A CT perfusion image post-processing method and device

Abstract

The application discloses a CT perfusion image post-processing method and device, wherein the method comprises the following steps: acquiring a CT perfusion image in aneurysmal subarachnoid hemorrhage, performing whole-brain voxelized parenchymal segmentation and midline calculation processing on the CT perfusion image; performing registration processing on the CT perfusion image subjected to the parenchymal segmentation and the midline calculation processing; identifying arteriovenous points in the CT perfusion image based on a deep learning model and calculating CT perfusion parameter values. The CT perfusion image is processed through flexible registration and rigid registration, which can improve image quality, facilitate quantitative analysis, improve image feature visualization, accurately position and realize cross-modality image fusion, and the like, and is helpful to improving the diagnosis and treatment level in the medical imaging field.

Description

Technical Field

[0001] This invention relates to the field of CT image processing technology, and in particular to a method and apparatus for post-processing CT perfusion images. Background Technology

[0002] CT perfusion is a medical imaging technique used to assess blood supply and hemodynamic parameters of organs or tissues. It involves performing multiple consecutive CT scans to observe and record the distribution and propagation of contrast agents in the blood to obtain blood flow information. CT perfusion is commonly used in the diagnosis and treatment planning of diseases of the brain, heart, and other organs. CT perfusion technology can help assess perfusion defects or abnormalities in diseases such as stroke and brain tumors, determine the presence and extent of ischemia in brain tissue, and guide early intervention and treatment decisions. However, it is important to note that CT perfusion is affected by factors such as resolution, contrast agent selection, and calculation methods. In clinical applications, multiple aspects need to be considered comprehensively, and a holistic analysis combining other imaging and clinical information is necessary.

[0003] Hypoperfusion is considered the most important pathophysiological process in aneurysmal subarachnoid hemorrhage (aSAH), and CT perfusion is considered an important imaging tool for identifying these changes in brain perfusion. However, most post-processing software currently used in international research for aneurysmal subarachnoid hemorrhage has the following problems: 1. It requires manual drawing of ROIs of interest; 2. There are large differences in the selection of perfusion parameters, and there is a lack of a gold standard for parameter selection; 3. The selection of analysis planes is limited, and there is a lack of whole-brain quantitative studies; 4. There is a lack of CT perfusion parameters that are highly specific to the aSAH field, and the software is not based on the pathophysiological characteristics of aSAH. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for post-processing CT perfusion images. Processing CT perfusion images through flexible and rigid registration can improve image quality, facilitate quantitative analysis, enhance the visualization of image features, provide precise localization, and enable cross-modal image fusion, thereby helping to improve the diagnostic and treatment levels in the field of medical imaging.

[0005] To address the aforementioned technical problems, a first aspect of this invention provides a method for post-processing CT perfusion images, comprising the following steps:

[0006] Step S100: Obtain CT perfusion images of aneurysmal subarachnoid hemorrhage, and perform whole-brain voxelization of the CT perfusion images for brain parenchyma segmentation and midline calculation. The brain parenchyma segmentation specifically includes the following steps:

[0007] Step S111: Denoise, remove skull bones, and / or normalize the CT perfusion images;

[0008] Step S112: Extract features that can distinguish brain structures from the processed CT perfusion images. Based on the characteristics of brain tissue and other tissues in the CT perfusion images, extract features that can distinguish brain parenchyma. The features include perfusion curves, local blood flow, and local blood volume.

[0009] Step S113: Based on the preset segmentation model, the CT perfusion image is segmented in combination with the features, and the segmented brain regions are converted into three-dimensional voxel representations.

[0010] Step S114: Post-process the segmented image, including: removing isolated points in small regions, smoothing boundaries, removing artifacts, repairing broken areas and / or filling holes;

[0011] The centerline calculation process specifically includes the following steps:

[0012] Step S121: Preprocess the CT perfusion image, including denoising and grayscale normalization;

[0013] Step S122: Select regions of interest on the arteries and veins in the preprocessed CT perfusion images;

[0014] Step S123: Extract the corresponding blood flow curve from each region of interest. The blood flow curve is calculated based on the perfusion curve and represents the density or blood flow that changes with vascular perfusion over time.

[0015] Step S124: Based on the shape and characteristics of the blood flow curve, the shape of the blood flow curve includes: the peak value and the trough value of the blood flow curve. The blood flow curve is divided into multiple time periods, and the peak point and trough point in each time period are found. The midline of the artery and vein is calculated.

