Method and apparatus for fully automatic selection of arterial input function in CT perfusion imaging

Abstract

This application discloses a method and apparatus for fully automatic selection of arterial input functions in CT perfusion imaging. The method includes: obtaining a candidate layer of middle arteries after skull removal, wherein voxels are of the first type; selecting a second type of voxels with a first expected proportion based on the first type of voxels; normalizing the area under the curve of the density-time curve of the second type of voxels to obtain the roughness of the normalized curve, and selecting a third type of voxels with a second expected proportion; performing a first cluster analysis on the third type of voxels based on the first cluster target number, and selecting a fourth type of voxels with the smallest average first moment; statistically analyzing the distribution density of the fourth type of voxels to obtain a fifth type of voxels, performing a second cluster analysis on the second cluster target number, and selecting a category with the smallest average peak time, wherein the voxels form an arterial candidate point group; performing connected component analysis on the arterial candidate point group to obtain an arterial result point group, whose average density-time curve is the arterial input function.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method and apparatus for fully automatic selection of arterial input functions in CT perfusion images. Background Technology

[0002] CT perfusion imaging is commonly used to examine cerebral blood perfusion in stroke patients, assessing the patient's condition by analyzing perfusion parameters such as cerebral blood volume (CBV) and cerebral blood flow (CBF). Based on contrast agent dilution theory, the contrast agent density-time curve (Ct) within the tissue can be obtained by convolving the arterial input function (AIF) with the contrast agent residual curve (r). Further calculations on curve (r) yield various perfusion parameters. Curves Ct and AIF can be obtained from CT perfusion images. AIF selection is a crucial step in CT perfusion analysis. In CT perfusion analysis of human brain tissue, multiple studies recommend selecting the density-time curve within the middle cerebral artery (MCA) as the AIF.

[0003] Existing AIF selection methods can be broadly categorized into manual and semi-automatic methods, with few fully automatic methods. Manual AIF selection relies on operator experience, is time-consuming and inefficient, and has low repeatability, failing to meet the requirements of rapid detection. Semi-automatic selection methods require pre-selection of the region of interest, followed by subsequent calculations using feature-weighted models or clustering methods, which are also inefficient. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and apparatus for fully automatic selection of arterial input functions in CT perfusion images to address the aforementioned technical problems.

[0005] This application provides a method for fully automatic selection of arterial input functions in CT perfusion imaging, including:

[0006] A candidate layer of the middle cerebral artery is obtained after skull removal. The candidate layer contains voxels of type I. Density-time curve baseline correction is performed on the type I voxels. The corrected density-time curve is used to perform the following steps:

[0007] Integrate the corrected density-time curve, and select a second type of voxels with a first expected proportion from the first type of voxels.

[0008] The area under the curve of the density-time curve of the second type of voxel is normalized to obtain the roughness of the normalized curve, and the third type of voxel with the second expected ratio is screened to obtain the roughness of the normalized curve.

[0009] The third type of voxels undergoes a first cluster analysis based on the first cluster target number to screen out the fourth type of voxels with the smallest average first moment.

[0010] The distribution density of the fourth type of voxels is statistically analyzed. In the candidate layer of the middle artery, a desired number of layers are selected, and the voxels in the desired number of layers are used as the fifth type of voxels.

[0011] The density-time curves of the fifth type of voxels are obtained, and a second clustering analysis of the second clustering target number is performed to screen out the category with the smallest average peak time. The voxels in this category form the artery candidate point group.

[0012] A mask image of the candidate artery point group is obtained, and the connected region with the largest area is obtained by the maximum connected component algorithm as the artery result point group. The average density time curve of the artery result point group is the artery input function.

[0013] The process of obtaining the candidate layer for the middle artery during skull removal specifically includes:

[0014] The raw CT perfusion images were obtained, and preprocessing and reading of brain tissue masks after skull removal were performed sequentially.

