Fully automatic post-processing method for brain CT perfusion images

Abstract

The application relates to a kind of brain CT perfusion image full-automatic post-processing methods, comprising: reading CT perfusion image;The CT perfusion image is preprocessed, and the skull perfusion image with skull is obtained, and the brain tissue image containing lateral ventricle after removing skull;Obtain lateral ventricle segmentation result binary graph, comprising: using the skull perfusion image with skull, obtain the optimal image layer of the largest connected domain area in skull;In the brain tissue image, from the optimal image layer, find the first distance towards the direction of skull top, find the second distance towards the direction of skull bottom, obtain the image layer containing lateral ventricle;The image layer containing lateral ventricle is converted into feature image, and lateral ventricle segmentation is carried out on each layer feature image, and lateral ventricle segmentation result binary graph is obtained;Using the brain tissue image, arterial input function and perfusion parameter map are sequentially obtained;The perfusion parameter map is binarized, and the lesion area is obtained in combination with the lateral ventricle segmentation result binary graph.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a full-automatic post-processing method for brain CT perfusion images. BACKGROUND

[0002] Stroke is an acute cerebrovascular disease, which is caused by the sudden rupture of cerebral blood vessels or the blockage of blood vessels, resulting in brain tissue damage. It includes ischemic and hemorrhagic stroke, and has a very high disability rate and a high mortality rate. Acute ischemic stroke (cerebral infarction) is the most common type of stroke, accounting for 60% to 80% of all strokes. The treatment of ischemic stroke mainly includes thrombolysis, intravascular thrombectomy and conservative treatment, etc.

[0003] Perfusion imaging is most widely used in acute stroke and oncology. When used for the diagnosis of stroke, the purpose of perfusion imaging is to determine the extent of the lesion and to delineate the ischemic tissue that can be reperfused. With the development of medical imaging technology and computer technology, brain CT perfusion imaging (CTP) has become an important imaging method for the examination of acute ischemic stroke. Brain CT perfusion imaging is an imaging technique for evaluating the perfusion state of brain parenchyma. It is performed by dynamic brain perfusion examination after intravenous injection of contrast medium to obtain the corresponding brain perfusion images. Through quantitative analysis of the obtained CT perfusion images based on deconvolution or non-deconvolution algorithms, the relevant cerebral hemodynamic perfusion parameters of the patient can be obtained, such as cerebral blood volume (CBV), cerebral blood flow (CBF), mean transit time (MTT), etc. And according to the perfusion parameter map, normal brain tissue, lesion tissue (infarction core) and reperfusable ischemic tissue (ischemic penumbra) can be identified.

[0004] Accurate identification of the location and extent of the infarction core and ischemic penumbra is crucial for the treatment of patients with acute stroke. For patients with acute stroke, whether intravenous thrombolysis or mechanical thrombectomy, the earlier the treatment is initiated, the greater the benefit to the patient. However, the infarction core region and ischemic penumbra region of patients with acute ischemic stroke are usually connected with the lateral ventricle, which affects the judgment of the infarction or ischemic tissue region. Due to the poor speed and accuracy of the existing technology for segmenting the lateral ventricle from CT perfusion images, the actual volume of the infarction core and ischemic penumbra in the CT perfusion images cannot be accurately identified. SUMMARY

[0005] Therefore, it is necessary to provide a full-automatic post-processing method for brain CT perfusion images in view of the above technical problems.

[0006] The application discloses a full-automatic post-processing method for brain CT perfusion images, and relates to the technical field of medical image processing.