[0016] Step S300: The CT perfusion image processed by brain parenchyma segmentation and midline calculation is registered. The registration process includes rigid registration and flexible registration, specifically including:

[0017] Step S310: Based on a convolutional neural network, extract the texture, color, and shape of the CT perfusion image for flexible registration;

[0018] Step S320: Perform rigid registration processing on CT perfusion images based on key point detection and feature matching;

[0019] Step S500: Based on a deep learning model, identify arteriovenous points in the CT perfusion image and calculate CT perfusion parameter values.

[0020] Furthermore, before performing registration processing on the CT perfusion images that have undergone brain parenchyma segmentation and midline calculation processing, the method further includes:

[0021] The CT perfusion images are subjected to geometric correction and / or grayscale correction.

[0022] Accordingly, a second aspect of the present invention provides a CT perfusion image post-processing apparatus, comprising:

[0023] The first image acquisition module is used to acquire CT perfusion images of aneurysmal subarachnoid hemorrhage, and to perform whole-brain voxelization of brain parenchyma segmentation and midline calculation processing on the CT perfusion images.

[0024] The first image processing module includes: a first preprocessing unit, a first feature extraction unit, a first image segmentation unit, and a first post-processing unit. The first preprocessing unit is used to perform noise reduction, skull removal, and / or standardization on the CT perfusion image. The first feature extraction unit is used to extract features that can distinguish brain structures from the processed CT perfusion image. Based on the characteristics of brain tissue and other tissues in the CT perfusion image, features that can distinguish brain parenchyma are extracted. The features include perfusion curve, local blood flow, and local blood volume. The first image segmentation unit is used to segment the CT perfusion image based on a preset segmentation model and in combination with the features, and convert the segmented brain regions into three-dimensional voxel representations. The first post-processing unit is used to perform post-processing on the segmented image. The post-processing includes: removing isolated points in small regions, smoothing boundaries, removing artifacts, repairing broken areas, and / or filling holes.

[0025] The first image processing module further includes: a second preprocessing unit, a second region selection unit, a second blood flow curve extraction unit, and a second midline calculation unit. The second preprocessing unit is used to preprocess the CT perfusion image, including denoising and grayscale normalization. The second region selection unit is used to select regions of interest on arteries and veins in the preprocessed CT perfusion image. The second blood flow curve extraction unit is used to extract the corresponding blood flow curve from each region of interest. The blood flow curve is calculated based on the perfusion curve and represents the density or blood flow rate that changes with vascular perfusion over time. The second midline calculation unit is used to divide the blood flow curve into multiple time periods based on the shape and characteristics of the blood flow curve, the shape of which includes the peak and trough values ​​of the blood flow curve, and find the peak and trough points in each time period to calculate the midline of the arteries and veins.

[0026] A registration processing module is used to perform registration processing on the CT perfusion images processed by brain parenchyma segmentation and midline calculation. The registration processing module performs rigid registration processing and flexible registration processing on the CT perfusion images. The registration processing module includes: a flexible registration processing unit and a rigid registration processing unit. The flexible registration processing unit extracts the texture, color and shape of the CT perfusion images based on a convolutional neural network to perform flexible registration. The rigid registration processing unit performs rigid registration processing on the CT perfusion images based on key point detection and feature matching.

[0027] The parameter calculation module is used to identify arteriovenous points in the CT perfusion image based on a deep learning model and calculate CT perfusion parameter values.

[0028] Furthermore, the CT perfusion image post-processing device also includes:

[0029] The second image processing module is used to perform geometric correction processing and / or grayscale correction processing on the CT perfusion image.

[0030] Accordingly, a third aspect of the present invention provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to cause the at least one processor to perform the CT perfusion image post-processing method described above.

[0031] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the CT perfusion image post-processing method described above.

[0032] The above-described technical solutions of the embodiments of the present invention have the following beneficial technical effects:

[0033] 1. Improve image quality: Registration processing can reduce or eliminate geometric distortions such as positional offset, rotation, and scaling in an image, thereby improving the spatial alignment and overall shape of the image;

[0034] 2. Facilitates quantitative analysis: Registration processing can make CT perfusion images at different time points or in different modalities spatially consistent, which facilitates quantitative analysis and comparison, such as blood flow calculation and perfusion parameter analysis.