[0015] The area of ​​the brain tissue mask is obtained, and a first distance is found from the layer with the maximum area towards the skull base to screen and obtain the candidate layer of the middle artery.

[0016] The preprocessing includes motion correction, image filtering, and skull removal.

[0017] Among the first type of voxels, the density-time curve integral value of the second type of voxels is greater than that of the other voxels.

[0018] Among the second type of voxels, the roughness of the normalization curve of the third type of voxel is smaller than that of the other voxels.

[0019] The roughness of the normalized curve is obtained by the following formula: In the formula, C” norm (t) is the normalized curve C norm R(v) is the second derivative of (t), and R(v) is the integral of the square of the second derivative over the time dimension.

[0020] The first cluster analysis and the second cluster analysis both include:

[0021] Initialize the class center curves, calculate the high-dimensional distance between each voxel density time curve or normalized curve and each class center curve, and assign each voxel density time curve or normalized curve to the class with the smallest distance. Recalculate the class center curves, and repeat the distance calculation and assignment operations through multiple iterations until the classification results no longer change.

[0022] The first moment is obtained by calculating the following formula using the normalized curves of each voxel: μ=∑t*C norm (t).

[0023] The expected number of layers refers to the layers in the candidate layer of the middle artery where the distribution density of the fourth type of voxels is higher than a predetermined threshold.

[0024] This application also provides an apparatus for fully automatic selection of arterial input functions in CT perfusion images, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described method for fully automatic selection of arterial input functions in CT perfusion images.

[0025] This application has at least the following effects:

[0026] The method for fully automated selection of arterial input functions in CT perfusion images provided in this application is based on machine learning and clustering algorithms. The entire method can achieve fully automated calculation without human interaction or manual operation, with fast calculation speed and superior results, which can improve the accuracy and speed of deconvolution methods in calculating brain tissue perfusion parameters.

[0027] The arterial input function obtained in this application is superior, avoiding differences caused by manual selection. Furthermore, this application does not refer to morphological features in CT perfusion images when selecting the AIF curve, thus avoiding the influence of morphological differences in CT perfusion images. It also exhibits good robustness to minor disturbances caused by patient activity, and has achieved good results across various case datasets. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a fully automatic selection method for arterial input functions in CT perfusion imaging according to an embodiment of this application.

[0029] Figure 2 This is a flowchart of a fully automatic selection method for arterial input functions in CT perfusion imaging according to an embodiment of this application;

[0030] Figure 3 for Figure 1 Schematic diagram of candidate sites for the middle cerebral artery (indicated by arrows in the diagram);

[0031] Figure 4This is a schematic diagram of the arterial input function result obtained in one embodiment of this application;

[0032] Figure 5 This is a schematic diagram of an arterial result point group obtained in one embodiment of this application (the location indicated by the arrow in the figure). Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0035] In this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number or order of the indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0036] In this application, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a system, product, step, or apparatus that includes a series of units is not necessarily limited to those units that are explicitly listed, but may include other units that are not explicitly listed or that are inherent to such products or apparatus.

[0037] See Figure 1 and Figure 2 One embodiment of this application provides a method for fully automatic selection of arterial input functions in CT perfusion imaging, including steps S100 to S500.

[0038] Step S100: Obtain the middle artery candidate layer after skull removal. The middle artery candidate layer contains voxels of type I. Select the type I voxels to obtain a first expected proportion of type II voxels.

[0039] Step S200: Normalize the area under the curve of the density-time curve of the second type of voxels to obtain the roughness of the normalized curve, and screen to obtain the third type of voxels with the second expected ratio.

[0040] Step S300: Perform the first cluster analysis on the third type of voxels based on the first cluster target number, and screen out the fourth type of voxels with the smallest average first moment.

[0041] Step S400: Calculate the distribution density of the fourth type of voxels to obtain the fifth type of voxels. Perform a second cluster analysis on the second cluster target number to screen out the category with the smallest average peak time. The voxels in this category form the artery candidate point group.