[0007] reading the CT perfusion images;

[0008] preprocessing the CT perfusion images to obtain skull-perfused images and brain tissue images containing lateral ventricles after removing the skull;

[0009] obtaining a lateral ventricle segmentation result binary image, comprising:

[0010] obtaining an optimal image layer with the largest skull-in connected domain area from the skull-perfused images;

[0011] finding a first distance in a parietal direction and a second distance in a basilar direction from the optimal image layer on the brain tissue images to obtain an image layer containing lateral ventricles;

[0012] converting the image layer containing lateral ventricles into a feature image, performing lateral ventricle segmentation on each layer of the feature image, and obtaining a lateral ventricle segmentation result binary image;

[0013] obtaining an arterial input function and a perfusion parameter map in sequence from the brain tissue images;

[0014] performing binaryzation processing on the perfusion parameter map, combining the lateral ventricle segmentation result binary image, and obtaining a lesion region.

[0015] Optionally, the preprocessing comprises image registration and filtering, and the brain tissue images are obtained by removing the skull from the skull-perfused images.

[0016] Optionally, the parietal direction and the basilar direction are obtained according to the area ratio change trend of each layer of the brain tissue images and the convex hull images thereof.

[0017] Optionally, the optimal image layer with the largest skull-in connected domain area is obtained from the skull-perfused images, and specifically comprises:

[0018] extracting a skull region of the skull-perfused images, filling holes to obtain a candidate layer with two connected domains outside the skull region;

[0019] selecting a layer with the largest skull-in connected domain area in the candidate layer as the optimal image layer.

[0020] Optionally, the feature image is one of the following:

[0021] a maximum density projection image, a minimum density projection image, a baseline image, an average image, a difference image of the maximum density projection image and the minimum density projection image.

[0022] Optionally, the feature images of each layer are subjected to lateral ventricle segmentation to obtain a lateral ventricle segmentation result binary image, including:

[0023] The feature images of each layer are subjected to rough segmentation by using a global adaptive threshold to obtain a lateral ventricle rough segmentation binary image.

[0024] The lateral ventricle rough segmentation binary image of each layer is subjected to fine segmentation to obtain a lateral ventricle fine segmentation binary image.

[0025] According to the lateral ventricle fine segmentation binary image, a lateral ventricle segmentation result binary image is obtained.

[0026] Optionally, the global adaptive threshold is obtained by the following method:

[0027] The CT values of each pixel point of the feature images of each layer are subjected to point normalization processing in a first interval to obtain a contrast-enhanced image.

[0028] The gray values of each pixel point in all contrast-enhanced images are counted, and a global adaptive threshold is obtained by using an adaptive threshold method.

[0029] Optionally, the lateral ventricle fine segmentation binary image is obtained by performing lateral ventricle fine segmentation on the lateral ventricle rough segmentation binary image, specifically including:

[0030] The connected domains of the lateral ventricle rough segmentation binary image of each layer are extracted to obtain all two-dimensional connected domains of the current layer.

[0031] If the centroid of a two-dimensional connected domain deviates from a first expected position, the two-dimensional connected domain is deleted.

[0032] If the boundary of a two-dimensional connected domain exceeds a second expected position, the two-dimensional connected domain is subjected to segmentation again by using an adaptive threshold method.

[0033] The above steps are repeatedly executed until an expectation is met, and a lateral ventricle fine segmentation binary image is obtained.

[0034] Optionally, the lateral ventricle segmentation result binary image is obtained according to the lateral ventricle fine segmentation binary image, specifically including:

[0035] The lateral ventricle fine segmentation binary images of each layer are correspondingly mapped to the brain tissue image.

[0036] After combination, a three-dimensional maximum connected domain is extracted to obtain a lateral ventricle segmentation result binary image.

[0037] Optionally, the lesion region includes an infarction core region and an ischemic penumbra region, and does not include the lateral ventricle.

[0038] The full-automatic post-processing method of the brain CT perfusion image has at least the following effects:

[0039] The application can automatically complete the post-processing of the brain CT perfusion image after reading the brain CT perfusion image, including image preprocessing, lateral ventricle segmentation, perfusion parameter map calculation, and lesion area determination. The application ensures the reliability of the feature image sample through the determination of the optimal image layer and the image layer containing the lateral ventricle.