[0035] 3. Improve the visualization of image features: Registration processing can align feature points or feature regions between different images, making the structure, contours and anatomical relationships of the image clearer, which helps doctors make accurate judgments in the process of diagnosis and treatment.

[0036] 4. Precise positioning: Registration processing can establish spatial correspondence between different CT perfusion images, thereby more accurately locating and marking specific areas, such as in surgical navigation.

[0037] 5. Cross-modal image fusion: In some cases, registration processing can fuse CT perfusion images from different modalities to obtain more comprehensive and integrated information, thereby improving the accuracy of diagnosis. Attached Figure Description

[0038] Figure 1 This is a flowchart of the CT perfusion image post-processing method provided in an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of CT perfusion image post-processing provided in an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of CT perfusion image parameter calculation provided in an embodiment of the present invention;

[0041] Figure 4 This is a block diagram of the CT perfusion image post-processing device module provided in an embodiment of the present invention;

[0042] Figure 5 This is a block diagram of the first image processing module provided in an embodiment of the present invention;

[0043] Figure 6 This is a block diagram of the registration processing module provided in an embodiment of the present invention.

[0044] Figure label:

[0045] 1. First image processing module; 11. First preprocessing unit; 12. First feature extraction unit; 13. First image segmentation unit; 14. First post-processing unit; 15. Second preprocessing unit; 16. Second region selection unit; 17. Second blood flow curve extraction unit; 18. Second midline calculation unit; 2. Registration processing module; 21. Flexible registration processing unit; 22. Rigid registration processing unit; 3. Parameter calculation module; 4. Second image processing module. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0047] Aneurysmal subarachnoid hemorrhage (aSAH) refers to intraarachnoid hemorrhage caused by the rupture of a cerebral aneurysm. It is a serious neurosurgical condition with symptoms including severe headache, sudden loss of consciousness, nausea, vomiting, and neck stiffness. The cause is usually due to a weak wall of the cerebral aneurysm, leading to rupture and subsequent cerebral hemorrhage.

[0048] The following section uses aSAH as an example to illustrate this technical solution in detail. However, the technical solution of this application is not limited to aSAH and can also be applied to other neurosurgical vascular diseases with ischemic characteristics, such as hypertensive intracerebral hemorrhage, arteriovenous malformation, moyamoya disease, traumatic intracerebral hemorrhage, and hydrocephalus.

[0049] Please refer to Figure 1 , Figure 2 and Figure 3 The first aspect of this invention provides a post-processing method for CT perfusion images, comprising the following steps:

[0050] Step S100: Obtain CT perfusion images of aSAH, and perform whole-brain voxelization of brain parenchyma segmentation and midline calculation on the CT perfusion images.

[0051] Brain parenchyma segmentation in CT perfusion images aims to separate brain tissue from other tissue regions (such as the skull and cerebrospinal fluid) for analysis and quantitative assessment of brain diseases. The specific brain parenchyma segmentation method includes the following steps:

[0052] 1. Data Preprocessing: Preprocessing of CT perfusion images includes denoising, grayscale normalization, and skull removal. These steps help improve the accuracy of subsequent segmentation.

[0053] 2. Feature Extraction: Based on the characteristics of brain tissue compared to other tissues in CT perfusion images, features that can distinguish brain parenchyma are extracted. Commonly used features include perfusion curves (time-density curves), local blood flow, and local blood volume.

[0054] 3. Segmentation Algorithms: Selecting an appropriate segmentation algorithm to separate brain tissue from the image. Commonly used segmentation algorithms include thresholding, region growing, graph cuts, and deep learning methods.

[0055] 4. Post-processing: Further processing of the segmentation results, such as removing isolated points in small regions, filling holes, and smoothing boundaries. These steps help improve the accuracy and consistency of the segmentation results.

[0056] Step S300: Register the CT perfusion images processed by brain parenchyma segmentation and midline calculation.

[0057] Registration of CT perfusion images involves spatially aligning CT perfusion images from different time points or different image modalities to facilitate quantitative analysis and comparison. The following are methods for CT perfusion image registration:

[0058] 1. Data preparation: Collect CT perfusion image data that need to be registered, including image data acquired at different time points or in different modalities.

[0059] 2. Preprocessing: Preprocessing of CT perfusion images includes noise reduction and grayscale normalization. These steps help improve the accuracy of registration.

[0060] 3. Feature Extraction: Based on regions with contrast or distinctive features in the image, key feature points are extracted as reference points for registration. Commonly used features include vascular branches, anatomical structures, or perfusion curves.