[0042] Step S500: Perform connected component analysis on the artery candidate point group to obtain the artery result point group. The average density time curve of the artery result point group is the artery input function.

[0043] In the following embodiments, each of steps S100 to S400 includes optional sub-steps or detailed steps.

[0044] Step S100 includes steps S110 to S120. Wherein:

[0045] Step S110: Obtain the candidate layer of the middle cerebral artery after skull removal. The candidate layer of the middle cerebral artery contains voxels of type I. In step S110:

[0046] The raw CT perfusion images are obtained, and preprocessing is performed sequentially, followed by reading brain tissue masks after skull removal from each layer. Specifically, preprocessing includes motion correction, image filtering, and skull removal. The preprocessed four-dimensional CT perfusion images are then read as the raw CT perfusion images, which include three-dimensional CT perfusion images composed of cross-sections at different scan times. The four dimensions are time T, spatial layer number L, image height H, and image width W.

[0047] Obtaining candidate layers for the middle artery includes: reading brain tissue masks after removing the skull from each layer, obtaining the area of ​​the brain tissue mask, finding the first distance from the layer with the largest area towards the skull base, and filtering to obtain candidate layers for the middle artery.

[0048] The brain tissue mask obtained after skull removal from preprocessing is read and arranged in the direction from skull base to skull top. The area (L) of the brain tissue mask at each layer is calculated, and the maximum area is found by iterating through each layer. max Towards the skull base, based on the slice spacing ΔL of the CT perfusion images, n layers of brain tissue data are selected, such that the selected data covers approximately a first distance (e.g., 40 mm) of brain tissue in the slice dimension L, i.e., n = [40 / ΔL]. The selected n layers of brain tissue data constitute the middle artery candidate layer, and the voxels contained in the middle artery candidate layer are of type I voxels, which serve as input data for subsequent steps.

[0049] Step S120 involves screening the first type of voxels to obtain a first expected proportion of the second type of voxels. This specifically includes steps S121 to S122.

[0050] Step S121: Perform baseline correction of the density-time curve for the first type of voxel;

[0051] Specifically, for each voxel in each layer of brain tissue, the signal density time-space curve C(t) is obtained. The average signal density over the first m time points when the signal is stable is taken as the density baseline. m can be determined by assessing the degree of curve variation or by experience. The baseline value is subtracted from each point of the density time-space curve C(t) of all voxels to obtain the baseline-corrected curve, and all curve points with values ​​less than zero are set to zero.

[0052] Step S122: Integrate the corrected density-time curve, and select the second type of voxels with the first expected proportion from the first type of voxels;

[0053] Specifically, the signal density integral is obtained by integrating the density-time curve C(t) of all voxels. Sort the signal density integral S(v) of all voxels, filter out voxels with smaller S(v), and the filtering ratio parameter is p. S p S As a hyperparameter, the optimal p can be determined by adjusting the parameter and observing the precision of the results. S Parameters. The reference MCA has a relatively small proportion in brain tissue, p S A larger value can be selected, generally above 0.9, meaning the filtration ratio parameter is p. S The voxels.

[0054] Step S200: Normalize the area under the curve of the density-time curve of the second type of voxel to obtain the roughness of the normalized curve, and screen to obtain the third type of voxel with the second expected ratio.

[0055] Specifically, the area under the curve is normalized for the remaining voxels after integral screening to obtain the roughness of the normalized curve, which is obtained by the following formula: In the formula, C” norm (t) is the normalized curve C norm The second derivative of (t), and R(v) is the integral of the square of the second derivative over the time dimension, used to represent the roughness of the normalized curve.

[0056] During roughness screening, the curve roughness R(v) of the remaining voxels after integration screening is sorted, voxels with relatively large R(v) are filtered out, and voxels with relatively small R(v) are retained. The filtering ratio is p. R p RAs a hyperparameter, the optimal p can be determined by adjusting the parameter and observing the precision of the results. R Parameters. To filter out as many noise curves as possible without removing too many AIF candidate curves, p R Choose the smaller value, generally below 0.2, that is, the filtration ratio is p. R The voxels.