[0040] The cranial direction and the cranial base direction of the application are obtained through automatic processing, without the need for external specification, further improving the processing speed of the fully automatic post-processing method of the brain CT perfusion image; the application sequentially performs coarse lateral ventricle segmentation and fine lateral ventricle segmentation, the coarse lateral ventricle segmentation can automatically segment the lateral ventricle region according to the perfusion image adaptive threshold, improving the adaptive degree of the method to different perfusion images; at the same time, the sample range is reduced, the fine lateral ventricle segmentation further segments through the adaptive threshold method on the basis of reducing the sample, which can better exclude the influence of infarction and other low perfusion areas on the lateral ventricle segmentation result, and can quickly and accurately gradually approach the ideal lateral ventricle segmentation result binary graph. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 It is a flowchart of the fully automatic post-processing method of the brain CT perfusion image in an embodiment of the application;

[0042] Figure 2 It is a flowchart of the fully automatic post-processing method of the brain CT perfusion image in an embodiment of the application;

[0043] Figure 3 It is a flowchart of obtaining the lateral ventricle segmentation result binary graph in an embodiment of the application;

[0044] Figure 4a It is a cranial base section of the CT perfusion image not including the lateral ventricle in an embodiment of the application;

[0045] Figure 4b It is a layer diagram of the CT perfusion image in an embodiment of the application, the number of connected domains is 3;

[0046] Figure 4c It is a layer diagram of the CT perfusion image in an embodiment of the application, the number of connected domains is 2;

[0047] Figure 4d It is a binary graph of Figure 4a ;

[0048] Figure 4e It is a binary graph of Figure 4b ;

[0049] Figure 4f It is a binary graph of Figure 4c ;

[0050] Figure 5This is a flowchart of a method for fine segmentation of the lateral ventricle in one embodiment of this application;

[0051] Figure 6a This is a feature image from one embodiment of this application;

[0052] Figure 6b This is a coarse segmentation result of the lateral ventricle at one level in one embodiment of this application;

[0053] Figure 6c This is a subdivision result of the lateral ventricle at one level in one embodiment of this application;

[0054] Figure 6d This is the final result of lateral ventricle segmentation at one level in one embodiment of this application. Detailed Implementation

[0055] 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.

[0056] See Figure 1 and Figure 2 A fully automated post-processing method for brain CT perfusion images, including steps S100 to S500.

[0057] Step S100, Read CT perfusion images (completed) Figure 2 (Reading CT perfusion images);

[0058] CT perfusion images consist of layered CT perfusion images at different time points, after being time-ordered. The brain CT perfusion image sequence is read, and the images are layered based on the spatial information contained within them. Each layer contains perfusion images from all different time points; for each layer, the images are time-ordered based on the scan time information contained within them.

[0059] Step S200: Preprocess the CT perfusion images to obtain perfusion images with the skull and images of brain tissue including the lateral ventricles after removing the skull (this completes the process). Figure 2 (Image preprocessing in the process);

[0060] Preprocessing included image registration and filtering. Brain tissue images were obtained by removing the skull from images with skull perfusion.

[0061] Specifically, image preprocessing operations are performed on the time-ordered images, including: image registration, which registers images from different scanning times at each level to the image at the first moment of the same level, so that the voxel points on the images at all moments coincide in space; filtering, which filters all registered images to reduce noise and improve the signal-to-noise ratio of the images; and brain tissue segmentation, which removes the skull with high CT values ​​according to the threshold method or other segmentation methods to obtain brain tissue images.

[0062] See Figure 3 Step S300: Obtain the binary image of the lateral ventricle segmentation result (completed). Figure 2 The process (segmentation of the lateral ventricle) includes steps S310 to S330, which include reading the preprocessed perfusion image with skull and the brain tissue image after removing the skull, segmenting the brain tissue image into the lateral ventricle, and obtaining the segmentation result.

