed on medical image data at all time points after motion correction to identify time points with poor motion correction. After motion correction, usually only a small number of time points have registration problems. This is because the registration algorithm used generally has fixed internal parameter settings. For large-scale positional changes or scenarios exceeding the expected range of change during algorithm development, the optimization process may not converge, resulting in registration failure. Poor image quality can also lead to unreliable registration algorithm results. Identifying the least similar registered image from a set of images can be considered an anomaly image or an outlier detection problem. Therefore, similarity can be used to exclude registered image data whose artifact levels do not meet preset requirements. This embodiment of the invention proposes to design a similarity metric to quantify the similarity or difference between the registered reference image data and medical image data at other time points.
[0091] Similarity metrics can be characterized by, but are not limited to, the following parameters: mean squared error, feature vector distance calculated from extracted features, mutual information, normalized cross-correlation, etc.
[0092] In one embodiment, all similarity metrics are sorted (ascending or descending) using the quartile method, the first quartile (Q1) and the third quartile (Q3) are found, and the interquartile range (IQR, the difference between the third quartile and the first quartile) is calculated. The similarity threshold is Q1 - 1.5IQR. If the similarity is less than Q1 - 1.5IQR, the corresponding registered image data is determined to be outlier, and an alarm message needs to be output.
[0093] In one embodiment, the difference between each registered image data and the registered reference image data is calculated. All differences are sorted (ascending or descending) using the quartile method. The first quartile (Q1) and the third quartile (Q3) are found. The interquartile range (IQR, the difference between the third quartile and the first quartile) is calculated. The difference threshold is Q3 + 1.5IQR. If the difference is greater than Q3 + 1.5IQR, the corresponding registered image data is determined to be outlier, and an alarm message needs to be output.
[0094] The following specific example will further illustrate the motion correction process.
[0095] During brain perfusion image acquisition, sudden head movements can cause image blurring, necessitating the identification of images containing blur artifacts at various time points. The Laplacian variance algorithm is employed to calculate the amount of blur in the images, quantifying image sharpness using variance values. Specifically, the input image is first convolved using the Laplacian operator, and then the variance of the response map is calculated. The Laplacian operator measures the second derivative of the image, highlighting regions with rapid gradient changes. A high variance indicates a broad response, typical of a normally focused image. Conversely, a low variance indicates minimal response diffusion, suggesting few edges in the image. Since blurrier images tend to have fewer edges, variance can be used to detect image blur.
[0096] The equilibrium position analysis method is used to calculate the time point number with equilibrium position for all time point images. The example is a CT brain perfusion scenario. During contrast agent injection, the grayscale values of brain tissue and blood vessels change, and this change is independent of the patient's head movement. Therefore, the alignment of the head at different time points should be evaluated by observing the spatial position of internal or external landmarks such as bones that are unaffected by the contrast agent. In this example, the alignment effect of the skull is considered as an evaluation method for motion correction. Bones are extracted from images at each time point to generate corresponding mask images. The bone masks are generated using bone segmentation methods, which include threshold segmentation, region growing, deep learning segmentation algorithms, etc. The mean squared error (MSE) of the mask image at each time point and the mask images at other time points are calculated and summed to obtain the total MSE. The total MSE of all time points is sorted, and the threshold corresponding to the first quartile (Q1) is found according to the quartile method. Time points where the total MSE is lower than the Q1 threshold are marked as the "selectable range". This indicates that several time points within the "selectable range" have a high frequency of occurrence during the entire perfusion scan and are close to the equilibrium position. Using such images as reference images can reduce the difficulty of registration to the reference image at other time points. Next, from the "selectable range", the image with the largest variance, i.e., the clearest image, is selected as the reference image based on the artifact quantization results obtained in the artifact detection stage.
[0097] During the motion correction phase, different motion correction algorithm schemes are set according to the motion characteristics of the organs and the grayscale distribution of the image modalities. In this embodiment, CT brain perfusion images are rich in grayscale information. Motion correction adopts an image registration algorithm based on mutual information. Considering the rigid motion characteristics of the head, the rotation angle and the scale of translation are predicted, and appropriate registration algorithm parameters are set.