[0057] Roughness screening requires area normalization, which prevents direct integral screening. This embodiment performs integral screening first, followed by roughness screening, thus saving steps in the automated processing flow.

[0058] Step S300: Perform a first cluster analysis on the third type of voxels based on the first cluster target number, and screen out the fourth type of voxels with the smallest average first moment.

[0059] Specifically, this step uses a clustering algorithm belonging to unsupervised learning in machine learning to perform cluster analysis, including: initializing the class center curves, calculating the high-dimensional distance between each voxel density time curve or normalized curve and the class center curves, assigning each voxel density time curve or normalized curve to the class with the smallest distance, recalculating the class center curves, and repeating the distance calculation and assignment operations through multiple iterations until the classification result no longer changes.

[0060] The normalized curve C for the third type of voxel after roughness screening norm (t) is used for the first cluster analysis. k-means++ is used as the initialization method for the cluster centers, and the target number for the first cluster is set to k1. k1 is a hyperparameter, and the optimal k1 parameter can be determined by adjusting the precision of the observation results. For example, k1 = 5.

[0061] The first cluster analysis divided the data into categories based on the target number of the first cluster. For each voxel within each category, the first moment was calculated: μ = ∑t·C norm (t), and calculate the average first moment of each class after classification. Retain the average first moment Minimum category The voxels in the text thus obtained the fourth type of voxel.

[0062] Step S400 includes steps S410 to S420.

[0063] Step S410: Calculate the distribution density of the fourth type of voxels. In the candidate layer of the middle artery, select the expected number of layers. The voxels in the expected number of layers are used as the fifth type of voxels.

[0064] Specifically, the distribution of type 4 voxels on each candidate layer of the middle artery is counted. The candidate layer of the middle artery with the highest expected number of remaining voxels (e.g., 2 layers) is retained, and the voxels distributed on it are classified as type 5 voxels. The expected number can be adjusted according to the thickness of the original CT perfusion image.

[0065] Step S420: Obtain the density-time curve of the fifth type of voxels, perform a second cluster analysis on the second cluster target number, and screen out the category with the smallest average peak time. The voxel composition within this category is as follows: Figure 3 The group of candidate arterial points shown.

[0066] After restoring the normalized density-time curve area of ​​the fifth type of voxels, the density-time curve C(t) is obtained again, and a second cluster analysis is performed. The second cluster analysis uses k-means++ as the initialization method for the cluster centers, and the target number for the second cluster is set to k2. k2 is a hyperparameter, and the optimal k2 parameter can be determined by adjusting the accuracy of the observation results, for example, setting k2 = 5.

[0067] The second cluster analysis divided the target number into several categories. For each category, the peak time (time to reach peak value) t was calculated for each voxel. p And calculate the average peak time for each category after classification. Retain average peak time Minimum category The voxels in the data serve as candidate sites for arterial infarction (AIF), thus forming a group of candidate arterial infarction sites.

[0068] From the perspective of screening methods, average first-order moment screening is more effective than average peak time screening. Layered screening narrows the spatial range, which is beneficial for clustering by average peak time screening to result in more concentrated midpoint sets. Following the steps provided in this embodiment sequentially can improve the speed of automated processing.

[0069] Step S500: Obtain the mask image of the artery candidate point group, and use the maximum connected component algorithm to obtain the connected component with the largest area as the artery result point group (e.g., Figure 5 As shown), the average density-time curve of the arterial result point group is the arterial input function (e.g., Figure 4 (As shown).

[0070] Specifically, a mask image of the artery candidate point group is constructed. Using a maximum connected component algorithm, such as the 8-connected maximum connected component algorithm, c connected components of the artery candidate point group are calculated, and the area Area(c) of each connected component is calculated. The artery candidate point group in the connected component with the largest area is selected as the AIF result point group. The average density-time curve of the AIF result point group is the artery input function (AIF curve).