[0063] Step S310 involves using perfusion imaging of the skull to obtain the optimal image layer with the largest connected region area within the skull, specifically including steps S311 to S312. Among these steps...

[0064] Step S311: Extract the skull region with skull perfusion image, fill in the holes and obtain a candidate layer with two connected domains outside the skull region.

[0065] Step S312: Select the layer with the largest connected area within the skull from the candidate layers as the optimal imaging layer.

[0066] Steps S311 and S312 include reading the preprocessed CT perfusion images and determining the image slices containing the lateral ventricles. Reading the preprocessed CT perfusion images: These images have undergone registration and filtering but have not been segmented into brain tissue; that is, the images include the skull region. The optimal image slice containing the lateral ventricles is determined using a specific method. Images within a certain spatial range from the optimal slice are considered to contain the lateral ventricles, while images outside this range are considered not to contain the lateral ventricles. Subsequent processing only applies to image slices containing the lateral ventricles. This step eliminates the influence of sulci and gyri with lower CT values ​​near the top of the skull on lateral ventricle segmentation.

[0067] The method for determining the optimal imaging slice containing the lateral ventricle specifically includes (1) to (6). Among them, (1) to (4) correspond to step S311, and (5) to (6) correspond to step S312.

[0068] (1) Maximum density projection (MIP) image I of the i-th layer i (include Figures 4a to 4c Based on the thresholding method, the skull region is extracted to obtain a binary image B of the skull. i (include Figures 4d to 4f );

[0069] (2) Repair and fill the small holes on the skull binary image B i to obtain image T i ;

[0070] (3) Calculate the number of connected domains inside and outside the skull. If the number of connected domains is greater than 2 or less than 2, all are rejected, and only the layer with a connected domain number of 2 is reserved (as shown in Figure 4c );

[0071] (4) Perform the above (1) to (3) processing on all layer images. After screening, there are N layer planes as the optimal image layer planes to be selected;

[0072] (5) For the jth layer in the N selected layers, calculate the connected domain area A j inside the skull. Perform the same processing on all N layers;

[0073] (6) Compare the connected domain areas inside the skull of all N selected layers. If the connected domain area of the kth layer is the largest, the kth layer is the optimal image layer plane containing the lateral ventricle. In the above (1) to (6), i, j, N are all integers.

[0074] Step S320, on the brain tissue image (corresponding to the CT perfusion image after preprocessing in Figure 3 ), from the optimal image layer, find the first distance L1 in the direction of the top of the skull and the second distance L2 in the direction of the bottom of the skull, and obtain the image layer plane containing the lateral ventricle (corresponding to the image layer plane containing the lateral ventricle determined in Figure 3 );

[0075] In this step, the direction of the top of the skull and the direction of the bottom of the skull are obtained according to the area ratio change trend of each layer brain tissue image and its convex hull image. Specifically, it includes (7) to (11) to obtain the direction of the top of the skull and the direction of the bottom of the skull; (12) to obtain the image layer plane containing the lateral ventricle.

[0076] (7) Perform brain tissue segmentation processing on the brain tissue region image I i of the ith layer to obtain the binary image B i ;

[0077] (8) Calculate the convex hull image H i of the binary image B i ;

[0078] (9) Calculate the area A i of the binary image B i and the area A' i of the convex hull image H i , and calculate the ratio r i = A i / A' i;

[0079] (10) The above processing is performed on all layers of brain tissue images to obtain the area ratio of each layer of brain tissue image to the convex hull image;

[0080] (11) The top of the skull and the bottom of the skull are determined according to the area ratio of each layer of brain tissue image to the convex hull image: the area ratio of the top layer is close to 1, and the area ratio of the bottom layer is less than 1. Thus, the top of the skull and the bottom of the skull are obtained according to the trend of the area ratio of each layer of brain tissue image to the convex hull image.