[0061] 4. Registration Algorithms: Selecting a suitable registration algorithm for image registration. Commonly used methods include feature point-based registration, cross-correlation-based registration, and dominance point-based registration. These algorithms can transform the image based on the location and matching degree of feature points.

[0062] 5. Optimization and Evaluation: The registration results should be optimized and evaluated. Error metrics (such as root mean square error) can be used to assess the accuracy and stability of the registration. If the registration results are unsatisfactory, further optimization and adjustments can be made.

[0063] The goal of CT perfusion image registration is to maintain spatial consistency between different images to facilitate quantitative analysis and comparison. During registration, appropriate feature points and registration algorithms need to be selected, and other image processing and analysis techniques should be combined according to specific application requirements to obtain accurate and reliable registration results.

[0064] Step S500: Based on a deep learning model, identify arteriovenous points in CT perfusion images and calculate CT perfusion parameter values.

[0065] Deep learning models, trained on existing large-scale population data, can quickly identify arteries and veins in CT perfusion images, meeting the needs of clinical and emergency departments.

[0066] Specifically, identifying arteriovenous points in CT perfusion images typically requires blood flow curve analysis to distinguish between arteries and veins. First, a Region of Interest (ROI) is selected: an appropriate ROI is chosen on the CT perfusion image, usually within a vascular region or a specific brain region. ROIs can be drawn manually or automatically selected using image processing algorithms. Second, blood flow curves are extracted: a corresponding blood flow curve is extracted from each ROI, representing the density or blood flow rate that changes over time with vascular perfusion. The blood flow curve can be obtained by calculating the average absorbance or density value within the ROI. Third, the blood flow curves are analyzed: arteriovenous points are identified based on the morphology and characteristics of the blood flow curves. Arterial blood flow curves typically exhibit rapid rises and falls, while venous blood flow curves are relatively flat; thresholding, slope calculation, and peak detection methods can be used to identify arteriovenous points. Finally, the results are validated: the identification results are validated and corrected based on known vascular anatomy and other image information. This can be accomplished by comparing with other imaging modalities (such as cerebral angiography).

[0067] Whole-brain voxelization segmentation can be used for the analysis and quantitative assessment of brain diseases, such as for analyzing perfusion parameters and calculating cerebral hemodynamic parameters.

[0068] In one specific embodiment of the present invention, step S100, which involves whole-brain voxelization of the CT perfusion image for brain parenchyma segmentation, includes:

[0069] Step S111: Denoise, skull removal and / or standardization are performed on the CT perfusion images.

[0070] Preprocessing CT perfusion images, including denoising and grayscale normalization, helps improve the accuracy of subsequent segmentation.

[0071] Step S112: Extract features that can distinguish brain structures from the processed CT perfusion images. Based on the characteristics of brain tissue and other tissues in the CT perfusion images, extract features that can distinguish brain parenchyma. These features include perfusion curves, local blood flow, and local blood volume.

[0072] Step S113: Based on the preset segmentation model, the CT perfusion image is segmented by combining features, and the segmented brain regions are converted into three-dimensional voxel representations.

[0073] Based on a pre-defined segmentation model, segmentation algorithms are used to separate brain tissue from other tissue regions (such as skull, cerebrospinal fluid, etc.) in CT perfusion images. Segmentation methods include thresholding, region growing, graph cut, and deep learning methods. The segmented brain regions are then converted into three-dimensional voxel representations. Voxels are pixels in three-dimensional space, and their color or numerical value represents the type or attribute of the brain tissue. Mapping the brain tissue segmentation results to voxel representations facilitates subsequent analysis and processing.

[0074] Step S114: Post-process the segmented image. Post-processing includes: removing isolated points in small regions, smoothing boundaries, removing artifacts, repairing broken areas, and / or filling holes.

[0075] Further processing of the voxelized segmentation results, such as noise removal, hole filling, and boundary smoothing, helps improve the accuracy and consistency of the segmentation results.

[0076] Furthermore, by combining voxel-based parameter calculation with conventional methods (e.g., deconvolution) parameter calculation, various parameter values ​​of CT perfusion images, such as CBF (cerebral blood flow), CBV (cerebral blood flow volume), MTT (mean transit time), and Tmax (time to peak impulse residual function), are obtained. Additionally, in Figure 2 In the text, IRF TO (ischemic reaction factor) represents ischemic reaction factor.