[0071] In one embodiment of this application, the method for fully automatic selection of arterial input functions in CT perfusion imaging mainly includes the following steps one through eight. Steps one through three correspond to step S100. Step four corresponds to step S200. Step five corresponds to step S300. Steps six and seven correspond to step S400. Step eight corresponds to step S500.

[0072] The first step is to read the preprocessed four-dimensional CT perfusion images, calculate the area of ​​each brain tissue region, and select data covering a 40mm thick layer of brain tissue from the layer with the largest brain tissue area towards the skull base as the candidate layer for the middle artery; thus, the first type of voxel in step S100 is obtained.

[0073] The second step is to use the average signal density at the moment when the signal is stable as the baseline to perform baseline correction on the density-time curve of each voxel in each layer; this completes the baseline correction in step S100.

[0074] The third step is to integrate the density-time curves of all voxels to obtain the signal density integral, and then filter out the voxels with smaller signal density integrals; thus, the second type of voxels in step S100 are obtained.

[0075] The fourth step is to calculate the roughness of the remaining voxel density time normalization curve and filter out the voxels with larger roughness; thus, the third type of voxels in step S200 are obtained.

[0076] Fifth, the kmeans++ machine learning algorithm is used to perform the first cluster analysis on the density time curves of the remaining voxels, and the class with the smallest average first moment is retained; thus, the fourth class of voxels in step S300 is obtained.

[0077] Step 6: Calculate the distribution density of the remaining voxels on the candidate layer of the middle artery, and retain the expected number of voxels on the candidate layer of the middle artery with the largest number of remaining voxels; thus, the fifth type of voxels in S400 is obtained.

[0078] Step 7: Use the kmeans++ machine learning algorithm to perform a second cluster analysis on the density-time curves of the remaining voxels, and retain the class with the smallest average peak time; thus, the artery candidate point group in step S400 is obtained.

[0079] Step 8: Construct a mask image of artery candidate points on the middle artery candidate layer. Using the maximum connected component algorithm, select the artery candidate points in the connected component with the largest area as the artery result point group. The average density time curve of the artery result point group is the artery input function. Thus, the artery result point group and artery input function in S500 are obtained.

[0080] In one embodiment, an apparatus for fully automatic selection of arterial input functions in CT perfusion imaging is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements a method for fully automatic selection of arterial input functions in CT perfusion imaging, namely, the computer program performs the following steps when executed by the processor:

[0081] Step S100: Obtain the middle artery candidate layer after skull removal. The middle artery candidate layer contains voxels of type I. Select the type I voxels to obtain a first expected proportion of type II voxels.

[0082] Step S200: Normalize the area under the curve of the density-time curve of the second type of voxels to obtain the roughness of the normalized curve, and screen to obtain the third type of voxels with the second expected ratio.

[0083] Step S300: Perform the first cluster analysis on the third type of voxels based on the first cluster target number, and screen out the fourth type of voxels with the smallest average first moment.

[0084] Step S400: Calculate the distribution density of the fourth type of voxels to obtain the fifth type of voxels. Perform a second cluster analysis on the second cluster target number to screen out the category with the smallest average peak time. The voxels in this category form the artery candidate point group.

[0085] Step S500: Perform connected component analysis on the artery candidate point group to obtain the artery result point group. The average density time curve of the artery result point group is the artery input function.

[0086] The fully automated selection method for arterial input functions in CT perfusion imaging provided in the embodiments of this application is based on machine learning and clustering algorithms. The entire method is fully automated, requiring no manual intervention or operation. It offers fast computation speed and superior results, improving the accuracy and speed of deconvolution methods in calculating brain tissue perfusion parameters. The k-means++ algorithm used in each clustering analysis exhibits fast convergence, further reducing process time and significantly improving the speed of CT perfusion imaging analysis, thus accelerating the diagnostic process.