[0098] After motion correction, quality control is performed on all time-point images after motion correction, and an artifact warning list and a motion warning list are output. In this embodiment, during the acquisition of brain perfusion images, sudden head movements can cause image blurring, and such images containing blur artifacts need to be excluded in subsequent perfusion calculations. The Laplacian variance algorithm can be used to calculate the amount of blur in the image. First, the input image is convolved using the Laplacian operator, and then the variance of the response map is calculated. If the variance is lower than a predefined threshold, the image is considered blurry. The threshold setting is related to the image set used and is determined by factors such as different image sets and different purposes, resulting in subjective uncertainty in the threshold setting. In perfusion scanning, motion artifacts and blurred or ghosted anatomical structures typically occur only at a few time points, which can be considered outlier detection problems. Here, we propose using a quartile method to sort the quantified artifact measure (blurring quantity) from smallest to largest, finding the first quartile (Q1) and third quartile (Q3), and calculating the interquartile range (IQR, the difference between the third quartile and the first quartile). Outliers are defined as values less than Q1 - 1.5IQR, and time points meeting this definition are added to the artifact warning list. In brain perfusion, evaluating the alignment of the skull can serve as an evaluation method for motion correction. Therefore, bones from the motion-corrected images at each time point are extracted to generate corresponding mask images. The bone masks are generated using bone segmentation methods, including thresholding, region growing, and deep learning segmentation algorithms. The mean squared error is calculated using bone structure information as a similarity measure. The mean squared error (MSE) between the mask image at each time point and the mask image at the reference time point is calculated. Here, the quartile method is proposed to sort all the MSEs from smallest to largest, find the first quartile (Q1) and the third quartile (Q3), and calculate the interquartile range (IQR, the difference between the third quartile and the first quartile). Time points with a MSE greater than Q3 + 1.5IQR are included in the motion warning list.
[0099] Corresponding to the aforementioned image registration method and motion correction method for medical images, the present invention also provides embodiments of an image registration device and a motion correction system for medical images.
[0100] Figure 3 A schematic diagram of an image registration device provided as an exemplary embodiment of the present invention, the image registration device comprising:
[0101] The acquisition module 31 is used to acquire medical image data containing multiple time points of the target object;
[0102] The determination module 32 is used to determine the degree of motion artifacts in the medical image data and to determine several candidate image data frames based on the balance position of the target object in the medical image data.
[0103] Selection module 33 is used to select the medical image data with the lowest degree of motion artifacts among the candidate image data of the plurality of frames as the registration reference image data;
[0104] The registration module 34 is used to perform image registration on the medical image data using the registration reference image data.
[0105] Optionally, the determined module includes:
[0106] A clustering unit is used to extract features from each medical image data and perform cluster analysis on each medical image data based on the extracted features to obtain multiple clusters;
[0107] The determining unit is used to determine the cluster containing the most medical image data among the multiple clusters as a candidate cluster, and to determine the medical image data in the candidate cluster as the candidate image data.
[0108] Optionally, the determined module includes:
[0109] The calculation unit is used to calculate the feature values of each frame of medical image data according to the feature operator; the feature values represent the spatial location of the target object;
[0110] A segmentation unit is used to segment all medical image data according to the feature value to obtain at least one time point subset; the time point subset contains at least two frames of medical image data that are spatially close.
[0111] The determining unit is used to determine the subset of time points containing the largest number of medical image data as a candidate subset, and to determine the medical image data in the candidate subset as the candidate image data.
[0112] Optionally, the determined module includes:
[0113] The calculation unit is used to calculate the mean square error of each frame of medical image data and other frames of medical image data, and to sum the mean square errors to obtain the mean square error corresponding to each frame of medical image data.
[0114] The determining unit is configured to determine an error threshold based on the sum of the mean squared errors, and to determine images in each frame of medical image data whose sum of the mean squared errors is less than the error threshold as candidate image data for the frame.
[0115] Optionally, it also includes:
[0116] The feature extraction module is used to determine the skeletal regions in each frame of the image and to extract features from the skeletal regions.
[0117] Optionally, the registration module includes:
[0118] The determining unit is equivalent to determining the target registration algorithm that matches the object type of the target object based on the correspondence between object type and registration algorithm;
[0119] A registration unit is used to perform image registration on the medical image data using the registration reference image data according to the target registration algorithm.
[0120] Figure 4 A schematic diagram of a motion correction system for medical images, provided as an exemplary embodiment of the present invention, is shown. The motion correction system for medical images includes:
[0121] The image registration device 41 provided in the above embodiments is used to perform image registration on medical image data to obtain spatial registration relationships;
[0122] The correction device 42 is used to perform motion correction on the medical image data according to the spatial registration relationship.
[0123] Optionally, it also includes:
[0124] An alarm module is used to output alarm information for the medical image data; wherein the alarm information includes at least one of the following: degree of motion artifacts, identification of medical image data with poor motion correction, and similarity between each frame of medical image data and the registration reference image data.
[0125] Figure 5 This is a schematic diagram of the structure of an electronic device according to an example embodiment of the present invention, showing a block diagram of an exemplary electronic device 50 suitable for implementing embodiments of the present invention. Figure 5 The electronic device 50 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.