[0077] Specifically, by combining deconvolution parameter calculation and voxel-based parameter calculation through a fusion correction algorithm, more accurate and comprehensive image analysis and quantification can be achieved. First, deconvolution parameter calculation is primarily used to recover high-frequency details and fine structures in the image, improving resolution and sharpness. This algorithm removes blur from blurred images by establishing the point spread function (PSF) of the imaging system and the inverse process of blurring. Second, voxel-based parameter calculation is used for quantitative analysis of the image, obtaining numerical results related to image content, shape, and features. These parameters can include volume, density, diameter, surface area, connectivity, uniformity, etc., providing a quantitative description of different structures and tissues in the image.

[0078] The fusion correction algorithm combines these two calculation methods. The specific process is as follows: (1) First, the image to be processed is preprocessed, including denoising and grayscale normalization, to reduce noise and enhance image contrast. (2) The deconvolution algorithm is applied to correct the image to restore the details and high-frequency components of the image. The image after deconvolution correction is obtained, making subsequent calculations more accurate and reliable. (3) Feature extraction is performed on the corrected image to extract key feature points or feature regions, and the corresponding parameters are calculated using a voxel-based parameter calculation method. (4) The deconvolution correction and voxel-based parameter calculation results are fused, which can be achieved through weighted averaging, logical operations, or other fusion techniques. The fused result can provide more accurate and comprehensive image analysis results. By using the fusion correction algorithm to realize deconvolution parameter calculation and voxel-based parameter calculation, the advantages of image fidelity and quantitative analysis can be comprehensively utilized to improve image quality and the accuracy of analysis results. The specific implementation method and parameter selection need to be determined according to the specific application scenario and data characteristics to meet actual needs and goals.

[0079] Furthermore, midline calculation processing of CT perfusion images involves analyzing blood flow curves to calculate the midlines of arteries and veins, thereby quantitatively assessing hemodynamic parameters. The specific midline calculation processing method includes the following steps:

[0080] Step S121, Data Preprocessing: Preprocessing the CT perfusion images, including denoising and grayscale normalization. These steps help improve the accuracy of subsequent analysis.

[0081] Step S122, Region of Interest (ROI) selection: Select an appropriate ROI on the CT perfusion image, usually on arteries and veins; the ROI can be drawn manually or automatically selected based on image processing algorithms.

[0082] Step S123, Blood Flow Curve Extraction: Extract the corresponding blood flow curve from each ROI. These curves are usually calculated based on the time-density curve, which represents the density or blood flow that changes over time with vascular perfusion.

[0083] Step S124, Midline Calculation: Calculate the midlines of arteries and veins based on the shape and characteristics of the blood flow curve. This can be achieved by identifying peaks, troughs, or other characteristic points in the blood flow curve. The calculation method involves dividing the curve into multiple time intervals, finding the peak and trough points within each time interval, and calculating the midline based on the positions of these points.

[0084] Midline calculation can be used to quantify hemodynamic parameters, such as the perfusion difference between arteries and veins, and blood flow velocity. It is of great significance for assessing cerebral blood flow and for decision-making and monitoring vascular diseases. When performing midline calculation, it is necessary to select an appropriate Region of Interest (ROI), use suitable feature points and calculation methods, and combine them with other image processing and analysis techniques to obtain accurate and reliable results.

[0085] Optionally, before registering the CT perfusion images processed by brain parenchyma segmentation and midline calculation in step S300, the process further includes:

[0086] Step S200: Perform geometric correction and / or grayscale correction on the CT perfusion image.

[0087] Geometric correction of CT perfusion images refers to performing spatial geometric transformations on the images to correct image distortions caused by geometric distortions or other factors during image acquisition. The specific process is as follows: (1) Data preparation: collect CT perfusion image data that need to be geometrically corrected; (2) Image calibration: obtain a set of reference calibration images with known geometric relationships; (3) Feature extraction: extract feature points or feature regions on the calibration images and the images to be corrected; (4) Transformation model selection: select a suitable geometric transformation model according to the actual situation. The above models include translation, rotation, scaling, affine transformation, etc.; (5) Parameter estimation: estimate the parameters of the geometric transformation model using the feature points on the calibration images and the corresponding feature points on the images to be corrected through the least squares method or other optimization methods; (6) Transformation application: apply the obtained geometric transformation model parameters to the images to be corrected, and correct the geometric distortions in the images through transformation operations.