[0081] (12) The first distance L1 is found from the optimal image layer containing the lateral ventricle to the top of the skull to obtain layer 1, and the second distance L2 is found from the optimal image layer containing the lateral ventricle to the bottom of the skull to obtain layer 2, and all layers between layer 1 and layer 2 are image layers containing the lateral ventricle. The values of the first distance and the second distance can be determined by physiological parameters of the human body. In fact, the distribution of image layers containing the lateral ventricle on both sides of the optimal image layer is uneven, that is, the ideal first distance and the second distance are not the same. The first distance L1 can be usually taken as 40-50 mm, and the second distance L2 can be usually taken as 20-40 mm. This step automatically obtains the top of the skull and the bottom of the skull, and excludes as much as possible the non-target region not containing the lateral ventricle in the image layer containing the lateral ventricle, thereby ensuring the reliability of the sample and significantly improving the processing speed of the subsequent process.

[0082] Step S330 includes: converting the image layer containing the lateral ventricle into a feature image (step S331, corresponding to the extraction of the feature image in Figure 3 ); and performing lateral ventricle segmentation on each layer of feature image to obtain a lateral ventricle segmentation result binary image (step S332, corresponding to the lateral ventricle coarse segmentation and lateral ventricle fine segmentation in Figure 3 , in sequence);

[0083] In step S331, the feature image is one of the following: a maximum density projection image I max , a minimum density projection image I min , a baseline image (i.e. an image at the initial moment) I baseline , an average image I mean , a difference image I diff = (I max -I min ) of the maximum density projection image and the minimum density projection image.

[0084] In step S332, lateral ventricle segmentation is performed on each layer of feature image (for example, the maximum density projection image I max shown in Figure 6a ) to obtain a lateral ventricle segmentation result binary image, including:

[0085] S3321, performing coarse segmentation on each layer feature image using a global adaptive threshold to obtain a ventricle coarse segmentation binary image as shown in Figure 6b

[0086] Specifically, the feature images of all layers to be segmented are segmented using a global adaptive threshold to obtain the ventricle coarse segmentation result of each layer, which is a binary image. In the image, the foreground points are the ventricle regions obtained by coarse segmentation, and the value is 255; the background points are other regions except the ventricle, and the value is 0.

[0087] The global adaptive threshold is determined according to all layer feature images containing the ventricle layer, and the global adaptive threshold is obtained by the following method:

[0088] (A) Point normalization processing is performed on the CT values of each pixel point of each layer feature image in the first interval to obtain a contrast-enhanced image. Specifically, the points with CT values in the range of 0-50HU (first interval) in the feature image of each layer are normalized to 0-255HU, the points with CT values higher than 50HU are assigned a value of 255HU, and the points with CT values lower than 0HU are assigned a value of 0HU to obtain a contrast-enhanced image. This operation can enlarge the CT value range of the region where the ventricle is located and enhance the contrast between the ventricle and the surrounding brain tissue.

[0089] (B) The gray values of each pixel point in all contrast-enhanced images are counted, and a global adaptive threshold is obtained using an adaptive threshold method. Specifically, the CT values of each pixel point in the contrast-enhanced images of all layers to be segmented are counted, the number of points with different gray values is counted, and a global gray histogram of all layers to be segmented is obtained. A global adaptive threshold is obtained using an adaptive threshold method, such as the maximum between-class variance method (i.e. Otsu method).

[0090] S3322, performing ventricle fine segmentation on each layer coarse segmentation binary image to obtain a ventricle fine segmentation binary image, which specifically includes steps (13) to (16).

[0091] (13) Extracting connected domains from each layer coarse segmentation binary image to obtain all two-dimensional connected domains (such as the N connected domains shown in Figure 5 ) of the current layer. Specifically, for the ventricle coarse segmentation result of one layer, the connected domains are extracted to obtain all connected domains after coarse segmentation of the layer.

[0092] (14) If the centroid of a two-dimensional connected domain deviates from the first expected position, delete the two-dimensional connected domain. Specifically, first extract the centroid M0(x0, y0) of the brain tissue of the layer, which is obtained by the following formula.