[0087] The arterial input function (AIF curve) obtained in each embodiment is superior, avoiding differences caused by manual selection. Furthermore, this method does not refer to morphological features in CT perfusion images when selecting the AIF curve, thus avoiding the influence of morphological differences in CT perfusion images. It also exhibits good robustness to minor disturbances caused by patient activity and has achieved good results across various case datasets.

[0088] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0089] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered to be within the scope of this specification. When technical features of different embodiments are embodied in the same drawing, it can be regarded as the drawing also disclosing examples of combinations of the various embodiments involved.

[0090] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for fully automatic selection of arterial input functions in CT perfusion imaging, characterized in that, include: A candidate layer of the middle cerebral artery is obtained after skull removal. The candidate layer contains voxels of type I. Density-time curve baseline correction is performed on the type I voxels. The corrected density-time curve is used to perform the following steps: The corrected density-time curve is integrated to obtain the signal density integral, which is then sorted. Smaller voxels in the first type of voxels are filtered out based on the filtering ratio parameter to obtain the second type of voxels with the first expected ratio. The area under the curve of the density-time curve of the second type of voxels is normalized to obtain the roughness of the normalized curve, and the third type of voxels with the second expected proportion are selected. The roughness of the normalized curve is obtained by the following formula: In the formula, C” norm (t) is the normalized curve C norm The second derivative of (t) is given by R(v), where R(v) is the integral of the square of the second derivative over the time dimension. During roughness screening, the curve roughness R(v) of the remaining voxels after integral screening is sorted, filtering voxels with relatively large R(v) and retaining voxels with relatively small R(v), with a filtering ratio of p. R ; The third type of voxels undergoes a first cluster analysis based on the first cluster target number to screen out the fourth type of voxels with the smallest average first moment. The distribution density of the fourth type of voxels is statistically analyzed. In the candidate layer of the middle artery, a desired number of layers are selected, and the voxels in the desired number of layers are used as the fifth type of voxels. The density-time curves of the fifth type of voxels are obtained, and a second clustering analysis of the second clustering target number is performed to screen out the category with the smallest average peak time. The voxels in this category form the artery candidate point group. A mask image of the candidate artery point group is obtained, and the connected region with the largest area is obtained by the maximum connected component algorithm as the artery result point group. The average density time curve of the artery result point group is the artery input function.

2. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 1, characterized in that, Obtaining the candidate layer of the middle cerebral artery during cranial removal, specifically including: The raw CT perfusion images were obtained, and preprocessing and reading of brain tissue masks after skull removal were performed sequentially. The area of ​​the brain tissue mask is obtained, and a first distance is found from the layer with the maximum area towards the skull base to screen and obtain the candidate layer of the middle artery.

3. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 2, characterized in that, The preprocessing includes motion correction, image filtering, and skull removal.

4. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 1, characterized in that, Among the first type of voxels, the density-time curve integral value of the second type of voxels is greater than that of the other voxels.

5. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 1, characterized in that, In the second type of voxels, the roughness of the normalization curve of the third type of voxels is smaller than that of the other voxels.

6. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 1, characterized in that, Both the first and second cluster analyses include: Initialize the class center curves, calculate the high-dimensional distance between each voxel density time curve or normalized curve and each class center curve, and assign each voxel density time curve or normalized curve to the class with the smallest distance. Recalculate the class center curves, and repeat the distance calculation and assignment operations through multiple iterations until the classification results no longer change.

7. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 1, characterized in that, The first moment is obtained by calculating the following formula using the normalized curves of each voxel: μ=∑t*C norm (t).

8. The method for fully automatic selection of arterial input functions in CT perfusion imaging according to claim 1, characterized in that, The expected number of layers refers to the layers in the candidate layer of the middle artery where the distribution density of the fourth type of voxels is higher than a predetermined threshold.

9. A device for fully automatic selection of arterial input functions in CT perfusion imaging, 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 steps of the fully automatic selection method for arterial input functions in CT perfusion imaging as described in any one of claims 1 to 8.

Images