[0088] Furthermore, grayscale correction of CT perfusion images involves adjusting the image's grayscale values ​​to more accurately reflect the actual gray levels. The goal of grayscale correction is to improve image contrast and accuracy, enabling clearer differentiation and identification of different tissues and structures within the image. When performing grayscale correction, it is necessary to select a reference calibration image, feature extraction method, and grayscale correction calculation method, and adjust the parameters according to actual needs to obtain accurate and reliable correction results.

[0089] In one specific embodiment of the present invention, step S300 involves registering the CT perfusion images processed by brain parenchyma segmentation and midline calculation, including:

[0090] Rigid registration and flexible registration were combined in the CT perfusion images.

[0091] Furthermore, the flexible registration processing of the CT perfusion images in step S300 includes:

[0092] Step S310: Based on a convolutional neural network, extract the texture, color, and shape of the CT perfusion image for flexible registration.

[0093] Specifically, flexible registration of CT perfusion images involves considering image deformation during the registration process to more accurately align variations between different images. In flexible registration, feature extraction is performed, extracting key feature points or regions with contrast or distinctive features from the images as registration references. These features can be vascular branches, anatomical structures, or other specific regions. Furthermore, an appropriate image deformation model can be selected to describe the deformation and differences between CT perfusion images; the model can be an elastic deformation model, a non-rigid deformation model, etc., or a suitable model can be chosen based on the specific circumstances. Optimization methods (such as least squares, gradient descent, etc.) are used to adjust the parameters of the deformation model to minimize the differences between images. The optimization goal is to make the registered image as similar as possible to the reference image. The results of flexible registration can be evaluated using error metrics (such as root mean square error) to assess the accuracy and stability of the registration. If the registration results are unsatisfactory, further optimization adjustments can be made.

[0094] Flexible registration can more accurately account for deformations and differences between CT perfusion images, making it suitable for image registration tasks requiring significant shape changes and localized deformations. When performing flexible registration, it is necessary to select appropriate feature points or regions, deformation models, and optimization methods, and adjust parameters according to specific needs to obtain accurate and reliable registration results.

[0095] Further, the rigid registration process for the CT perfusion image in step S300 includes:

[0096] Step S320: Perform rigid registration processing on CT perfusion images based on key point detection and feature matching.

[0097] Rigid registration of CT perfusion images involves performing rigid transformations such as rotation, translation, and scaling on different images to align them spatially. During rigid registration, key feature points or regions are extracted from areas of contrast or distinctiveness in the images as registration references. These features can be vascular branches, anatomical structures, or other contrasting regions. Rigid transformation models can be selected, typically including rotation, translation, and scaling. For CT perfusion images, the rigid transformation model can be implemented by calculating global transformation parameters between images. Furthermore, optimization methods, such as least squares and gradient descent, can be used to adjust the parameters of the rigid transformation model to minimize differences between images; the optimization goal is to make the registered image as similar as possible to the reference image.

[0098] Rigid registration can perform simple rigid transformations such as translation, rotation, and scaling on CT perfusion images to handle cases where the overall shape differences between images are small. When performing rigid registration, it is necessary to select appropriate feature points or regions, rigid transformation models, and optimization methods, and adjust parameters according to specific needs to obtain accurate and reliable registration results.

[0099] Accordingly, please refer to Figure 4 A second aspect of the present invention provides a CT perfusion image post-processing apparatus, comprising:

[0100] The first image processing module 1 is used to acquire CT perfusion images of the whole brain and perform brain parenchyma segmentation and midline calculation processing on the CT perfusion images.

[0101] Registration processing module 2 is used for registration processing of CT perfusion images that have undergone brain parenchyma segmentation and midline calculation processing;

[0102] The parameter calculation module 3 is used to identify arteriovenous points in CT perfusion images based on a deep learning model and calculate CT perfusion parameter values.

[0103] Further, please refer to Figure 5 The first image processing module 1 performs whole-brain voxel segmentation on the CT perfusion images. The first image processing module 1 specifically includes:

[0104] The first preprocessing unit 11 is used to perform noise reduction, correction and / or normalization on CT perfusion images;

[0105] The first feature extraction unit 12 is used to extract features of distinguishable brain structures in the processed CT perfusion image;

[0106] The first image segmentation unit 13 is used to segment CT perfusion images based on a preset segmentation model and combined with features;

[0107] The first post-processing unit 14 is used to perform post-processing on the segmented image, including: removing artifacts, repairing broken areas and / or filling holes.