[0093]

[0094] ​

[0095] where i, j are the coordinates of all points in the x, y direction of the current slice image; m, n are the number of points in the x, y direction of the current slice image; I ij is the gray value of the point with coordinates (i, j) in the current slice lateral ventricle rough segmentation result image.

[0096] the centroid M i (x i ,y i ) of each two-dimensional connected domain after the rough segmentation of the current slice is calculated. i If M L is out of the range of the first expected position, the connected domain is deleted, otherwise the next step is entered.

[0097] The range of the first expected position is defined by the following formula:

[0098] x0-α·(x0-x R )<x<x0+α·(x U -x0)

[0099] y0-α·(y0-y B )<y<y0+α·(y L -y0)

[0100] where x, y are the centroid positions of the two-dimensional connected domain, α is a given coefficient, for example, α is taken as 0.4-0.6, x R is the left boundary of the brain tissue, x U is the right boundary of the brain tissue, y B is the upper boundary of the brain tissue, and y i is the lower boundary of the brain tissue. If the centroid of the two-dimensional connected domain does not meet the above formula, i.e., the centroid of the two-dimensional connected domain deviates from the first expected position. As shown in the formula, if the distance Li between the centroid M i of the two-dimensional connected domain and the centroid M0 of the brain tissue of the current slice is greater than a given value, i.e., the centroid M i of the two-dimensional connected domain is not in the above range, i.e., the centroid M L of the two-dimensional connected domain deviates from the first expected position. Since the centroid of the noise region not connected with the lateral ventricle is not in the range of the first expected position, the noise region not connected with the lateral ventricle can be excluded through step (14).

[0101] (15) If the boundary of a two-dimensional connected domain exceeds the ROI range shown in the second expected position, the adaptive threshold method is used for segmentation again; Figure 5

[0102] The range of the second expected position is defined by the following formula:

[0103] x0-β·(x0-x R )<y<x0+β·(x​R - x0)

[0104] y0- β · (y0- y U < y < y0+ β · (y B - y0)

[0105] where x, y are the boundary position of the two-dimensional connected domain, β is a given coefficient, for example, 0.6-0.8. For each two-dimensional connected domain whose centroid after the current layer rough segmentation satisfies the first expected position range, it is determined whether the boundary position of the two-dimensional connected domain is within the second expected position. If the above formula is met, the boundary of the two-dimensional connected domain is within the second expected position. It can be understood that if the range exceeds the second expected position, there is a noise region connected with the lateral ventricle.

[0106] (16) The steps (13)-(15) are repeatedly executed until the expectation is met, and the lateral ventricle segmentation binary graph as shown in FIG. 8 is obtained. Figure 6c

[0107] If the boundary of the two-dimensional connected domain exceeds the range of the second expected position, the adaptive threshold method (such as Otsu method) is used to re-segment the two-dimensional connected domain, and steps (13)-(15) are executed again after segmentation. Until both the centroid position of the two-dimensional connected domain is within the first expected position and the boundary of the two-dimensional connected domain is within the second expected position (until the expectation is met). The connected domains satisfying the above conditions are merged to obtain the lateral ventricle segmentation results of each layer, and the lateral ventricle segmentation binary graph is obtained.

[0108] S3323, according to the lateral ventricle segmentation binary graph, the lateral ventricle segmentation result binary graph is obtained, specifically including:

[0109] The lateral ventricle segmentation binary graph of each layer is mapped to the brain tissue image accordingly;

[0110] After combination, the three-dimensional maximum connected domain is extracted, and the lateral ventricle segmentation result binary graph as shown in FIG. 8 is obtained (to complete the extraction of three-dimensional connected domain in Figure 6d Figure 3 ).