[0126] like Figure 5 As shown, the electronic device 50 can be manifested in the form of a general-purpose computing device, such as a server device. The components of the electronic device 50 may include, but are not limited to: at least one processor 51, at least one memory 52, and a bus 53 connecting different system components (including memory 52 and processor 51).
[0127] Bus 53 includes a data bus, an address bus, and a control bus.
[0128] The memory 52 may include volatile memory, such as random access memory (RAM) 521 and / or cache memory 522, and may further include read-only memory (ROM) 523.
[0129] The memory 52 may also include a program tool 525 (or utility) having a set (at least one) program module 524, such program module 524 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.
[0130] The processor 51 performs various functional applications and data processing, such as the methods provided in any of the above embodiments, by running computer programs stored in the memory 52.
[0131] Electronic device 50 can also communicate with one or more external devices 54 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 55. Furthermore, the model-generated electronic device 50 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 56. As shown, network adapter 56 communicates with other modules of the model-generated electronic device 50 via bus 53. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated electronic device 50, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.
[0132] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0133] This invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in any of the above embodiments.
[0134] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0135] In a possible implementation, the present invention can also be implemented as a program product comprising program code, wherein when the program product is run on a terminal device, the program code is used to cause the terminal device to execute the method implementing any of the above embodiments.
[0136] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.
[0137] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.
Claims
1. An image registration method, characterized in that, include: Acquire medical image data containing the target object at multiple time points; The degree of motion artifacts in the medical image data is determined, and the medical image data at multiple time points are grouped according to the spatial location of the target object in the medical image data. The medical image data in the group containing the most medical image data is determined as several candidate image data frames. The medical image data with the lowest degree of motion artifacts among the candidate image data frames is selected as the registration reference image data. The medical image data is registered using the registration reference image data.
2. The image registration method according to claim 1, characterized in that, Based on the spatial location of the target object in the medical image data, the medical image data at multiple time points are grouped, and the medical image data in the group containing the largest number of medical image data are determined as several candidate image data frames, including: Feature extraction is performed on each frame of medical image data, and cluster analysis is performed on each frame of medical image data based on the extracted features to obtain multiple clusters; The cluster containing the most medical image data among the multiple clusters is identified as a candidate cluster, and the medical image data in the candidate cluster is identified as the candidate image data.
3. The image registration method according to claim 1, characterized in that, Based on the spatial location of the target object in the medical image data, the medical image data at multiple time points are grouped, and the medical image data in the group containing the largest number of medical image data are determined as several candidate image data frames, including: Based on the feature operator, the feature values of each frame of medical image data are calculated; the feature values represent the spatial location of the target object. Based on the aforementioned feature values, all medical image data are divided to obtain at least one time point subset; the time point subset contains at least two frames of medical image data that are spatially close. The subset of time points containing the largest number of medical image data is determined as the candidate subset, and the medical image data in the candidate subset is determined as the candidate image data.
4. The image registration method according to claim 1, characterized in that, Image registration of the medical image data using the registration reference image data includes: Based on the correspondence between object type and registration algorithm, determine the target registration algorithm that matches the object type of the target object; The medical image data is registered using the registration reference image data according to the target registration algorithm.
5. A motion correction method for medical images, characterized in that, include: The image registration method described in any one of claims 1-4 is used to perform image registration on medical image data to obtain spatial registration relationships; Motion correction is performed on the medical image data based on the spatial registration relationship.
6. The motion correction method for medical images according to claim 5, characterized in that, Also includes: Output alarm information for the medical image data; wherein the alarm information includes at least one of the following: degree of motion artifacts, identification of medical image data with poor motion correction, and similarity between each frame of medical image data and the registration reference image data.
7. An image registration device, characterized in that, include: The acquisition module is used to acquire medical image data containing multiple time points of the target object; The determination module is used to determine the degree of motion artifacts in the medical image data, and to group the medical image data at multiple time points according to the spatial location of the target object in the medical image data, and to determine the medical image data in the group containing the most medical image data as several frames of candidate image data. The selection module is used to select the medical image data with the lowest degree of motion artifacts among the candidate image data of the plurality of frames as the registration reference image data. The registration module is used to perform image registration on the medical image data using the registration reference image data.
8. A motion correction system for medical images, characterized in that, include: The image registration device according to claim 7 is used to perform image registration on medical image data to obtain spatial registration relationships; A correction device is used to perform motion correction on the medical image data according to the spatial registration relationship.
9. The motion correction system for medical images according to claim 8, characterized in that, Also includes: An alarm module is used to output alarm information for the medical image data; wherein the alarm information includes at least one of the following: degree of motion artifacts, identification of medical image data with poor motion correction, and similarity between each frame of medical image data and the registration reference image data.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 6.