[0108] The first image processing module 1 further includes: a second preprocessing unit 15, a second region selection unit 16, a second blood flow curve extraction unit 17, and a second midline calculation unit 18;

[0109] The second preprocessing unit 15 is used to preprocess CT perfusion images, including noise reduction and grayscale normalization.

[0110] The second region selection unit 16 is used to select regions of interest on arteries and veins in the preprocessed CT perfusion images;

[0111] The second blood flow curve extraction unit 17 is used to extract the corresponding blood flow curve from each region of interest. The blood flow curve is calculated based on the perfusion curve and represents the density or blood flow rate that changes with vascular perfusion over time.

[0112] The second midline calculation unit 18 is used to divide the blood flow curve into multiple time periods based on the shape and characteristics of the blood flow curve, including the peak and trough values ​​of the blood flow curve, and to find the peak and trough points in each time period, and to calculate the midline of the artery and vein.

[0113] Further, please refer to Figure 6 The registration processing module 2 performs rigid registration and flexible registration processing on CT perfusion images.

[0114] Furthermore, the registration processing module 2 includes: a flexible registration processing unit 21;

[0115] The flexible registration processing unit 21 is based on a convolutional neural network to extract the texture, color and shape of CT perfusion images for flexible registration.

[0116] Furthermore, the registration processing module 2 also includes: a rigid registration processing unit 22;

[0117] The rigid registration processing unit 22 performs rigid registration processing on CT perfusion images based on key point detection and feature matching.

[0118] Furthermore, the CT perfusion image post-processing device also includes:

[0119] The second image processing module 4 is used to perform geometric correction processing and / or grayscale correction processing on CT perfusion images.

[0120] Accordingly, a third aspect of the present invention provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by a processor, the instructions being executed by the processor to cause the at least one processor to perform the CT perfusion image post-processing method described above.

[0121] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the CT perfusion image post-processing method described above.

[0122] This invention aims to protect a method and apparatus for post-processing CT perfusion images. The method includes the following steps: acquiring CT perfusion images of aneurysmal subarachnoid hemorrhage; performing brain parenchyma segmentation and midline calculation on the CT perfusion images; performing registration processing on the segmented and midline-calculated CT perfusion images; and identifying arteriovenous points in the CT perfusion images based on a deep learning model, and calculating CT perfusion parameter values. The above technical solution has the following advantages:

[0123] 1. Improve image quality: Registration processing can reduce or eliminate geometric distortions such as positional offset, rotation, and scaling in an image, thereby improving the spatial alignment and overall shape of the image;

[0124] 2. Facilitates quantitative analysis: Registration processing can make CT perfusion images at different time points or in different modalities spatially consistent, which facilitates quantitative analysis and comparison, such as blood flow calculation and perfusion parameter analysis.

[0125] 3. Improve the visualization of image features: Registration processing can align feature points or feature regions between different images, making the structure, contours and anatomical relationships of the image clearer, which helps doctors make accurate judgments in the process of diagnosis and treatment.

[0126] 4. Precise positioning: Registration processing can establish spatial correspondence between different CT perfusion images, thereby more accurately locating and marking specific areas, such as in surgical navigation.

[0127] 5. Cross-modal image fusion: In some cases, registration processing can fuse CT perfusion images from different modalities to obtain more comprehensive and integrated information, thereby improving the accuracy of diagnosis.