[0111] In this step, the lateral ventricle segmentation results of all layers containing the lateral ventricle are mapped to the three-dimensional model, and the maximum connected domain of the three-dimensional model is extracted to obtain the three-dimensional model of the entire lateral ventricle. The three-dimensional model is mapped to each layer respectively to obtain the final lateral ventricle segmentation result of each layer. This step can determine the final lateral ventricle region according to the spatial connectivity of the segmentation result, that is, the maximum connected domain is taken as the final lateral ventricle region. This step makes the lateral ventricle region obtained by segmentation have better continuity in space and be more consistent with the anatomical characteristics of the lateral ventricle.

[0112] ​​Step S400, using the brain tissue image, sequentially obtaining an arterial input function and a perfusion parameter map (complete Figure 2 The arterial input function AIF is obtained and the perfusion parameter map is calculated.

[0113] For example, a point on the middle cerebral artery (MCA) is automatically selected as a global arterial input function (AIF), and a perfusion equation is solved: The residual function R(t) of each voxel of the brain tissue is calculated, and then the perfusion parameter map is obtained. In the formula, C(t) is the concentration-time curve of each voxel of the brain tissue, AIF(t) is the arterial input function, is a convolution symbol.

[0114] The perfusion parameter map obtained from the residual function R(t) includes: the time integral of the R(t) curve, the cerebral blood volume CBV. The peak value of the R(t) curve, the cerebral blood flow CBF. The peak value corresponding time of the R(t) curve, the residual function peak time Tmax. The average transit time,

[0115] Step S500, binarizing the perfusion parameter map, and combining the lateral ventricle segmentation result binary image to obtain a lesion area. Complete Figure 2 The infarct core and the ischemic area (ischemic penumbra area) are identified.

[0116] The lesion area includes the infarct core area and the ischemic penumbra area, and does not include the lateral ventricle. In step S500, the parameter Figure Two The position and volume of the infarct core (relative CBF <30%) and the low perfusion (Tmax>6s) area are obtained by binarizing the parameter. CBF is the cerebral blood flow, relative CBF is the ratio of the cerebral blood flow of each voxel to the normal CBF, Tmax is the peak time of the residual function. The ischemic penumbra is the area in the low perfusion area that does not belong to the infarct core. The lateral ventricle area segmented in this step is excluded from the infarct core and the ischemic penumbra area.

[0117] The full-automatic post-processing method of brain CT perfusion image proposed in each embodiment of the present application can read the brain CT perfusion image and perform preprocessing, and automatically complete the post-processing of the brain CT perfusion image, including image preprocessing, perfusion parameter map calculation, infarct, lateral ventricle segmentation, and ischemic volume calculation. The method can quickly and accurately identify the position and volume of the infarct core and the penumbra, and accurately exclude the influence of the lateral ventricle on the identification of the infarct core and the penumbra. Therefore, more imaging information can be provided to help doctors better diagnose acute stroke patients and provide more information for clinical treatment.

[0118] It should be understood that, although Figures 1 to 3The steps in the flowcharts of the above embodiments are shown in sequence according to the arrows, but the steps are not necessarily executed in the order shown by the arrows. Unless otherwise specified herein, the steps are not necessarily executed in strict sequence, and the steps can be executed in other sequences. Moreover, Figures 1 to 3 At least part of the steps in the flowcharts of the above embodiments can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or sub-steps or stages of other steps.

[0119] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of each method. Any reference to memory, storage, database or other medium used in the embodiments provided in the present 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. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0120] The technical features of the above embodiments can be combined in any way. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present disclosure. When the technical features in different embodiments are embodied in the same figure, it can be considered that the figure also discloses the combination of each embodiment involved.

[0121] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific and detailed manner, but should not be construed as limiting the scope of the patent. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these are all within the scope of the present application. Therefore, the scope of protection of the patent of the present application should be subject to the appended claims.