[0128] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0130] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0131] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for post-processing CT perfusion images, characterized in that, Includes the following steps: Step S100: Obtain CT perfusion images of aneurysmal subarachnoid hemorrhage, and perform whole-brain voxelization of the CT perfusion images for brain parenchyma segmentation and midline calculation. The brain parenchyma segmentation specifically includes the following steps: Step S111: Denoise, skull removal and / or standardization are performed on the CT perfusion images; Step S112: Extract features that can distinguish brain structures from the processed CT perfusion images. Based on the characteristics of brain tissue and other tissues in the CT perfusion images, extract features that can distinguish brain parenchyma. The features include perfusion curve, local blood flow, and local blood volume. Step S113: Based on the preset segmentation model, the CT perfusion image is segmented in combination with the features, and the segmented brain regions are converted into three-dimensional voxel representations. Step S114: Post-process the segmented image. Post-processing includes: removing isolated points in small regions, smoothing boundaries, removing artifacts, repairing broken areas, and / or filling holes. The centerline calculation process specifically includes the following steps: Step S121: Preprocess the CT perfusion image, including denoising and grayscale normalization; Step S122: Select regions of interest on the arteries and veins in the preprocessed CT perfusion images; Step S123: Extract the corresponding blood flow curve from each region of interest. The blood flow curve is calculated based on the perfusion curve and represents the density or blood flow that changes with vascular perfusion over time. Step S124: Based on the shape and characteristics of the blood flow curve, the shape of the blood flow curve includes: the peak value and the trough value of the blood flow curve. The blood flow curve is divided into multiple time periods, and the peak point and trough point in each time period are found. The midline of the artery and vein is calculated. Step S300: The CT perfusion image processed by brain parenchyma segmentation and midline calculation is registered. The registration process includes rigid registration and flexible registration, specifically including: Step S310: Based on a convolutional neural network, extract the texture, color, and shape of the CT perfusion image for flexible registration; Step S320: Perform rigid registration processing on CT perfusion images based on key point detection and feature matching; Step S500: Based on a deep learning model, identify arteriovenous points in the CT perfusion image and calculate CT perfusion parameter values.

2. The CT perfusion image post-processing method according to claim 1, characterized in that, Before registering the CT perfusion images processed by brain parenchyma segmentation and midline calculation, the method further includes: The CT perfusion images are subjected to geometric correction and / or grayscale correction.

3. A CT perfusion image post-processing device, characterized in that, include: The first image acquisition module is used to acquire CT perfusion images of aneurysmal subarachnoid hemorrhage, and to perform whole-brain voxelization of brain parenchyma segmentation and midline calculation processing on the CT perfusion images. The first image processing module includes: a first preprocessing unit, a first feature extraction unit, a first image segmentation unit, and a first post-processing unit. The first preprocessing unit is used to perform noise reduction, skull removal, and / or standardization on the CT perfusion image. The first feature extraction unit is used to extract features that can distinguish brain structures from the processed CT perfusion image. Based on the characteristics of brain tissue and other tissues in the CT perfusion image, features that can distinguish brain parenchyma are extracted. The features include perfusion curve, local blood flow, and local blood volume. The first image segmentation unit is used to segment the CT perfusion image based on a preset segmentation model and in combination with the features, and convert the segmented brain regions into three-dimensional voxel representations. The first post-processing unit is used to perform post-processing on the segmented image. The post-processing includes: removing isolated points in small regions, smoothing boundaries, removing artifacts, repairing broken areas, and / or filling holes. The first image processing module further includes: a second preprocessing unit, a second region selection unit, a second blood flow curve extraction unit, and a second midline calculation unit. The second preprocessing unit is used to preprocess the CT perfusion image, including denoising and grayscale normalization. The second region selection unit is used to select regions of interest on arteries and veins in the preprocessed CT perfusion image. The second blood flow curve extraction unit is used to extract the corresponding blood flow curve from each region of interest. The blood flow curve is calculated based on the perfusion curve and represents the density or blood flow rate that changes with vascular perfusion over time. The second midline calculation unit is used to divide the blood flow curve into multiple time periods based on the shape and characteristics of the blood flow curve, the shape of which includes the peak and trough values ​​of the blood flow curve, and find the peak and trough points in each time period to calculate the midline of the arteries and veins. A registration processing module is used to perform registration processing on the CT perfusion images processed by brain parenchyma segmentation and midline calculation. The registration processing module performs rigid registration processing and flexible registration processing on the CT perfusion images. The registration processing module includes: a flexible registration processing unit and a rigid registration processing unit. The flexible registration processing unit extracts the texture, color and shape of the CT perfusion images based on a convolutional neural network to perform flexible registration. The rigid registration processing unit performs rigid registration processing on the CT perfusion images based on key point detection and feature matching. The parameter calculation module is used to identify arteriovenous points in the CT perfusion image based on a deep learning model and calculate CT perfusion parameter values.

4. The CT perfusion image post-processing apparatus according to claim 3, characterized in that, Also includes: The second image processing module is used to perform geometric correction processing and / or grayscale correction processing on the CT perfusion image.

5. An electronic device, characterized in that, include: At least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the processor to cause the at least one processor to perform the CT perfusion image post-processing method as described in claim 1 or 2.

6. A computer-readable storage medium, characterized in that, It stores computer instructions, which, when executed by a processor, implement the CT perfusion image post-processing method as described in claim 1 or 2.

Images