Claims

1. A fully automatic post-processing method for brain CT perfusion images, characterized in that, The method comprises the following steps: reading CT perfusion images; preprocessing the CT perfusion images to obtain skull perfusion images and brain tissue images containing lateral ventricles after removing the skull; obtaining a lateral ventricle segmentation result binary image, comprising: using the skull perfusion images to obtain an optimal image layer with the largest skull internal connected domain area; on the brain tissue images, finding a first distance from the optimal image layer to the top of the skull and a second distance to the bottom of the skull to obtain an image layer containing lateral ventricles; converting the image layer containing lateral ventricles into a feature image, performing lateral ventricle segmentation on each layer of the feature image, and obtaining a lateral ventricle segmentation result binary image; using the brain tissue images to sequentially obtain an arterial input function and a perfusion parameter map; performing binary processing on the perfusion parameter map and combining the lateral ventricle segmentation result binary image to obtain a lesion region.

2. The method of fully automatic post-processing of brain CT perfusion images according to claim 1, characterized in that, The preprocessing comprises image registration and filtering, and the brain tissue images are obtained by removing the skull from the skull perfusion images.

3. The method of fully automatic post-processing of brain CT perfusion images according to claim 1, characterized in that, The top direction and the bottom direction are obtained according to the area ratio change trend of each layer of the brain tissue images and the convex hull image thereof.

4. The method of claim 1, wherein the method is characterized by, Using the skull perfusion images to obtain an optimal image layer with the largest skull internal connected domain area, specifically comprising: extracting the skull region of the skull perfusion images, filling the holes to obtain a selected layer outside the skull region with two connected domains; selecting the layer with the largest skull internal connected domain area in the selected layer as the optimal image layer.

5. The method of claim 1, wherein the method is characterized by, The feature image is one of the following: maximum density projection image, minimum density projection image, baseline image, average image, difference image of maximum density projection image and minimum density projection image.

6. The method of claim 1, wherein the method is characterized by, Performing lateral ventricle segmentation on each layer of the feature image to obtain a lateral ventricle segmentation result binary image, comprising: performing rough segmentation on each layer of the feature image using a global adaptive threshold to obtain a lateral ventricle rough segmentation binary image; performing lateral ventricle fine segmentation on each layer of the rough segmentation binary image to obtain a lateral ventricle fine segmentation binary image; obtaining a lateral ventricle segmentation result binary image according to the lateral ventricle fine segmentation binary image.

7. The fully automated post-processing method of brain CT perfusion images according to claim 6, characterized in that, The global adaptive threshold is obtained by the following method: point normalizing the CT values of each pixel point of each layer of the feature image in a first interval to obtain a contrast-enhanced image; statistically counting the gray values of each pixel point in all contrast-enhanced images, and obtaining a global adaptive threshold using an adaptive threshold method.

8. The fully automatic post-processing method of brain CT perfusion images according to claim 6, characterized in that, Performing lateral ventricle fine segmentation on each layer of the rough segmentation binary image to obtain a lateral ventricle fine segmentation binary image, specifically comprising: extracting connected domains from each layer of the rough segmentation binary image to obtain all two-dimensional connected domains of the current layer; if the centroid of a two-dimensional connected domain deviates from a first expected position, the two-dimensional connected domain is deleted; if the boundary of a two-dimensional connected domain exceeds a second expected position, the two-dimensional connected domain is segmented again using an adaptive threshold method; the above steps are repeatedly executed until the expected condition is met to obtain a lateral ventricle fine segmentation binary image.

9. The method of fully automatic post-processing of brain CT perfusion images according to claim 6, characterized in that, Obtaining a lateral ventricle segmentation result binary image according to the lateral ventricle fine segmentation binary image, specifically comprising: mapping each layer of the lateral ventricle fine segmentation binary image to the brain tissue images accordingly; The three-dimensional maximum connected domain is extracted after combination to obtain a binary graph of the lateral ventricle segmentation result.

10. The method of fully automatic post-processing of brain CT perfusion images according to claim 1, characterized in that, The lesion region includes an infarction core region and an ischemic penumbra region, and does not include the lateral ventricle.

Images